Computer-implemented method for providing a starting sequence, in particular a partially text-based starting sequence

Abstract

The invention relates to a computer-implemented method (10), using an algorithm (12), in particular machine learning, having the following steps: - inputting (14) at least one input sequence (22) into the algorithm (12), said input sequence describing a product (32), system and / or method which is novel, in particular at least for the algorithm (12), preferably patentable; - processing (16) the input sequence (22) by collecting meta data (76) on the product (32), system and / or method using the algorithm (12); - generating (18) a starting sequence (30) by means of the algorithm (12), said starting sequence being based on at least some of the collected meta data (34) and in particular being at least partially text-based; and - providing (20) the starting sequence (30).

Description

[0001] Computer-implemented method for providing an output sequence, in particular at least partially text-based

[0002] State of the art

[0003] The invention relates to a computer-implemented method with an algorithm according to claim 1, a computer-implemented method for learning an algorithm according to claim 22, a computer-implemented method for continuously learning an algorithm according to claim 23, a system according to claim 24, a computer program with program code according to claim 25 and a computer-readable storage medium according to claim 26.

[0004] The automated generation of patent applications, especially those supported by AI, brings with it a number of challenges, for example with regard to quality and / or data security.

[0005] The object of the invention is, in particular, to provide a computer-implemented method with advantageous properties regarding algorithm-based processing of, in particular, technically novel, preferably patentable, input sequences. This object is achieved according to the invention by the features of patent claim 1, while advantageous embodiments and further developments of the invention can be found in the subclaims.

[0006] Advantages of the invention

[0007] A computer-implemented method using an algorithm, in particular machine learning, with the following steps:

[0008] Input of at least one input sequence into the algorithm which describes a product, system and / or method which is novel, in particular at least for the algorithm, and preferably patentable;

[0009] - processing the input sequence by collecting metadata about the product, system and / or process using the algorithm;

[0010] - generating an output sequence based on at least part of the collected metadata, in particular at least partially text-based, by means of the algorithm; and

[0011] - Providing the output sequence is proposed. In particular, the input sequence is entered in digital form. The input sequence is particularly preferably entered as a text-based input sequence, in particular in the form of patent claims. Alternatively or additionally, the input sequence is entered as a graphic and / or a voice sequence and / or a mind map that clearly describes the product, system and / or method and / or a computer program with program code that clearly describes the product, system and / or method. If the input sequence is designed as a computer program with program code, this can preferably be analyzed and / or executed directly, and in particular on the basis of this, metadata relating to an analysis result and / or an execution result can be collected during the processing of the input sequence.Preferably, in such a configuration, the computer program with program code can directly output an output sequence based on the execution of the computer program with program code. Particularly preferably, the input sequence describes an inventive concept of a technically implementable product, system, and / or method.

[0012] The computer-implemented method is preferably carried out by a computing device. The algorithm, in particular the machine learning algorithm, is particularly preferably executed by the computing device. A “computing device” is to be understood in particular as a device with an information input, an information processing device, and an information output. The computing device advantageously has at least one processor, a memory unit, an input and output means, further electrical components, an operating program, control routines, control routines, and / or calculation routines. The memory unit of the computing device particularly preferably comprises a computer program with program code for executing the algorithm. The algorithm, in particular the machine learning algorithm, in particular the computer program with program code, is particularly preferably executed with the processor. The computing device is preferably at least part of a computer.Preferably, the components of the computing device are arranged on a common circuit board and / or advantageously in a common housing. Alternatively, however, the computing device can also be designed as a distributed, particularly virtual, computing device, such as a cloud.

[0013] The article of manufacture is preferably a device, machine, equipment, product or composition, which is preferably manufactured by a human. The system is preferably an arrangement of articles of manufacture, in particular parts or components, which preferably interact with one another to particularly preferably perform a specific function or performance. Alternatively or additionally, the system is a software application, a network or another type of technical structure. The method is preferably a method for performing a specific task or for solving a technical problem. The method is, for example, a production process, a process for manufacturing products and / or a method for processing data or information.The fact that the product, system and / or method is “novel at least for the algorithm” should be understood to mean that the product, system and / or method is not present in any database accessible to the algorithm and in particular is not present in any trained database of the algorithm, in particular is missing from the accessible database and the trained database. Preferably, the product, system and / or method is not present in any database from which the metadata is collected. Particularly preferably, the product, system and / or method is missing from a global, in particular freely accessible, patent database, in particular in the complete prior art. In particular, the product, system and / or method described by the input sequence is at least unknown to the algorithm and / or cannot be found in the databases to which the algorithm can access.The fact that the product, system and / or process is “patentable” should be understood in particular to mean that the product is technical, novel, preferably based on an inventive step and in particular is industrially applicable, preferably according to the German Patent Act Paragraph 1 Paragraph 1.

[0014] Preferably, the input sequence is input into the algorithm for processing the input sequence. Particularly preferably, the input sequence is input into the algorithm using the processor from the memory unit of the computing device. Preferably, metadata of the input sequence is input into the algorithm for the input sequence. The metadata of the input sequence is preferably a file type, an editor, in particular a name of an editor and / or a law firm and / or a company, a creation date, a modification date, and / or a sequence size. Particularly preferably, a sequence type of the input sequence is recorded as metadata of the input sequence, which defines a data structure of the input sequence and / or a data type of the input sequence.Particularly preferably, the sequence type of the input sequence is recognized by the algorithm in a processing step, in particular based on the metadata of the input sequence, preferably based on content data of the input sequence. Alternatively and / or additionally, the sequence type is entered by a user via a user interface, in particular into the input means of the computing device. If the sequence type of the input sequence is additionally entered by a user, the detected sequence type is preferably compared with the input sequence type during processing. Particularly preferably, the result of the comparison of the sequence types is output to a user via the user interface. If at least two input sequences are received, the relationship between the input sequences is preferably detected, in particular by the algorithm, preferably based on the metadata of the input sequence.Alternatively or additionally, the relationship between the input sequences is entered by a user via the user interface. Preferably, the sequence type of the input sequence and / or the relationship between the at least two input sequences is determined based on the sequence type and the processor as metadata.

[0015] In particular, during processing of the input sequence, the product, system and / or method, in particular the technical nature of the product, system and / or method, is determined, preferably starting from the input sequence. Particularly preferably, during processing of the input sequence, preferably starting from the input sequence, a function, a procedure, an assembly of the product, system and / or method is determined. Preferably, metadata is collected which has at least one, in particular technical, property and / or advantage of the product, system and / or method. Preferably, during processing of the input sequence, the collected metadata is assigned to the input sequence, in particular to the product, system and / or method.Particularly preferably, the metadata relating to the product, system and / or method is collected, which describes the product, system and / or method in further detail, in particular technically, preferably based on the advantages and / or properties and / or based on technical configurations. In particular, the collected metadata, in particular technical advantages, properties, functions and / or technical configurations of the product, system and / or method, are at least substantially missing from the input sequence. Preferably, the metadata are metasequences, in particular text sequences, particularly preferably at least one word and / or at least one sentence in each case.Particularly preferably, the collected metadata are a plurality of metasequences, wherein each metasequence describes a property, an advantage, a function and / or a technical embodiment, in particular as a technical term, of the product, system and / or method. A plurality of metasequences should be understood to mean in particular at least 10, preferably at least 100, advantageously at least 500 and particularly preferably at least 1000. In particular, the number of metadata, in particular the metasequences, is determined by a complexity of the product, system and / or method, wherein the metadata describe the product, system and / or method at least sufficiently precisely, preferably all functions of the product, system and / or method.

[0016] Preferably, the algorithm uses the collected metadata relating to the product, system and / or method to determine the portion of the collected metadata from the collected metadata relating to the product, system and / or method during processing of the input sequence, in particular based on the quality of the metadata, in particular based on a weighting of the quality of the metadata, in particular by sorting out a portion of the collected metadata. Collected metadata which is at least partially duplicated, which is partially incomplete, immoral and / or technically faulty, is particularly preferably sorted out. The quality of the metadata is preferably determined by technical properties and / or advantages of the product, method and / or system. In particular, the portion of the collected metadata comprises only the technically applicable metadata of the product, system and / or method and / or metadata which can be combined with the product, system and / or method.

[0017] An “algorithm” should be understood in particular as at least essentially a sequence of computing operations on the computing device, which preferably cause the computer to manipulate data and / or control system components. The algorithm preferably forms, in particular, at least one machine learning model, preferably at least two machine learning models, wherein the input sequence is processed with at least one model and the output sequence is generated with at least one model. Particularly preferably, the algorithm forms a neural network, in particular a multi-input neural network and / or a transformer neural network, as a machine learning algorithm, at least for processing the input sequence and / or for generating the output sequence. Alternatively or additionally, the algorithm at least partially comprises a Naive Bayes classifier, a Support Vector Machine and / or a Random Forest.Particularly preferably, the algorithm forms at least one text generation model, in particular as a machine learning model, with which at least the output sequence is generated. Preferably, the algorithm, in particular as a text generation model, at least partially forms a natural language model, which, in particular during the generation of the output sequence, generates a coherent text that, in particular, forms the output sequence. Preferably, the text generation model, with the portion of the collected metadata, generates a text, in particular a coherent text, as the output sequence.

[0018] Particularly preferably, when generating the output sequence, a text, in particular a description of the input sequence, preferably a description of the product, system and / or method, which is novel in particular for an algorithm and preferably patentable, is generated, in particular with the text generation model of the algorithm. Preferably, the output sequence, in particular in the case of a text-based input sequence, particularly preferably in the case of an input sequence as claims, at least partially, preferably completely, comprises the input sequence. In particular, the output sequence is generated as a description of the product, system and / or method. Particularly preferably, an application text for a patent application is generated as the output sequence, which application text describes the product, system and / or method, in particular claims it.When an output sequence is used as the application text for the product, system, and / or method, the input sequence is preferably incorporated in its entirety into the claims of the patent application and / or at least partially into a description of the patent application, in particular the description of the claims of the patent application, preferably the description of advantages of the claims of the patent application, and / or at least partially into a description of the figures of the application. Preferably, when generating the output sequence,

[0019] Patent application at least one abstract of the product, system and / or method is generated. In particular, based on the part of the collected metadata, the product, system and / or method is further detailed, in particular the function, structure, properties and / or advantages are included. In particular, the part of the collected metadata is linked to the product, system and / or method, in particular text-based, during the generation of the output sequence. Particularly preferably, the part of the collected metadata is determined by processing the input sequence, in particular by sorting, in particular weighting, the collected metadata. Particularly preferably, the product, system and / or method is disclosed in the output sequence so clearly and completely that a person skilled in the art can carry it out.The fact that the output sequence is in particular “at least partially text-based” should be understood to mean that the output sequence preferably comprises a graphic representation, particularly preferably a patent drawing, of the product, system and / or method, in particular a schematic illustration and / or a flow chart of the product, system and / or method.

[0020] Particularly preferably, the output sequence is provided following the generation of the output sequence. The provided output sequence is preferably a coherent text, in particular a coherent description, preferably a patent application, of the product, system, and / or method. The output sequence is preferably provided, in particular stored, at least in the memory unit of the computing device. Particularly preferably, the output sequence is provided to the output means of the computing device and preferably output in the user interface.

[0021] In an alternative embodiment, the input sequence is checked for structural and / or content optimization potential during processing of the input sequence with the algorithm. In particular, metadata relating to a required structure and / or required content for the given input sequence, in particular for the product, system and / or method, is collected based on the input sequence and in particular based on the metadata of the input sequence during processing of the input sequence. In particular, in a processing step, the existing input sequence is compared with the metadata of the required structure. Particularly preferably, during generation of the output sequence, the input sequence is revised in terms of content and / or structure with comments and / or markings, in particular based on a comparison between the input sequence and the required structure, and output in text-based form.Particularly preferably, in the case of a text-based input sequence, in particular in the case of the input sequence as claims of the product, system and / or method, a structural design of the input sequence and in particular the content description of the product, method and / or system are checked during the processing of the input sequence. In particular, when generating the output sequence, the text-based input sequence is supplemented and / or shortened based on the portion of the collected metadata, the structural design and / or the content description of the product, system and / or method of the input sequence.

[0022] In a further embodiment of the method according to the invention, the generation of the output sequence is supported by a user. Preferably, the metadata for the product, system and / or method is suggested to a user, who then selects the metadata that is applicable to the given product, system and / or method of the input sequence. Particularly preferably, a user is involved step by step in the generation of the output sequence, wherein in particular for each feature and / or each function the user is supported using the metadata of the product, system and / or method. By means of the method according to the invention, an at least partially text-based output sequence with advantageously further metadata can be advantageously generated and provided for a product, system and / or method that is novel in particular for an algorithm and is preferably patentable.Furthermore, by collecting metadata, preferably as an advantage and / or property of the product, system and / or method, in particular compared to generating metadata, the speed of data processing can be increased and / or more cost-effective processors can be used. Furthermore, by generating the output sequence based on at least part of the metadata, the quality of the output sequence can be increased. Furthermore, a patent application, in particular with claims as an input sequence, can advantageously be generated and provided. Furthermore, by collecting the metadata, it can advantageously be ensured that the text-generated output sequence is at least substantially uniform.

[0023] It is further proposed that the input sequence describes at least one sub-element and / or method step and / or a combination of at least two sub-elements of the product, system and / or method. In particular, the at least one sub-element and / or the at least one method step and / or the combination of at least two sub-elements is a characterising feature of the product, system and / or method, by means of which the product, system and / or method becomes novel, preferably patentable. This should be understood to mean that at least one sub-element of the described product, system and / or method is missing in the prior art or a combination of at least two sub-elements, in particular the disclosed prior art of a trained database, advantageously a local and / or global patent database, preferably in the entire prior art.Preferably, the metadata is collected during the processing of the input sequence to form the at least one sub-element, the at least one method step, and / or the combination of at least two sub-elements. In particular, the product, system, and / or method comprises a plurality of sub-elements, so that the product, system, and / or method, and in particular the structure and / or sequence of the product, system, and / or method, can preferably be clearly described. In particular, the number of sub-elements depends on the number of sub-elements required to describe the novelty of the product, system, and / or method. In particular, the input sequence is divided into sub-sequences.Preferably, the partial sequences each describe at least one function, one feature, the at least one method step and / or the at least one partial element and / or the combination of the at least two partial elements of the product, system and / or method. Metadata is preferably collected which has at least one property and / or advantage of the at least one partial element. In particular, when generating the output sequence, at least one output partial sequence is generated for each partial sequence of the input sequence, which partial sequence is linked in particular to the metadata. This advantageously allows the product, system and / or method to be recorded in further detail and advantageously allows the function of the product, system and / or method to be described.It is also proposed that the algorithm have at least one technology recognition model with which the product, system and / or method, in particular the at least one sub-element and / or the at least one method step, is detected. The technology recognition model preferably detects the technicality of the product, system and / or method during processing of the input sequence. Particularly preferably, the sub-elements are detected with the technology recognition model, preferably by extracting the sub-elements from the input sequence. Particularly preferably, the product, system and / or method, in particular the at least one sub-element, preferably all sub-elements, are classified with the technology recognition model.In particular, the product, system and / or method becomes advantageous in that it assigns the product, system and / or method, in particular the at least one sub-element, to a technology area or a technology group. Preferably, during the processing of the input sequence for the technology area of ​​the product, system and / or method, in particular the at least one sub-element, the metadata is collected with the technology recognition model, in particular from a database accessible to the algorithm, wherein preferably the product, system and / or method, in particular the at least one sub-element, is described with the collected metadata, in particular during the generation of the output sequence, preferably in the provided output sequence.Preferably, the technology recognition model is used to capture a structure of the product and / or system and / or a sequence of the system, in particular by capturing an arrangement of the sub-elements and / or a sequence of the method steps. Particularly preferably, the technology recognition model is used to determine the part of the metadata from the collected metadata. Particularly preferably, the algorithm forms at least the technology recognition model as a machine learning model with which at least the input sequence is processed. Particularly preferably, the classification of the sub-elements is captured using the technology recognition model as a machine learning model. Particularly preferably, the collected metadata is determined using the technology recognition model as a machine learning model. In particular, the technology recognition model is a support vector machine model and / or a naive Bayes model and / or a random forest model.Particularly preferably, the technology recognition model is a neural network, in particular a multi-input neural network, a deep graph network and / or a transformer neural network and / or a Naive Bayes classifier, a Support Vector Machine (SVM), a Latent Dirichlet Allocation (LDA), and / or a cross-encoder. Particularly preferably, the technology recognition model is provided for data processing. In particular, the technology recognition model is designed separately from the text generation model. Particularly preferably, the processed and / or collected metadata of the technology recognition model, in particular as part of the collected metadata, is provided to the text generation model, in particular for generating the output sequence.

[0024] Particularly preferably, the algorithm comprises the technology recognition model for collecting the metadata, in particular for processing the at least one input sequence, preferably for collecting and evaluating the metadata based on the processed input sequence from the at least one accessible database, and preferably a natural language model for generating the output sequence based on the part of the metadata from the technology recognition model. Preferably, the technology recognition model is provided solely for data processing, in particular processing the at least one input sequence, and / or data analysis, in particular evaluating the metadata, preferably determining the part of the metadata for generating the output sequence. This advantageously allows the technicality of the product, system and / or method, in particular of the sub-elements, to be recorded.Furthermore, the data quality of the metadata, in particular for generating the output sequence, can be advantageously increased with the technology recognition model, which is only intended for data processing.

[0025] It is further proposed that the sub-elements of the product of the input sequence be evaluated using the technology recognition model on the basis of structural features within the input sequence. The structural features are preferably format-specific and / or text-specific structural features. In particular, the sub-sequences are clearly separated from one another via the format-specific structural features. The format-specific structural features are preferably given by a structural arrangement of the sub-sequences of the input sequence. The format-specific structural features are preferably a control character and / or a special character, in particular punctuation marks. Particularly preferably, the format-specific structural features are at least one number as unique features, wherein the sub-sequences are preferably numbered consecutively.Alternatively or additionally, each subsequence comprises only one sentence, with punctuation in particular being recorded as a format-specific structural feature. In particular, the input sequence, preferably each subsequence of the input sequence, has at least one text-specific structural feature. Preferably, a unique assignment of the subelements and / or the method steps and / or the subsequences to one another is recorded on the basis of the text-specific structural features. In particular, the assignment of the subsequences is recorded by reference to at least one of the format-specific structural features of the subsequences, in particular by reference to the number of consecutively numbered subsequences. For example, the text-specific structural features are "according to claim [number]", and / or "according to claims [number] and [number]" and / or "according to one of the preceding claims" and / or features that are more comparable to a person skilled in the art.Preferably, the subsequences are structured hierarchically, in particular based on the assignment of the subsequences. Particularly preferably, an evaluation criterion is assigned to the subsequences based on the assignment, which preferably reflects the hierarchical structure of the subsequences. Preferably, the subelements and / or method steps are evaluated using assignment terms as text-specific structural features of the input sequence. For example, an assignment term of the input sequence is "characterized by" and / or "characterized in that" and / or "with" or "in a [number]. method step" and / or another assignment term that appears meaningful to a person skilled in the art. Preferably, the subelements and / or method steps are evaluated hierarchically based on the text-specific feature, in particular by assigning an evaluation criterion to the subelement and / or method step based on the assignment term.Preferably, each subelement is uniquely indexed based on the evaluation criterion for the assignment of the subsequences and / or the evaluation criterion according to the assignment concept. In particular, the algorithm, in particular the technology recognition model, is designed to assign the evaluation criterion to the subelement, and in particular to the subsequences. This advantageously allows the product, system, and / or method, which is particularly novel for an algorithm and preferably patentable, to be reconstructed based on a textual input sequence. In particular, the assignment of the individual subelements to one another, in particular the interaction of these subelements, can be advantageously captured.Furthermore, the computing power can advantageously be reduced in order to reconstruct a product, system and / or method which is missing in particular in a trained database, since the individual sub-elements are advantageously clearly identified with respect to one another and their hierarchical structure is recorded.

[0026] It is further proposed that the product, method and / or system, in particular the at least one sub-element, be described in abstract form in the introductory sequence. An “abstract form” should be understood in particular to mean that the product, system and / or method is defined by an abstract, preferably descriptive, term and / or an abstract description which applies in particular to a technology area or technology group. In particular, the introductory sequence in abstract form is text-based. Preferably, the introductory sequence describes only the novel, preferably patentable, core of the product, system and / or method, with the outgoing sequence preferably describing the product, system and / or method in further detail, in particular with properties, advantages and / or alternative implementations.Particularly preferably, a multitude of concrete, in particular technical, advantageously implementable, implementations can be derived from the abstract form of the product, system and / or method. Particularly preferably, the abstract form of the product, system and / or method, in particular of the input sequence, is an abstracted description of a technically preferred implementation of the product, system and / or method. In particular, an input sequence is entered which defines and delimits the scope of protection of the product, system and / or method, which is novel at least for one algorithm and is preferably patentable. The input sequence preferably reflects the essential core of the novelty of the product, system and / or method. Particularly preferably, the input sequence describes an invention, in particular an inventive concept, in particular in the form of claims of the product, system and / or method.Preferably, the input sequence, in addition to its abstract form, also meets the requirements of clarity of disclosure, novelty, and, in particular, inventive step of the product, system, and / or method. This advantageously allows a patent application to be output as the output sequence. Furthermore, a technology area can advantageously be efficiently captured using the technology recognition model.

[0027] Furthermore, it is proposed that at least one concrete input sequence be entered, which describes the product, system, and / or method in concrete form. A "concrete form" is to be understood, in particular, as the concrete input sequence describing a technically advantageously applicable implementation of the product, system, and / or method of the input sequence. In particular, the concrete input sequence is a text-based description of the product, system, and / or method. Preferably, the concrete input sequence is entered together with the input sequence in abstract form.

[0028] In particular, the concrete input sequence is a particularly preferred concrete, in particular technical, preferably implementable, in particular inventive, implementation of the product, system and / or method, preferably of the product, method and / or system, in particular of the input sequence, in abstract form. In particular, the concrete input sequence is an invention disclosure of the product, system and / or method that is novel, at least for one algorithm, and preferably patentable. Preferably, the concrete input sequence can be assigned to a technology area, in particular to the technology area of ​​the input sequence in abstract form. In particular, the metadata is collected based on the input sequence in abstract form and is preferably at least partially determined and / or restricted based on the concrete input sequence.Particularly preferably, the specific input sequence comprises advantages and / or properties of the product, system and / or method. Particularly preferably, the advantages and / or properties of the specific input sequence are recorded, in particular by the technology recognition model, for determining and / or restricting the metadata. Preferably, the specific output sequence is at least partially, preferably completely, included in the generation of the output sequence. By taking into account the specific input sequence, in particular as an invention disclosure of the novel, preferably patentable, product, system and / or method, the part of the collected metadata can advantageously be determined based on the preferred implementation of the product, system and / or method. This can advantageously ensure higher data quality when generating the output sequence.

[0029] It is further proposed that at least one graphic input sequence is entered which graphically represents the product, system and / or method and / or the at least one sub-element of the product, system and / or method, in particular in the form of a technical drawing. Alternatively or additionally, the graphic input sequence is a schematic representation of the product and / or system and / or a diagram, in particular a flow chart, of the method. In particular, the graphic input sequence graphically represents the novel, preferably patentable, core of the product, system and / or method. Particularly preferably, a structure of the product and / or system and / or a sequence of the method, in particular an arrangement of the sub-elements and / or a sequence of method steps, is represented on the basis of the graphic input sequence.Preferably, the structure and / or the sequence are captured using the algorithm, in particular using the technology recognition model. Preferably, during processing of the input sequence, the at least one sub-element and / or the at least one method step are evaluated based on the graphical input sequence in a processing step. Particularly preferably, the structure and / or the sequence are compared with the structural features of the input sequence, and preferably, the evaluation variables of the sub-elements and / or the method steps are further evaluated. Preferably, during generation of the output sequence, at least part of the output sequence is constructed based on the structure and / or the sequence.Particularly preferably, a description, in particular a figure description, of the graphic input sequence is generated based on the graphic input sequence, in particular based on the structure of the product and / or system and / or the sequence of the method, in particular the structure of the sub-elements and / or the sequence of the method steps. In an alternative embodiment of the method according to the invention, in particular if a graphic input sequence is missing, a graphic representation, in particular a structure of the sub-elements and / or a sequence of the method steps, is generated in the output sequence based on the input sequence, in particular based on the structural features of the input sequence. This advantageously allows the structure of the product and / or system and / or the sequence of the method to be recorded in more detail.

[0030] Furthermore, it is proposed that the input sequence be linked to the graphical input sequence. The graphical input sequence preferably has at least one reference symbol that uniquely identifies the product, system, and / or method, in particular the at least one sub-element and / or the at least one method step. The at least one reference symbol is preferably alphanumeric. Particularly preferably, the input sequence has the reference symbols of the graphical input sequence. In particular, the product, system, and / or method, in particular the at least one sub-element and / or the at least one method step, of the graphical input sequence can be uniquely identified using the reference symbols of the input sequence.Preferably, the structure of the product and / or the system and / or the process sequence, in particular an arrangement of the sub-elements and / or a sequence of the process steps, is detected using the graphical input sequence based on the reference symbols of the input sequence. Particularly preferably, when generating the output sequence, the at least one sub-element and / or the at least one process step is linked to reference symbols. This advantageously allows the detection of the sub-elements and / or the process steps in the input sequence with the technology recognition model with less computing power. Furthermore, an output sequence can advantageously be generated which, at least in a figure description, refers to the graphical input sequence, in particular to the technical drawings.

[0031] It is further proposed that at least one further input sequence be entered which describes a further product, system and / or method, in particular one which is known to the public. By “known” it should be understood in particular that the further product, system and / or method, in particular at least one sub-element and / or at least one method step of the further product, system and / or method, which is present in at least one further input sequence, in particular within sequences of a trained database, advantageously a local and / or global patent database, is preferably present in the complete disclosed prior art, in particular is disclosed. The at least one further input sequence is preferably a detailed description of the further product, system and / or method, in particular a patent specification and / or application specification.Particularly preferably, the further input sequence is the closest prior art of the product, system and / or method of the input sequence, which is novel at least for one algorithm and is preferably patentable. In particular, a plurality of further input sequences are entered which form the closest prior art to the input sequence. Preferably, a number of the further input sequences depends on the product, system and / or method of the input sequence. In particular, the number of further input sequences is defined by the user. In at least one embodiment of the inventive method, a user enters a difference between the input sequence and the at least one further input sequence, in particular with regard to various sub-elements and / or method steps and / or advantages and / or functions of the respective products, systems and / or methods.In particular, the difference between the input sequence and the at least one further input sequence at least partially forms the novel, preferably patentable, core of the input sequence. Metadata collected during the processing of the input sequence relating to the difference is preferably given priority when determining the portion of the collected metadata. Particularly preferably, advantages and / or properties of at least partial elements and / or method steps of the product, system, and / or method of the input sequence that differ from the further partial elements and / or method steps of the further product of the at least one further input sequence are given priority when determining the portion of the collected metadata.As a result, the closest prior art to the product, system and / or method of the input sequence can advantageously be incorporated into the processing and generation of the output sequence, and in particular, can be taken more into account.

[0032] It is also proposed that, in particular with the algorithm, at least one, in particular private, property database is accessed, which stores metadata, in particular information on, preferably technical, properties, of at least one element. A "private database" should be understood in particular to mean that the database is only accessible by an authorized user or authorized members of an organization. Preferably, the private database is hosted on a private server that is protected by authentication and access controls. Particularly preferably, the database is hosted on a private dedicated server, a local server, a virtual private server, a private cloud and / or a managed server. In particular, the property database at least partially contains the metadata, particularly preferably the advantages and / or properties of a large number of elements.The property database preferably has a plurality of property entries, each of which describes an element of the property database. Preferably, the algorithm collects the metadata from the property database when processing the input sequence. In particular, the at least one element of the property database is a product, system and / or a sub-element of products and / or systems, in particular at least partially the same sub-element of the product, system and / or method of the input sequence. Preferably, the property database has definitions for the elements. Particularly preferably, the property database has method steps, in particular with properties and / or advantages. Preferably, the sub-elements of the input sequence are recorded using the technology recognition model and passed to the algorithm for collecting the metadata from the property database.Particularly preferably, the technology areas of the product, system, and / or process, in particular of the at least one sub-element and / or process step, are transferred to the algorithm for collecting the metadata from the properties database. The algorithm preferably compares the at least one sub-element of the product, system, and / or process of the input sequence with the elements of the properties database and preferably collects the metadata for which the sub-element matches the element, in particular a technology area of ​​the sub-element matches a technology area of ​​the element. Particularly preferably, the metadata of the properties database is transferred from the algorithm to the technology recognition model for processing the metadata, in particular for evaluating the metadata.Preferably, during the generation of the output sequence, the metadata, in particular the advantages and / or properties of the elements, of the property database are linked to at least one of the sub-elements of the product, system and / or process. Preferably, the property database comprises at least one checklist for at least one technology area, preferably at least partially the technology areas of the product, system and / or process, in particular of the at least one sub-element and / or process step. Particularly preferably, the checklist defines rules for the advantages and / or properties of the technology areas of the product, system and / or process, in particular of the at least one sub-element and / or process step.In particular, based on the rules of the checklist, elements and / or process steps of the property database, in particular with the properties and / or advantages of the elements and / or processes, are passed to the algorithm for generating the output sequence, particularly as metadata. This advantageously reduces the computing power required to collect the metadata. Furthermore, using a private database to collect the metadata can advantageously increase data security.

[0033] Furthermore, it is proposed that, in particular with the algorithm, at least one, in particular private, sequence database is accessed, containing a plurality of, in particular text-based, sequences, each describing a further product, system and / or method, in particular with metadata relating to the product, system and / or method. The sequence database is preferably hosted on the, in particular private, server, in particular with the properties database. In particular, the server has at least one overall database containing at least the properties database and / or the sequence database. The sequences are preferably descriptions, in particular patent applications, of further products, systems and / or methods. In particular, a “plurality of sequences” should mean at least two, preferably at least ten, advantageously at least 100, particularly preferably at least 1,000, and particularly advantageously at least 10.000 sequences. The sequences of the sequence database are particularly preferably preselected by a user and / or an organization. The source sequence is particularly preferably stored in the sequence database when the source sequence is provided. Preferably, at least all source sequences that are processed by the algorithm, in particular according to the computer-implemented method, are stored in the sequence database. Preferably, the algorithm collects the metadata from the sequence database when processing the input sequence. Preferably, the algorithm is used to compare the product, system and / or method, preferably the technology area, with the products, systems and / or methods of the sequence database, preferably with the technology areas of the sequence database, wherein the metadata is preferably collected for the sequences that have the same technology area.In particular, the technology recognition model is used to capture the at least one sub-element and / or the at least one method step of the product, system and / or method of the sequence, and the advantages and / or properties are collected as metadata. Preferably, the sub-elements of the input sequence are captured using the technology recognition model and transferred to the algorithm for further processing. Particularly preferably, the technology areas of the product, system and / or method, in particular of the at least one sub-element and / or method step, are transferred to the algorithm for collecting the metadata from the sequence database. Preferably, the algorithm compares the at least one sub-element with the sub-elements of the sequences of the sequence database and preferably collects the metadata for which the sub-elements match.Particularly preferably, the metadata of the sequence database is passed from the algorithm to the technology recognition model for processing, in particular for determining the portion of the collected metadata. This advantageously allows for the provision of a higher-quality output sequence. Furthermore, higher-quality metadata can advantageously be collected with the sequences.

[0034] Furthermore, it is proposed that the sequences in the sequence database be divided into classes and that the input sequence be assigned to one of the classes based on metadata of the input sequence. In particular, the metadata of the input sequence for assigning the input sequence to the, in particular, associated class is entered when the input sequence is entered. Alternatively or additionally, the, in particular associated, class of the input sequence is recorded when the input sequence is processed, preferably by parsing the input sequence. Preferably, the metadata of the input sequence for assigning it to a class is a company name, in particular a client name, and / or a technology class of the input sequence. Preferably, the sequence database is divided into classes, in particular company classes, in at least part of the database.The sequence database preferably comprises further sequences which are divided into further classes, in particular technology classes. Particularly preferably, the sequences of the further classes, in particular technology classes, are missing from the classes, in particular from the company classes. In at least one embodiment of the method according to the invention, a user and / or an organization can evaluate the sequences of the sequence database, in particular in the respective class, preferably with regard to relevance. Preferably, the metadata, in particular the advantages and / or properties, of the sequences of the sequence database are given different priority when determining the portion of the collected metadata, depending on the evaluation during processing of the input sequence.In at least one further embodiment of the method according to the invention, the sequences of the sequence database, in particular in the respective class, in particular company class, are divided by the user and / or the organization into categories, in particular technological ones. Particularly preferably, during processing of the input sequence, the metadata is collected from the sequences of the sequence database for which the technology area of ​​the input sequence, in particular the technology area of ​​the product, system, and / or process of the input sequence, corresponds to the category, in particular technological, of the assigned class. This advantageously allows the data quality of the collected metadata to be further increased.

[0035] Furthermore, it is proposed that the metadata of the product, system, and / or method be at least partially extracted from the assigned class of the sequence database. Particularly preferably, when processing the input sequence, only metadata from, in particular, the sequences in the sequence database that correspond to the assigned class, in particular the assigned company class, of the input sequence are collected. Preferably, metadata, in particular the advantages and / or properties as metadata, of the sequences in the sequence database with the same further class, in particular the technology class, are given priority when determining the portion of the collected metadata. This can advantageously improve the data quality, in particular the quality of the collected metadata.Preferably, when determining the portion of the collected metadata for generating the output sequence, only the advantages and / or properties of the sequences in the sequence database that correspond to the assigned class, in particular the assigned company class, of the input sequence are used. Particularly preferably, the output sequence is stored in the class, in particular company class, that corresponds to the assigned class in the sequence database of the input sequence upon provision. Particularly preferably, the technology recognition model is trained separately for each class, in particular company class, of the sequence database. This advantageously increases security, since only advantages and / or properties of sequences of the assigned class, and thus of the same company, are used when processing the input sequence.Furthermore, the data security of the technology recognition model can be advantageously increased, since it is trained according to the individual classes with sequences from the sequence database.

[0036] Furthermore, it is proposed that trend properties for the respective classes be stored in the sequence database, which are then transferred to the technology recognition model for processing the input sequence with the algorithm. The trend properties are preferably stored, in particular by a user, in the sequence database for the respective class. In at least one embodiment of the method according to the invention, the trend properties of each class are determined using the algorithm, in particular using the technology recognition model, with the respectively assigned sequences. Preferably, the algorithm, in particular using the technology recognition model, is used to analyze, in particular parse, the respectively assigned sequences of the respective class according to a predominant trend property.The trend properties are preferably properties and / or technology areas that are applicable to a technical product, system and / or process, in particular across technologies, and / or that can be combined with a technical product, system and / or process. This should be understood in particular that a product, system and / or process, in particular the product, system and / or process of the input sequence, has the trend property as a property of the product, system and / or process through application and / or combination with the trend property. The trend properties are particularly preferably properties and / or technology areas that are in focus in the class of the sequence database, preferably in the focus of the company of the class of the sequence database.Preferably, the trend properties are, in particular for the respective company, promising technology areas and / or properties, preferably sustainable technologies, such as sustainable raw materials and / or sustainable processes, and / or digital technologies, such as machine learning, simulations, blockchain, 5G, quantum computing and / or cloud computing and / or another technology that appears appropriate to a person skilled in the art, particularly in the class of the sequence database. Preferably, the trend properties are different from the technology areas of the sequences, in particular partially different from the technology area of ​​the product, system and / or process of the input sequence. Particularly preferably, when processing the input sequence, the collected metadata that at least partially exhibit the trend properties are given preference when determining the part of the collected metadata.This advantageously allows for more technologies and / or properties that are the focus of the respective company to be included in the output sequence. Furthermore, technology areas that are outside the expertise of the user, especially the inventor, and preferably the company, can be advantageously included in the output sequence, particularly through a combination with the product, system, and / or process of the input sequence.

[0037] It is further proposed that at least one implementation of the product, system and / or process and / or of the at least one sub-element be derived using the technology recognition model based on the abstract form of the product, system and / or process with the metadata. In particular, the implementation of the technology recognition model is a further implementation of the product, system and / or process that is different from the implementation of the concrete input sequence and is preferably comparable to the implementation of the concrete input sequence at least in terms of a level of technical detail, in particular a level of abstraction and / or applicability. In particular, the implementation is derived based on the technology areas of the product, system and / or process and / or of the at least one sub-element.Particularly preferably, the implementation is an embodiment of the product, system and / or method, in particular an embodiment of the at least one sub-element and / or method step, which is within the corresponding technology field. Preferably, the implementation of the product, system and / or method and / or of the at least one sub-element is at least one level of abstraction lower than the abstract form of the product, system and / or method and / or of the at least one sub-element. Particularly preferably, the implementation is a concretization of the product, system and / or method in abstract form and / or an alternative embodiment of the product, system and / or method, in particular of the at least one sub-element, preferably within the scope of the novelty or patentability of the product, system and / or method according to the input sequence.Particularly preferably, at least the structural features within the input sequence are processed to derive the implementation. Preferably, an alternative embodiment is derived for at least one sub-element according to the hierarchical structure by detecting a sub-element of a higher hierarchy of the sub-element and thereby deriving a different sub-element of a lower hierarchy, particularly in relation to the sub-element of a lower hierarchy. This advantageously allows various concrete, novel implementations to be generated starting from an abstract product, system, and / or method.

[0038] It is further proposed that the technology recognition model links the metadata with the sub-elements of the product, system and / or process, and that the implementation is derived using property relationships and / or advantage relationships of the metadata. Preferably, a property relationship and / or an advantage relationship is a relationship between at least two sub-elements which at least partially, preferably completely, have the same property and / or the same advantage. Preferably, the implementation is derived based on the properties and / or advantages of the product, system and / or process, particularly preferably of the sub-element and / or process step, of the input sequence. Particularly preferably, at least one concretization is derived based on the properties and / or advantages of the metadata for the at least one sub-element and / or process step.A concretization of a property of a sub-element, particularly one formulated abstractly, is, for example, "the design of an operating element as a touch control element and / or a button element and / or a folding element and / or another element suitable for operation," where the touch control element and / or the button element and / or the folding element are the concretization based on a property. A concretization of an advantage of a sub-element, particularly one formulated abstractly, is, for example, "an operating element which is advantageously fire-resistant, in particular made of a fire-resistant material such as stainless steel, aluminum and / or copper," where the fire-resistant material is the concretization of the material based on the advantage of fire resistance.Preferably, an alternative design is derived based on the at least one sub-element of the product with the property and / or advantage, in particular based on the structural features at the same hierarchical level. An alternative design of a sub-element based on a property and / or advantage is, for example, recycled plastic, which is particularly sustainable, or bioplastics, which are also considered particularly sustainable. Particularly preferably, a large number of properties and / or advantages are combined for the concretization in order to limit the number of concretizations and / or the number of alternative properties. Particularly preferably, the implementations are determined based on the sequences of the sequence database and / or the at least one further input sequence.This makes it possible to generate particularly advantageous and relevant alternative implementations starting from an abstract product, system and / or process.

[0039] It is also proposed that the technology recognition model be used to generate a relationship matrix based on the property relationship and / or advantage relationships of the metadata of the sub-elements, which creates a link between at least the sub-elements of the product, system and / or method, in particular the sub-elements of the sequences of the sequence database. The relationship matrix is ​​to be understood in particular as a data construct that defines the collected metadata, in particular with the implementations of the product, system and / or method, preferably the sub-elements, and preferably a relationship between the respective metadata, in particular the implementations. The relationship matrix preferably represents a unique assignment of the metadata to one another and to the sub-elements of the product, system and / or method of the input sequence.Preferably, the relationship matrix is ​​supplemented with the at least one further input sequence, in particular the further product, system and / or method, in particular their further sub-elements. In particular, a relationship matrix is ​​generated with the implementations, in particular with the concretizations and / or the alternative embodiments. Particularly preferably, a plurality of relationship matrices are stored in the private database, in particular in the sequence database, preferably in the respective classes of the sequence database. Particularly preferably, based on the at least one sub-element of the input sequence with the technology recognition model, at least one relationship matrix is ​​selected which comprises the sub-element. This advantageously allows a relationship between sub-elements to be generated and, in particular, documented.Furthermore, data processing speed can be advantageously increased since the conversions are already present in a generated relationship matrix.

[0040] In addition, it is proposed that the links in the relationship matrix be weighted at least based on the metadata of the input sequence. In particular, the link of the at least one implementation to the product, system and / or method, in particular to the sub-element and / or method step, is weighted. In particular, the links between the implementations are weighted based on the advantages and / or properties of the product, system and / or method, in particular of the at least one sub-element and / or method step. Particularly preferably, the links are weighted as a function of the trend properties, the sequences of the sequence database, the further input sequence, the properties and / or advantages of the properties database and / or the at least one checklist for the technology area of ​​the product, system and / or method, in particular as influencing factors.When weighting the link, each influencing factor is preferably given varying degrees of weight. This allows the determined implementations of the product, system, and / or process to be weighted advantageously, particularly based on the given parameters.

[0041] Furthermore, it is proposed that, based on the weighting of the links, at least the implementation of the product, system, and / or method, in particular of the at least one sub-element, be integrated into the generation of the output sequence. Particularly preferably, the portion of the collected metadata is determined based on the weighting. Particularly preferably, the portion of the collected metadata comprising the implementations of the product, system, and / or method, in particular of the at least one sub-element, is integrated into the generation of the output sequence. Preferably, each associated implementation is weighted for each sub-element of the product, system, and / or method from the input sequence.Preferably, the implementations, in particular of the respective sub-element of the product, system, and / or method of the input sequence, are co-generated during the generation of the output sequence, which has the strongest weighting of the link, in particular after a weighting, preferably numerical. Particularly preferably, for at least a majority of the sub-elements and / or a majority of the process steps of the product, system, and / or method, an implementation, in particular based on the weighting of the link for the respective sub-element and / or the respective process step, is integrated during the generation of the output sequence.Particularly preferably, for the at least one sub-element and / or for the at least one method step, a plurality of, preferably at least two, advantageously at least three, and particularly preferably at least four, implementations forming part of the collected metadata for generating the output sequence are output from the technology recognition model to the algorithm. Particularly preferably, the number of integrated implementations is adapted depending on the technology area of ​​the sub-element and / or method step.Preferably, for the sub-elements that are captured in the input sequence, preferably in a sub-sequence of the input sequence, particularly preferably in an independent claim of the input sequence, and / or have a technology area that is at least partially a trend property, a number of implementations as part of the collected metadata, in particular for generating the output sequence, is increased by at least one, preferably at least two, particularly preferably at least three implementations. Particularly preferably, the relevance of the respective sub-element is determined based on the stated advantages of the specific input sequence, with the sub-elements with a stated advantage having the highest relevance.In at least one embodiment of the invention, the portion of the collected metadata comprising the implementations and the sub-elements and / or method steps of the product, system and / or method is combined in a text-based manner with a K1 text generator, such as a GPT model, during the generation of the output sequence. This advantageously allows an output sequence to be generated which describes an advantageous implementation, in particular a concretization and / or alternative embodiment, for the product, system and / or method, in particular for the at least one sub-element and / or the at least one method step. Furthermore, a patent application with a description of the claims and alternative embodiments can advantageously be generated.

[0042] It is further proposed that, in particular with the algorithm, at least one further, in particular public, sequence database, preferably a patent database, be accessed which has a large number of sequences. In particular, the further sequence database is freely accessible, in particular via the Internet. In particular, a "large number of sequences" in this context is to be understood as at least 100, preferably at least 10,000, advantageously at least 1,000,000 and particularly preferably at least 100,000,000 sequences. Preferably, at least a large part of the current state of the art, in particular public patents, patent applications, utility models and other intellectual property rights, is stored in the sequence database. Preferably, the current state of the art is recorded by the algorithm as a large number of sequences. Preferably, further implementations of the product of the input sequence are determined using the large number of sequences.This can advantageously increase the data quality, since a larger number of sequences can be used to process the input sequence.

[0043] It is also proposed that the algorithm generates a sub-element data set based on the product, system, and / or method, by means of which a subset of the plurality of sequences is transferred to the, in particular private, sequence database for generating a further relationship matrix. Preferably, the sub-element data set is generated for a sub-element that is missing and / or insufficiently present in the sequence database and / or the property database. In particular, further advantages and / or properties of the sub-element and / or method step and / or alternative embodiments and / or further specifications are recorded. The sub-element data set is preferably determined by an overarching technology term of the at least one sub-element and / or method step.Particularly preferably, the novel, preferably patentable, core of the product, system, and / or method of the input sequence is removed with the overarching technology term during the generation of the sub-element data set. Preferably, the sub-element data set, in particular the overarching technology term, is an IPC class and / or CPC class and / or another overarching technology term of the product, system, and / or method, in particular of the at least one sub-element and / or method step. In particular, the sub-element data set is used to determine the subset of the plurality of sequences of the further sequence database.Preferably, the subset of the plurality of sequences in the further sequence database is at least a majority of the sequences, in particular at least a majority of the intellectual property rights, preferably at least a majority of the patent applications, patents and / or utility models, of the overarching technology concept, in particular the IPC class and / or CPC class of the further sequence database. Particularly preferably, the subset of the plurality of sequences in the further sequence database is transferred to the, in particular private, sequence database for generating the further relationship matrix. Particularly preferably, the subset of the plurality of sequences in the further sequence database is processed with the technology recognition model. The further relationship matrix preferably comprises the links between the products, systems and / or methods, in particular the subelements, of the subset of the plurality of sequences in the further sequence database.In particular, the further relationship matrix at least partially forms the collected metadata from which the portion of the collected metadata is determined. In particular, after the further relationship matrix has been generated, the subset of the plurality of sequences in the further sequence database is removed from the sequence database. This advantageously increases data security, since no information from the input sequence, in particular the claims, is sent to a public database. Furthermore, deleting the sequences can advantageously optimize storage space in the sequence database.

[0044] In a further embodiment of the invention, it is proposed that the output sequence be generated sequentially, in particular based on the portion of the collected metadata. This advantageously allows the quality of the output sequence to be further optimized. In particular, the portion of the collected metadata can be enlarged, and the sequential generation can further restrict the portion of the collected metadata sequentially. Furthermore, due to the sequential adaptation to only a portion of the collected metadata, an adaptation of the output sequence based on a reduced data basis can advantageously be provided, which leads to accelerated data processing. The output sequence is particularly preferably provided sequentially, in particular via the user interface.Sequential generation should be understood in particular to mean that the output sequence is generated in an iterative process, in particular wherein the generation of the output sequence occurs sequence by sequence and preferably after each sequence an evaluation factor is generated and included, which in particular influences the generation of the subsequent sequence. Preferably, the sequential evaluation factor is assigned to the part of the collected metadata. Preferably, the evaluation factor is entered by an operator via the user interface. Preferably, the sequential evaluation factor is assigned to each property, each advantage and / or each implementation of the product, system, method and / or sub-element, in particular in the part of the collected metadata.Particularly preferably, the evaluation factor, in particular the algorithm, is used to perform data refinement, preferably of the portion of the collected metadata, during the sequential generation of the output sequence. Particularly preferably, the evaluation factor is taken into account for each sequence during the generation of the output sequence, so that metadata from the portion of the metadata is prioritized, in particular included or removed, during the generation of the further sequences of the output sequence.

[0045] In an alternative embodiment of the invention, the input sequence is processed sequentially. This advantageously allows the quality of the output sequence to be further optimized. Preferably, the input sequence is processed sequentially, in particular iteratively, for each subsequence, each subelement, and / or each method step. In particular, the output sequence is generated and preferably provided by the sequential processing of the input sequence, wherein the provided output sequence, in particular via the user interface, preferably assigns an evaluation factor to each property, each advantage, and / or each implementation of the product, system, method, and / or subelement, which is incorporated, in particular iteratively, into the processing of the input sequence.In particular, during the iterative processing of the further input sequence, the metadata is further weighted to determine the part of the collected metadata at least based on the evaluation factor.

[0046] Particularly preferably, an evaluation mask is created based on the evaluation factor from the sequential generation of the output sequence or the sequential generation of the input sequence, wherein, in particular within the evaluation mask, each evaluation factor is assigned to the metadata, in particular to each property, each advantage, and / or each implementation of the product, system, method, and / or sub-element. This advantageously provides data refinement during the processing of the input sequence and / or the generation of the output sequence, which advantageously enables a reduction in the amount of data to be processed by eliminating redundant data. Furthermore, the data access time can advantageously be reduced due to the sequential generation of the output sequence or the sequential processing of the input sequence.Furthermore, the reliability and accuracy of the computer-implemented method can be advantageously increased, since anomalies and / or redundant data during processing and / or generation are advantageously detected and sequentially corrected and / or removed by providing the input sequence. Preferably, the evaluation mask is stored in the sequence database, in particular company-specific and / or person-specific. Particularly preferably, the evaluation mask is included when entering a further input sequence, preferably during processing of the input sequence, in particular as a weighting of the metadata.

[0047] In a further embodiment of the invention, a data lake is generated from the metadata, preferably company-specific. This advantageously allows a company-specific data profile to be generated, which advantageously increases the quality and accuracy of the generated output sequence and enables efficient and effective use of the data by the operator. Furthermore, a continuously growing company-specific database can advantageously be created, with the know-how being continuously stored in a data profile. Particularly preferably, the metadata is collected by the data lake during processing of the input sequence. Particularly preferably, at least one company-specific data profile is generated with the data lake.Preferably, the data lake is generated from at least the sequences of at least one class of the sequence database, at least one, preferably company-specific, property from the property database, the created relationship matrices, the company-specific evaluation mask, and / or other collected metadata. In particular, further newly incoming sequences, in particular printed documents, for example patents or patent applications, preferably through competition monitoring, technology monitoring, proprietary applications, searches, and / or the like, and / or further new properties of the property database and / or further new metadata are added to the data lake, in particular continuously. Data integration is preferably performed using ETL (Extract, Transform, Load) processes, which in particular extract, transform, and load the metadata from the various sources into the data lake.In particular, the metadata is harmonized and / or standardized during the transformation. Within the data lake, the integrated metadata is preferably processed through additional algorithms and / or analysis tools, including but not limited to machine learning, statistical analyses, and heuristic algorithms, to generate a company-specific data profile. This company profile includes, in particular, calculated key performance indicators and indicators, aggregated and condensed information that represents, for example, the company's strengths, weaknesses, opportunities, and threats.

[0048] Furthermore, a computer-implemented method for training the machine learning algorithm is presented, comprising the following steps:

[0049] Inputting at least one input sequence into the machine learning algorithm which describes a product, system and / or method;

[0050] - processing the input sequence by collecting metadata about the product, system and / or process using the algorithm; and

[0051] - Adaptation of algorithm parameters based on at least part of the collected metadata is proposed.

[0052] Particularly preferably, the algorithm is trained for inputting and processing the at least one input sequence, which preferably describes a product, system and / or method that is novel, at least for the algorithm, and preferably patentable, and is provided for generating an output sequence, in particular one that is at least partially text-based. Preferably, the input sequence for training the algorithm describes sub-elements and / or method steps of the product, system and / or method that define a structure of the product, system and / or method. Particularly preferably, the input sequence is entered in text-based abstract form, preferably into the algorithm, preferably in the form of claims. In particular, the input sequence is stored in the sequence database and is entered for training the algorithm.Preferably, a plurality of input sequences, in particular from at least a portion of the sequences in the sequence database, in particular from one of the classes, preferably company classes, of the sequence database, are input to train the algorithm, in particular into the algorithm. The input sequences are preferably a description of the core of a product, system, and / or method of the sequences, in particular claims of the sequences, in the sequence database. Preferably, the at least one input sequence, in particular the plurality of input sequences, form training data for training the algorithm.Preferably, the sequences of the sequence database have a validation part for training the algorithm, which has at least one detailed description of the respective product, system, and / or method of the sequences of the sequence database, in particular associated with the input sequence, in particular a description of the claims and / or a description of the figures of the sequences, which are provided in particular for validating the collected metadata of the algorithm. Particularly preferably, the validation part of the sequences describes the advantages and / or properties and / or implementations, in particular as comparison values ​​for validating the part of the collected metadata, the product, system, and / or method of the input sequence of the respective sequence.In particular, the parameters of the algorithm are adjusted based on the validation of the part of the collected metadata of the respective input sequence with the respective validation part of the sequences.

[0053] In particular, the technology recognition model of the algorithm collects the metadata relating to the input sequence for training purposes. Particularly preferably, the algorithm parameters are adapted by validating the collected metadata of the technology recognition model to the product, system, and / or method of the input sequence, in particular to the sub-element and / or method step of the product, system, and / or method of the input sequence. Particularly preferably, the technology recognition model, in particular a Kl module of the technology recognition model, of the algorithm is trained. Particularly preferably, parameters of the technology recognition model, in particular of the Kl module, of the algorithm are adapted.Preferably, the validation is performed by calculating a difference between the collected metadata, in particular the part of the collected metadata, the technology recognition model, and the further validation part of the sequences, in particular the detailed description of the product. Particularly preferably, the validation part of the sequences for validating the part of the collected metadata of the algorithm corresponds at least substantially, in particular at least format-specifically, to an output sequence of a trained algorithm.

[0054] Preferably, in at least one step for training the algorithm, in particular the algorithm's technology recognition model, the input sequences, in particular as training data, are divided into smaller subsets. In particular, the subsets are fed to the algorithm one after the other. This advantageously reduces the computational load and makes training more efficient.

[0055] Preferably, in at least one further step for training the algorithm, in particular the technology recognition model of the algorithm, metadata is collected for the input sequences with the current parameters of the algorithm, in particular the technology recognition model. In particular, a portion of the collected metadata is determined from the collected metadata using the technology recognition model. In particular, the portion of the collected metadata is determined using predefined parameters of the algorithm, in particular the technology recognition model, in particular parameters preset by a user. The portion of the collected metadata is provided for validation.

[0056] Preferably, in at least one further step for training the algorithm, in particular the algorithm's technology recognition model, a loss function is calculated for the portion of the collected metadata of the algorithm, in particular the algorithm's technology recognition model. In particular, the loss function specifies the difference between the portion of the collected metadata and the validation portion of the sequences for the respective input sequence. Preferably, the loss function is used to calculate a quality of the portion of the collected metadata, in particular in comparison to the validation portion of the sequences.

[0057] Preferably, in at least one further step for training the algorithm, in particular the technology recognition model of the algorithm, a gradient of a loss is calculated with respect to each parameter of the algorithm, in particular the technology recognition model of the algorithm. Preferably, the gradient of the loss is calculated using the derivative of the loss function. This advantageously makes it possible to determine how sensitive the loss is to changes in the individual parameters of the algorithm, in particular the technology recognition model, and to provide information on how these must be adapted to minimize the loss. Preferably, in at least one further step for training the algorithm, in particular the technology recognition model, an optimization algorithm is applied to adapt the parameters of the algorithm, in particular the technology recognition model, preferably based on the gradient of the loss.Examples of optimization algorithms are Stochastic Gradient Descent, Adam or RMSprop.

[0058] Preferably, the steps of inputting, collecting, outputting, and adapting for training the algorithm are repeated cyclically, in particular until a number of cycles predetermined, in particular by a user, is reached, or a local or global optimum is achieved by adjusting the algorithm parameters, and in particular at least approximately no improvement in performance is recorded on the training data set, or another stopping criterion deemed reasonable by a person skilled in the art is met. In particular, with each cyclical repetition of the steps, an at least partially new portion of the collected metadata is determined for each input sequence from the collected metadata. This advantageously reduces the number of input sequences, in particular the required amount of training data.

[0059] Alternatively and / or additionally, the algorithm is trained unsupervised. Preferably, for unsupervised learning, a plurality of input sequences are input which describe a product, system and / or method, in particular in abstract form. The plurality of input sequences for unsupervised learning are sets of claims of the products, systems and / or methods, in particular of sequences from the sequence database. Particularly preferably, the metadata relating to the advantages and / or properties of the plurality of products, systems and / or methods, in particular relating to sub-elements and / or method steps of the plurality of products, systems and / or methods, are collected from the property database and / or the sequence database. In particular, during unsupervised learning, the portion of the collected metadata is determined from the collected metadata.In particular, the parameters are adapted, in particular calculated, by validating the portion of the collected metadata to the plurality of input sequences. Preferably, the evaluation detects a pattern in the portion of the collected metadata, in particular using clustering methods, with which the algorithm is trained.

[0060] The method according to the invention can advantageously be used to train a machine learning algorithm which can advantageously generate and provide an at least partially text-based output sequence with advantageously further metadata for a product, system and / or method, in particular one that is novel for an algorithm and is preferably patentable. Furthermore, an algorithm can advantageously be provided which, by collecting metadata, preferably as an advantage and / or property of the product, system and / or method, in particular compared to generating metadata, advantageously increases the speed of data processing and / or enables the use of a more cost-effective processor. Furthermore, an algorithm can advantageously be provided which generates and provides a patent application, in particular with claims as an input sequence.

[0061] Furthermore, a computer-implemented method for continuously training a machine learning algorithm is presented, comprising the following steps:

[0062] Input of at least one input sequence into the algorithm which describes a product, system and / or method which is novel, in particular at least for the algorithm, and preferably patentable;

[0063] - processing the input sequence by collecting metadata about the product, system and / or process using the algorithm;

[0064] - generating an output sequence based on at least part of the collected metadata, in particular at least partially text-based, by means of the algorithm;

[0065] - Providing the output sequence;

[0066] - Evaluation of the output sequence; and

[0067] - Adjustment of the algorithm parameters based on the evaluation, proposed.

[0068] In this context, “continuous training” should be understood in particular to mean that the parameters of the algorithm are continuously adapted during each run of the method by evaluating the output sequence. Preferably, the parameters of the algorithm are further optimized by each continuous, in particular each renewed, evaluation of each additional new output sequence. Particularly preferably, the evaluation of the output sequence is carried out by the user, in particular via the user interface. In particular, when evaluating the part of the collected metadata, the links of the advantages, properties and / or implementations to the product, system and / or method, in particular to the sub-element and / or method step of the input sequence, are evaluated. Particularly preferably, the evaluation is carried out for each link to each sub-element and / or each method step of the product, system and / or method.Preferably, the parameters of the algorithm, in particular of the technology recognition model, are adapted on the basis of the evaluation, wherein preferably the advantages, properties and / or implementations of the product, system and / or method, in particular of the sub-elements and / or method steps to which a positive evaluation is assigned, are given preference in a continuous, in particular repeated, implementation of the computer-implemented method, when determining the part of the collected metadata, and wherein preferably the advantages, properties and / or implementations of the product, system and / or method, in particular of the sub-elements and / or method steps to which a negative evaluation is assigned, are given disadvantage in a continuous, in particular repeated, implementation of the computer-implemented method, when determining the part of the collected metadata.Particularly preferably, a suitable, in particular technically correct, preferably applicable and / or combinable, property, advantage and / or implementation of the product, system and / or method, in particular of the sub-element and / or process step, of the part of the collected metadata is positively evaluated. Particularly preferably, an inappropriate, in particular technically incorrect, preferably faulty, property, advantage and / or implementation of the product, system and / or method, in particular of the sub-element and / or process step, of the part of the collected metadata is negatively evaluated.

[0069] In at least one embodiment of the invention, the relationship matrix, in particular an identifier of the advantages, properties, and / or implementations, which preferably marks the negative or positive evaluation, is adapted based on the evaluation, in particular a positive or negative evaluation. In particular, the advantages, properties, and / or implementations marked with the identifier are favored or disadvantaged when determining the portion of the collected metadata when the computer-implemented method is repeated and linked to the same subelement, depending on the evaluation.

[0070] In at least one embodiment of the method according to the invention, in which the generation of the output sequence is supported by a user, the linking of the metadata to at least one sub-element and / or method step is continuously weighted. In particular, a plurality of the advantages and / or properties of the product, system and / or method is displayed to the user as metadata collected, in particular by processing the input sequence, from which at least the part of the collected metadata is selected. The method of determining the subset of the collected metadata by the user influences the weighting of the links for determining the part of the collected metadata during continuous, in particular repeated, execution of the computer-implemented method, wherein the linking of the metadata selected by the user is preferably given greater weight.

[0071] The method according to the invention advantageously allows the machine learning algorithm, in particular the technology recognition model, to be continuously optimized. Furthermore, an at least partially text-based output sequence with advantageously further metadata can be advantageously generated and provided for a product, system and / or method, in particular for an algorithm that is novel and preferably patentable. By collecting metadata, preferably as an advantage and / or property of the product, system and / or method, in particular in comparison to generating metadata, the speed of data processing can advantageously be increased and / or more cost-effective processors can be used. Furthermore, the quality of the output sequence can advantageously be increased by generating the output sequence based on at least part of the metadata.Furthermore, a patent application, in particular with claims as an input sequence, can advantageously be generated and provided. Furthermore, the evaluation of the metadata can advantageously continuously generate improved properties, advantages, and / or implementations for the product, system, and / or method.In addition, a system for inputting and processing an input sequence and for generating and providing an output sequence, in particular an at least partially text-based one, and / or for training a machine learning algorithm for inputting and processing an input sequence and for adapting the parameters of the algorithm and / or for continuously training a machine learning algorithm for inputting and processing an input sequence and for generating, providing and evaluating an output sequence, in particular an at least partially text-based one, and for adapting the parameters of the algorithm is proposed, comprising: a computing device which is configured to carry out at least one of the methods.

[0072] Preferably, the input sequence is input to the computing device for processing. In particular, the computing unit comprises the algorithm, in particular machine learning. The algorithm in particular comprises a technology recognition model, which is preferably intended to capture a technicality, in particular a technology area, of the product, system and / or method. Particularly preferably, the technology recognition model is intended to capture at least one sub-element and / or method step of the product, system and / or method. Furthermore, the system has at least one, advantageously private, properties database. The properties database preferably contains a plurality of properties and / or advantages for a plurality of products, systems and / or methods, in particular for sub-elements and / or method steps.Alternatively and / or additionally, the property database comprises at least one checklist for a technology area. Furthermore, the system comprises at least one further, advantageously private, sequence database. Preferably, the sequence database comprises a plurality of, preferably at least partially text-based, sequences, in particular intellectual property rights, particularly preferably patent applications, patents and / or utility models. Particularly preferably, the sequences in the sequence database are divided into classes, in particular according to company and / or client. Furthermore, the algorithm, in particular the technology recognition model, is provided to create at least one relationship matrix which links, in particular, the advantages and / or properties and / or implementations of the product, system and / or method, in particular the sub-element and / or method step, in a weighted manner. The sequence database preferably comprises the relationship matrices.Particularly preferably, the sequence database has at least a plurality of relationship matrices relating to a plurality of products, systems, and methods, in particular sub-elements and / or method steps. In particular, the algorithm is provided to use at least the relationship matrix comprising the sub-elements and / or method steps of the input sequence. Furthermore, the system has at least one further, in particular public, sequence database with a plurality of further sequences. In particular, the algorithm is provided to create at least one relationship matrix with the plurality of sequences. The sequence database has the relationship matrices of the, in particular public, sequence database. Particularly preferably, the system has at least one user interface. The user interface is provided for parallel generation of the output sequence with the user.In particular, the user interface is designed to allow a user to evaluate the collected metadata. In at least one embodiment of the invention, the sequence database and the property database are part of an overall database, in particular a private one.

[0073] The system according to the invention can advantageously provide an at least partially text-based output sequence for a product, system and / or method, in particular for an algorithm that is novel and preferably patentable. Furthermore, by collecting metadata, preferably as an advantage and / or property of the product, system and / or method, in particular compared to generating metadata, the speed of data processing can be increased and / or more cost-effective processors can be used. Furthermore, by generating the output sequence based on at least part of the metadata, the quality of the output sequence can advantageously be increased. Furthermore, a patent application, in particular with claims as an input sequence, can advantageously be generated and provided.

[0074] Furthermore, a computer program with program code is proposed, comprising instructions which, when the program code is executed by a computer, cause the computer to carry out at least one of the aforementioned methods. The computer program is preferably intended to execute at least the algorithm, in particular machine learning. Preferably, the computer is at least part of the system for inputting and processing an input sequence and for generating an output sequence, in particular an at least partially text-based one, and / or for training a machine learning algorithm for inputting and processing an input sequence and for generating an output sequence, in particular an at least partially text-based one, and / or for continuously training a machine learning algorithm for inputting and processing an input sequence and for generating and evaluating an output sequence, in particular an at least partially text-based one.

[0075] Furthermore, a computer-readable storage medium is proposed on which the computer program with program code is stored. The computer-readable storage medium forms at least partially the storage unit of the computing device.

[0076] The methods according to the invention, the system according to the invention, and the computer program according to the invention are not intended to be limited to the application and embodiment described above. In particular, the methods according to the invention, the system according to the invention, and the computer program according to the invention can have a number of individual elements, components, units, and method steps that differs from the number stated herein in order to fulfill a functionality described herein. Furthermore, for the value ranges stated in this disclosure, values ​​lying within the stated limits are also to be considered disclosed and can be used arbitrarily. In particular, for the examples stated in this disclosure, especially for the text-specific features, any language is to be considered usable.

[0077] Drawings

[0078] Further advantages will become apparent from the following description of the drawings. The drawings illustrate an exemplary embodiment of the invention. The drawings, the description, and the claims contain numerous features in combination. Those skilled in the art will also expediently consider the features individually and combine them into useful further combinations.

[0079] They show:

[0080] Fig. 1 A schematic representation of a system for inputting and processing an input sequence and for generating and providing an output sequence,

[0081] Fig. 2 is a simplified schematic representation of the processing of the input sequence,

[0082] Fig. 3 is a flowchart of a computer-implemented method for inputting and processing the input sequence and for generating and providing the output sequence,

[0083] Fig. 4 is a flowchart for processing the input sequence,

[0084] Fig. 5 is a flowchart of a computer-implemented method for continuously learning an algorithm and

[0085] Fig. 6 is a flowchart of a computer-implemented method for training the algorithm

[0086] Description of the embodiment

[0087] Figure 1 shows a schematic representation of a system 68 for inputting 14 and processing 16 an input sequence 22 and for generating 18 and providing 20 an output sequence 30. The system 68 is intended to execute a computer-implemented method 10 with an algorithm 12.

[0088] The system 68 is at least partially part of a computer. The system 68 has at least one computing device 70. The computing device 70 at least partially forms the computer. The computing device 70 is provided to execute at least the computer-implemented method 10. The computing device 70 is provided to execute all method steps of the computer-implemented method 10. The computer-implemented method 10 is executed by the computing device 70. The algorithm 12 is executed by the computing device 70. The computing device 70 is provided to execute the algorithm 12. The computing device 70 has at least one processor. The algorithm 12 is executed using the processor of the computing device 70. The computing device 70 has at least one memory unit 128. The memory unit 128 is a computer-readable storage medium.A computer program is stored on the computer-readable storage medium. The computer program is stored on the storage unit 128. The computer program comprises program code. The computer program with program code comprises instructions which, when the program code is executed by the computer, cause the computer to execute at least the computer-implemented method 10. The computer program with program code comprises instructions which, when the program code is executed by the computing device 70, cause the computing device 70 to execute at least the computer-implemented method 10. The system 68 comprises the computer program with program code. The algorithm 12 is executed with the computer program. The processor executes the computer program from the storage unit 128 to execute the algorithm 12. The components of the computing device 70 are arranged on a common circuit board in the computer.The components of the computing device 70 are arranged in a housing of the computer. Alternatively, the computing device 70 is configured as a distributed and virtual computing device 70. The computing device 70 is configured as a virtual server and / or a virtual cloud computer.

[0089] The system 68 has at least one user interface 126 for inputting, controlling, and / or outputting data and / or sequences to a user of the system 68. The user interface 126 has at least one module that enables a user to communicate with the system 68 (not shown). The user interface 126 and the at least one module are at least partially a front-end application. The user interface 126 is an external computer with a further computing device for executing the user interface 126. Alternatively, the computing device 68 that executes the algorithm 12 at least partially comprises the user interface 126.

[0090] The system 68 is provided for user input of the input sequence 22. The user interface 128 at least partially comprises an input module for user input of the input sequence 22 into the storage unit 128. The input module is provided for interaction with the user. The computing device 70 has an input means 130. The input means 130 stores data in the storage unit 128 of the computing device 70. The input means 130 stores the input sequence 22, which is entered via the input module, in the storage unit 128 of the computing device 70. The system 68 is provided for the input 14 of the input sequence 22. The system 68 is provided for the input 14 of the input sequence 22 from the storage unit 128 of the computing device 70 into the algorithm 12. The system 68 is provided for the input 14 of a specific input sequence 24.The system 68 is provided for the input 14 of the concrete input sequence 24 from the memory unit 128 of the computing device 70 into the algorithm 12. The system 68 is provided for the input 14 of a graphical input sequence 26. The system 68 is provided for the input 14 of the graphical input sequence 26 from the memory unit 128 of the computing device 70 into the algorithm 12. The system 68 is provided for the input 14 of at least one further input sequence 28. The system 68 is provided for the input 14 of the at least one further input sequence 28 from the memory unit 128 of the computing device 70 into the algorithm 12.

[0091] The input sequence 22 describes a product 32, system and / or method. The product 32 of the input sequence 22 is novel at least for the algorithm 12. The product 32 of the input sequence 22 is patentable. The system of the input sequence 22 is novel at least for the algorithm 12. The system of the input sequence 22 is patentable. The method of the input sequence 22 is novel at least for the algorithm 12. The method of the input sequence 22 is patentable. The concrete input sequence 24 describes the product 32, system and / or method in concrete form. The graphical input sequence 26 describes the product 32, system and / or method in graphical form. The at least one further input sequence 28 describes at least one further product, system and / or method that at least substantially has the same technological field as the product 32 of the input sequence 22.

[0092] The product 32, system and / or method is arbitrary and cannot be restricted to a single product, system and / or method. With each further new input sequence 22 for carrying out the computer-implemented method 10, the product 32, system and / or method also differs. In the further description of the exemplary embodiment, only the product 32 will be described further. However, the description of the product 32 is equally applicable to the system and / or the method. In the method, a sequence of the method is determined according to the structure of the product 32. In the system, a structure and relationship of several products in the system is determined according to the structure of the product 32.

[0093] The input sequence 22 describes at least one sub-element 36 of the product 32. Alternatively and / or additionally, the input sequence 22 describes at least a combination of at least two sub-elements 36 of the product 32. Alternatively and / or additionally, the input sequence 22 describes at least one method step of the product 32. The product 32 can have a different number of sub-elements 36 and / or method steps for different input sequences 22, wherein the product 32 has at least one, but also any number of sub-elements 36 and / or method steps. When describing the method in the input sequence 22, there can be a different number of method steps analogous to the sub-elements 36 of the product 32. In the further description of the exemplary embodiment, only the sub-element 36 will be described further.The process steps are considered analogously to the sub-elements 36 of the product 32 and are treated accordingly.

[0094] The product 32 of the input sequence 22 is based on an inventive step. The product 32 of the input sequence 22 is industrially applicable. The product 32 of the input sequence 22 is missing within sequences of a trained database. The product 32 of the input sequence 22 is missing from a global patent database. The at least one subelement 36 and / or the combination of at least two subelements 36 is a characterizing feature of the input sequence 22. The characterizing feature makes the product 32 novel. The characterizing feature makes the product 32 patentable.

[0095] System 68 is provided to generate at least output sequence 30 using input sequence 22. System 68 is provided for processing 16 input sequence 22. System 68 includes algorithm 12. Algorithm 12 is provided for processing 16 input sequence 22. Algorithm 12 includes a technology recognition model 38 for processing 16 input sequence 22. Technology recognition model 38 is at least partially a machine learning model.

[0096] During processing 16 of the input sequence 22, the algorithm 12 is provided to collect metadata 76 for the input sequence 22. During processing 16 of the input sequence 22, the algorithm 12 is provided to collect metadata 76 for the product 32 of the input sequence 22. During processing 16, the technology recognition model 38 is provided to determine at least a portion of the collected metadata 34 from the collected metadata 76. The output sequence 30 is generated using the portion of the collected metadata 34. The system 68 has at least one database 44, 48, 124, 62 from which the algorithm 12 collects the metadata 76. The system 68 has at least three databases 44, 48, 124, 62 for collecting the metadata 76 (described further in Figures 2 and 4).

[0097] The system 68 has at least one property database 44. The property database 44 has a plurality of property entries 74. The property entries 74 are related to a plurality of products. The property entries 74 are related to a plurality of elements. The property database 44 has a plurality of definitions, benefits, and / or properties. The plurality of definitions, benefits, and / or properties are property entries 74 of the property database 44. The properties, benefits, and / or definitions are related to the plurality of products. The properties, benefits, and / or definitions are related to the plurality of elements. The property entries 74 at least partially form metadata 46 that the algorithm 12 collects. The property database 44 has at least one checklist 72. The checklist 72 is related to a technology area.The property database 44 has a plurality of checklists 72 for different technology areas. The checklists 72 are each a list of rules for a corresponding technology area. After the checklist 72, further property entries 74 of the property database 44 are collected as metadata 46. The further property entries 74 are required for the technology area of ​​the checklist 72. The further property entries 74 of the property database 44, depending on the product 32 of the input sequence 22, at least partially form the metadata 46 that the algorithm 12 collects. The system 68 has at least one sequence database 48. The sequence database 48 has a plurality of sequences 50. The sequences 50 of the sequence database 48 are intellectual property rights. The sequence database 48 is at least partially divided into classes 54.The sequence database 48 is at least partially divided into classes 54 according to company and / or client. The sequences 50 of the sequence database 48 are at least partially assigned to the individual classes 54. The sequence database 48 has at least one trend property 56. The sequence database 48 has at least one trend property 56 for each class 54. The trend properties 56 are intended at least to at least partially determine the portion of the collected metadata 34. The sequence database 48 has at least one relationship matrix 66. The relationship matrix 66 has a plurality of elements as products. The sequence database 48 has at least the plurality of relationship matrices 66 to a plurality of elements as products. The relationship matrices 66 at least partially form metadata 60 that the algorithm 12 collects.In Figure 1, for the sake of clarity, only one class 54 of the sequence database 48 is provided with a reference symbol. However, the other classes 54 are at least functionally identical but data-wise different from the reference-symbolized class 54 of the sequence database 48.

[0098] The sequence database 48 and the property database 44 form at least substantially a part of an overall database 124. The overall database 124 is located on a shared server. The overall database 124 is located on a shared private server. This allows for simplified communication between the algorithm 12 and the property database 44 and the sequence database 48.

[0099] The system 68 at least partially comprises a further sequence database 62. The further sequence database 62 comprises a plurality of further sequences 64. The further sequence database 62 is at least partially public. The algorithm 12 is intended to create at least one of the relationship matrices 66 with the plurality of sequences 64. The sequence database 48 comprises the relationship matrices 66 of the further sequence database 62. The relationship matrices 66 of the further sequence database 48 at least partially form the metadata 60 that the algorithm 12 collects.

[0100] The system 68 is provided for the generation 18 of the output sequence 30. The system 68 is provided for generating the output sequence 30 with the portion of the collected metadata 34. The system 68 is provided for the generation 18 of the output sequence 30 in text-based form. During the generation 18 of the output sequence 30, the system 68 is provided for converting the portion of the collected metadata 34 into text-based, concatenated form. During the generation 18 of the output sequence 30, the system 68 is provided for generating a continuous text. The algorithm 12 has a text generation model 140 for the generation 18 of the output sequence 30. The text generation model 140 is at least partially a machine learning model. The text generation model 140 and the technology recognition model 38 are designed separately from one another. The system 68 is intended to provide 20 the output sequence 30.The system 68 is provided for the provision 20 of the text-based output sequence 30. The provided output sequence 30 includes at least a portion of the collected metadata 34. The provided output sequence 30 describes a structure of the product 32. The provided output sequence 30 describes a property of the product 32. The provided output sequence 30 describes at least one advantage of the product 32. The provided output sequence 30 describes a function of the product 32. The provided output sequence 30 is a patent application for the product 32.

[0101] The output sequence 30 is provided by the algorithm 12 in the memory unit 128 of the computing device 70. The system 68 is provided for outputting the output sequence 30. The user interface 126 at least partially comprises an output module for outputting the output sequence 30 from the memory unit 128. The output module is provided for interaction with the user. The computing device 70 has an output means 132. The output means 132 outputs data from the memory unit 128 of the computing device 70 to the output module. The output means 132 outputs the output sequence 30, which is stored on the memory unit 128, to the output module.

[0102] In at least one further embodiment of the system according to the invention, the user interface 126 has at least one processing module. The processing module is intended to at least partially provide the output sequence 30 to a user. The output sequence 30 is a metadata construct, wherein the product 32 and the sub-elements 36 of the product 32 of the input sequence 22 are each linked to a plurality of metadata 34 of the part of the collected metadata 34. The processing module is intended for parallel generation with the computing device 70 for creating the output sequence 30 as continuous text with the user. The processing module is intended to provide the user with the linked metadata 34 of the output sequence 30 as suggestions. The user enters the preferred metadata 34 into the processing module and thus generates the final output sequence 30 as continuous text.

[0103] Figure 2 shows a simplified schematic representation of the processing 16 of the input sequence 22. The technology recognition model 38 is provided for processing 16 of the input sequence 22. The technology recognition model 38 receives the input sequence 22 from the algorithm 12 for processing 16 of the input sequence 22. During the processing 16 of the input sequence 22, the technology recognition model 38 is provided to prepare the input sequence 22. During the processing 16 of the input sequence 22, the technology recognition model 38 is provided to detect the product 32 of the input sequence 22. In the further method step, the product 32 is separated as an object from a continuous text of the input sequence 22. During the processing 16 of the input sequence 22, the technology recognition model 38 is provided to detect the sub-elements 36 of the product 32. The technology recognition model 38 detects a structure of the product 32.

[0104] During processing 16 of the input sequence 22, the technology recognition model 38 is provided to collect metadata 76 for the sub-elements 36. The technology recognition model 38 is provided to transfer 122 the sub-elements 36 for a database query. The technology recognition model 38 is provided to transfer 122 the sub-elements 36 to the algorithm 12 for the database query. The algorithm 12 collects the property entries 74 for the sub-elements 36 as metadata 46 from the property database 44. The algorithm 12 is provided to transfer 122 the collected property entries 74 as metadata 46 from the property database 44 to the technology recognition model 38. The technology recognition model 38 links the property entries 74 to the sub-elements 36.

[0105] The technology recognition model 38 is provided for creating a final relationship matrix 116. The technology recognition model 38 is provided for transferring 122 the sub-elements 36 with at least some of the property entries 74 to the algorithm 12. With the sub-elements 36 and the part of the property entries 74, further metadata 60 is collected at least from the sequence database 48. The algorithm 12 is provided for transferring 122 the further metadata 60 to the technology recognition model 38. The technology recognition model 38 creates at least the final relationship matrix 116 with the further metadata 60. The final relationship matrix 116 has a plurality of elements. The plurality of elements are at least partially implementations 58 of the product 32 and / or the sub-elements 36 of the product 32.The final relationship matrix 116 forms at least in part the entire collected metadata 76 of the property database 44 and the sequence database 48 and the further sequence database 62 during the processing 16 of the input sequence 22.

[0106] The technology recognition model 38 is intended to determine the portion of the collected metadata 34 from the collected metadata 76. The technology recognition model 38 determines the portion of the collected metadata 34 from the final relationship matrix 116. The technology recognition model 38 is intended to weight the collected metadata 76 and, from this, to determine the portion of the collected metadata 34. The portion of the collected metadata 34 is provided for the generation 18 of the output sequence 30.

[0107] Figure 3 shows a flowchart of the computer-implemented method 10 for providing 20 the output sequence 30. The computer-implemented method 10 comprises the algorithm 12. The algorithm 12 is a machine learning algorithm 12. The computer-implemented method 10 is at least partially carried out using the algorithm 12. The computer-implemented method 10 is executed by the computing device 70. In at least one method step, the input 14 of the input sequence 22 is carried out. The input sequence 22 is input into the algorithm 12 from the memory unit of the computing device 70. The input sequence 22 is entered in text form. The product 32 of the input sequence 22 is described in abstract form. The input sequence 22 is in abstract form. The input sequence 22 describes only the novel core of the product 32. The input sequence 22 describes only the patentable core of the product 32.The product 32 is described with an abstract description. The abstract description defines a technology area of ​​the product 32. The input sequence 22 defines the scope of protection of the patentable product 32. The input sequence 22 delimits the scope of protection of the patentable product 32. The input sequence 22 describes an inventive concept. The abstract description of the product 32 of the input sequence 22 is an abstracted description of a technically preferred implementation 58 of the product 32. The at least one sub-element 36 of the product 32 is described in abstract form. The at least one sub-element 36 is described using an abstract term. The abstract term defines a technology area of ​​the sub-element 36. The product 32 of the input sequence 22 has a plurality of sub-elements 36. The structure of the product 32 is at least essentially clearly described on the basis of the sub-elements 36.The number of sub-elements 36 depends on the number of sub-elements 36 necessary to describe the novelty of the product 32.

[0108] The input sequence 22 is divided into subsequences. The subsequences each describe at least one function, one feature of the product 32. The subsequences each describe at least one subelement 36 and / or one method step of the product 32. The input sequence 22 is entered in the form of claims. The claims describe the product 32. Each subsequence of the input sequence 22 is a claim of the claims of the product 32. The input sequence 22 has at least one independent claim. The input sequence 22 has at least one dependent claim. The at least one independent claim of the input sequence 22 has a characterizing part. The characterizing part of the independent claim describes a patentable core of the product 32. The at least one independent claim has at least one generic term. The generic term contains a designation of the technology field of the product 32.The preamble comprises technical features of the product 32. At least one dependent claim of the initial sequence 22 comprises a characterizing part. The characterizing part of the dependent claim describes a further feature of the product.

[0109] At least the concrete input sequence 24 is entered. The concrete input sequence 24 is text-based. The concrete input sequence 24 is entered together with the input sequence 22 in abstract form. The concrete input sequence 24 describes the product 32 in concrete form. The concrete input sequence 24 describes a technical implementation of the product 32. The concrete input sequence 24 describes a technical implementation of the input sequence 22. The concrete input sequence 24 is a particularly preferred concrete implementation of the input sequence 22. The concrete input sequence 24 is a technically feasible implementation of the product 32. The concrete input sequence 24 is an invention disclosure of the novel and patentable product 32. The concrete input sequence 24 has a technology area according to the technology area of ​​the input sequence 22 in abstract form. The concrete input sequence 24 has advantages of the product 32.The concrete input sequence 24 has properties of the product 32. A relationship between the concrete input sequence 24 and the input sequence 22 is entered in abstract form.

[0110] At least one graphic input sequence 26 is entered. The graphic input sequence 26 graphically represents the product 32 of the input sequence 22. The input sequence graphically represents at least one sub-element 36 of the product 32 of the input sequence 22. The graphic input sequence 26 graphically represents the input sequence 22 in abstract form. The graphic input sequence 26 graphically represents at least one implementation of the input sequence 22. The graphic input sequence 26 graphically represents at least the implementation of the concrete input sequence 24. The graphic input sequence 26 is in the form of a technical drawing. The graphic input sequence 26 represents the product 32 in the form of a technical drawing 42. The graphic input sequence 26 represents at least one sub-element 36 of the product 32 in the form of a technical drawing 42. A structure of the product 32 is represented in the graphic input sequence 26.The graphical input sequence 26 shows a structure of the sub-elements 36. The input sequence 22 is linked to the graphical input sequence 26. The graphical input sequence 26 has at least one reference symbol. The input sequence 22 has at least one reference symbol. The input sequence 22 has at least partially the reference symbols of the graphical input sequence 26. The reference symbol links the input sequence 22 to the graphical input sequence 26. The reference symbol uniquely identifies the product 32. The reference symbol uniquely identifies the at least one sub-element 36. The at least one reference symbol is alphanumeric. The product 32 of the graphical input sequence 26 can be uniquely identified using the reference symbol of the product 32 of the input sequence 22.The at least one subelement 36 of the graphic input sequence 26 can be uniquely identified using the reference symbol of the at least one subelement 36 of the input sequence 22. Based on the reference symbols of the input sequence 22, the structure of the product 32 is identified using the graphic input sequence 26. Based on the reference symbols of the input sequence 22, an arrangement of the subelements 36 is identified using the graphic input sequence 26.

[0111] At least one further input sequence 28 is entered. The at least one further input sequence 28 describes a further product. The further product of the at least one further input sequence has further sub-elements. The further product of the at least one further input sequence is obvious. The further product is present in the disclosed prior art in a global patent database. The further product of the at least one further input sequence 28 is different from the product 32 of the input sequence 22. The at least one further input sequence 28 is a detailed description of the further product. The at least one further input sequence 28 has advantages of the further product. The at least one further input sequence 28 has properties of the further product.The further product of the further input sequence 28 is at least partially in the same technology field as the product 32 of the input sequence 22. The further product of the at least one further input sequence 28 has at least partially the same sub-elements as the product 32 of the input sequence 22 in abstract form. The at least one further input sequence 28 is a patent specification and / or application specification. The at least one further input sequence 28 is the closest prior art of the novel and patentable product 32 of the input sequence 22. A relationship between the at least one further input sequence 28 and the input sequence 22 is entered in abstract form. If a plurality of further input sequences 28 are entered, a relationship to the input sequence 22 is entered in each case.

[0112] In at least one further embodiment of the method 10 according to the invention, at least the input sequence 22 is entered in an alternative file format. In the further embodiment, the input sequence 22 in the alternative file format replaces and / or supplements at least the input sequence 22, the concrete input sequence 24, and / or the graphic input sequence. The input sequence 22 in the alternative file format is entered as a graphic. The graphic represents the novel and patentable product 32. The graphic has at least one characterizing feature. The characterizing feature is highlighted in the graphic. The characterizing feature is the novel feature of the product 32. Alternatively or additionally, the input sequence 22 is entered in the alternative file format as a mind map and / or a voice sequence. The mind map and / or the voice sequence describe the novel and patentable product 32.Alternatively or additionally, the input sequence 22 is input in the alternative file form as a computer program with program code. The computer program with program code is intended for carrying out a novel and patentable method. The computer program with program code is executed directly by the algorithm 12 upon input of the input sequence. During processing 16 of the input sequence 22, the metadata 76 relating to the execution result of the computer program with program code is collected. The output sequence 30 at least partially describes the execution result of the input sequence 22 as a computer program with program code. All alternative embodiments or alternative file forms of the input sequence 22 are subsequently processed at least substantially the same as the input sequence 22 in abstract form.

[0113] In at least one method step, the processing 16 of the input sequence 22 is performed. The input sequence 22 is processed using the algorithm 12. The algorithm 12 has at least one technology recognition model 38. The input sequence 22 is processed using the technology recognition model 38. The product 32 is detected using the technology recognition model 38. The at least one sub-element 36 is detected using the technology recognition model 38.

[0114] During processing 16 of the input sequence 22, the product 32 is recorded using the technology recognition model 38. During processing 16 of the input sequence 22, a function of the product 32 is recorded using the technology recognition model 38. During processing 16 of the input sequence 22, metadata 76 is collected for the product 32. The metadata 76 is collected by the algorithm 12. The collected metadata 76 is processed using the technology recognition model 38. During processing 16 of the input sequence 22, the metadata 76 is collected for at least one sub-element 36 of the product 32 of the input sequence 22. Using the technology recognition model 38, at least one advantage of the product 32 is collected as metadata 76. Using the technology recognition model 38, at least one property of the product 32 is collected as metadata 76.The technology recognition model 38 collects at least one advantage of the at least one sub-element 36 of the product 32 as metadata 76. The technology recognition model 38 collects at least one property of the at least one sub-element 36 of the product 32 as metadata 76. The technology recognition model 38 collects at least the implementation 58 of the product 32 as metadata 76. The implementation 58 is a concretization or an alternative embodiment of the product 32. The technology recognition model 38 collects at least one implementation 58 of the at least one sub-element 36 as metadata 76. The implementation 58 is a concretization or an alternative embodiment of the sub-element 36. During processing 16 of the input sequence 22, the metadata 76 is assigned to the input sequence 22. The metadata 76 for the product 32 is weighted using the technology recognition model 38.The metadata 76 having the strongest weighting is output as part of the collected metadata 34 after processing 16 of the input sequence 22 for generation 18 of the output sequence 30. During processing 16 of the input sequence 22, metadata 76 is determined regarding a structure of the input sequence 22. The structure of the input sequence 22 is captured based on structural features of the input sequence 22.

[0115] During processing 16 of the input sequence 22, the concrete input sequence 24 is also processed. The advantages of the concrete input sequence 24 are collected as metadata 76 for the input sequence 22 using the technology recognition model 38. The properties of the concrete input sequence 24 are collected for the input sequence 22 using the technology recognition model 38. The collected metadata 76 for the input sequence 22 is at least partially weighted in abstract form using the concrete input sequence 24 using the technology recognition model 38. During processing 16 of the input sequence 22, the graphical input sequence 26 is also processed. The structure of the product 32 is recorded using the graphical input sequence 26 during processing 16 of the input sequence 22. During processing 16 of the input sequence 22, at least one further input sequence 28 is also processed.The technology recognition model 38 collects the advantages of the at least one further input sequence 28 as metadata 76 for the input sequence 22. The technology recognition model 38 collects the properties of the at least one further input sequence 28 as metadata 76 for the input sequence 22. During the processing 16 of the at least one further input sequence 28, a publication number is collected as metadata of the at least one further input sequence 28. During the processing 16 of the at least one further input sequence 28, an abstract is collected as metadata of the at least one further input sequence 28. The technology recognition model 38 is used to at least partially weight the collected metadata 76 for the input sequence 22 in abstract form based on the at least one further input sequence 28.

[0116] The computer-implemented method 10 includes at least one property database 44. The property database 44 is private. The property database 44 is accessible only by an authorized user or members of an organization. The property database 44 is hosted on a private server. The server is protected by authentication and access controls. The property database 44 stores the metadata 46. The property database 44 has the metadata 46 that at least partially forms the collected metadata 76 of the technology discovery model 38. During processing 16 of the input sequence 22, the metadata 46 is collected from the property database 44 using the technology discovery model 38. The metadata 46 are properties of an item. The properties are technical. The metadata 46 are advantages of the item. The metadata 46 is information about the item.The metadata 46 is stored with the property entries 74 in the property database. The property database 44 has a plurality of elements. The property database 44 has at least one element as a product. The property database 56 has at least one element as a sub-element of a product. Each element forms one of the property entries 74 of the property database 44. For each element as a property entry 74, the information is stored at least as a property and / or benefit.

[0117] The computer-implemented method 10 comprises at least one sequence database 48. The sequence database 48 is private. The sequence database 48 is accessible only by an authorized user or members of an organization. The sequence database 48 is hosted on a private server. The server is protected by authentication and access controls. The sequence database 48 has a plurality of sequences 50. The sequences 50 are text-based. The sequences 50 each describe a further product. The sequences 50 are intellectual property rights of the further products. The sequences 50 are patent applications, patents, and / or utility models of the further products. The sequences 50 of the sequence database 48 are preselected by a user. The sequences 50 of the sequence database 48 are the prior art of at least one company. The further products of the sequences 50 have subelements.The subelements of the further products have advantages. The subelements of the further products have properties. The sequences 50 have metadata 52. The sequence database 48 has the metadata 52, which at least partially forms the collected metadata 76 of the technology recognition model 38. During processing 16 of the input sequence 22, the metadata 52 is collected from the sequence database 46. The metadata 52 of the sequences 50 is related to the further product. The advantages of the sequences 50 form the metadata 52. The properties of the sequences 50 form the metadata 52.

[0118] The computer-implemented method 10 comprises at least one further sequence database 62. The further sequence database 62 is public. The further sequence database 62 is freely accessible via the Internet. The further sequence database 62 is a patent database. The further sequence database 62 has a plurality of sequences 64. The further sequence database 62 contains a large part of the current state of the art. The further sequence database 62 contains public patents, patent applications, utility models, and other intellectual property rights. The plurality of sequences 64 each describe a further product, system, and / or method. The plurality of sequences 64 have metadata. The further sequence database 48 has the metadata that at least partially forms the collected metadata 76 of the technology recognition model 38. The metadata are related to the further product 32. The further products of the sequences 64 have subelements.The sub-elements of the further products have advantages. The sub-elements of the further products have properties. The advantages of the sequences 62 are metadata. The properties of the sequences 64 form metadata, at least in part. During processing 16 of the input sequence 22, a subset of the plurality of sequences 64 of the further sequence database 62 is transferred to the sequence database 48 using the algorithm 12. In the sequence database 48, the subset of the plurality of sequences 64 is processed using the algorithm 12. The dashed arrow in Figure 3 represents that no direct database query of the further sequence database 62 is performed using the input sequence 22. The information of the input sequence 22 is at least partially modified for the database query of the further sequence database 62.

[0119] The processing 16 of the input sequence 22, the concrete input sequence 24, the graphical input sequence 26, and the further input sequence 28 is described in further detail in Figure 2. Furthermore, the collection of the metadata 76 from the property database 44, the sequence database 48, and the further sequence database 62 is described in further detail in Figure 4.

[0120] The generation 18 of the output sequence 30 is carried out in at least one method step. The output sequence 30 is generated starting from the input sequence 22. The output sequence 30 is generated using the algorithm 12. The output sequence 30 is generated using the text generation model 140 of the algorithm 12. The output sequence 30 is generated at least partially based on text. The output sequence 30 is generated based on at least the part of the collected metadata 34. The product 32 is detailed using the part of the collected metadata 34 during the generation 18 of the output sequence 30. The part of the collected metadata 34 is linked to the product in a text-based manner during the generation 18 of the output sequence 30. The part of the collected metadata 34 is linked to the sub-elements 36 of the product 32 in a text-based manner during the generation 18 of the output sequence 30.During generation 18 of the output sequence 30, the portion of the collected metadata 34, which is determined at least by the metadata 46 of the property database 22, is linked to the product 32 and generated in a text-based manner. During generation 18 of the output sequence 30, the portion of the collected metadata 34, which is determined at least by the metadata 60 of the sequence database 48, is linked to the product 32 and generated in a text-based manner. During generation 18 of the output sequence 30, the portion of the collected metadata 34, which is determined at least by the metadata 40 of the at least one further input sequence 28, is linked to the product 32 and generated in a text-based manner. Each subsequence of the input sequence 22 is linked to the portion of the collected metadata 34 during generation 18 of the output sequence 30. During the generation 18 of the output sequence 30, at least one output subsequence is generated for each subsequence of the input sequence 22.Each output subsequence comprises at least a portion of the portion of collected metadata 34. The at least one subelement 36 is linked to metadata 34 of the portion of collected metadata during generation 18 of the output sequence 30. During generation 18 of the output sequence 30, a description of the input sequence 22 is generated. An application text for a patent application is generated as the output sequence 30. The application text claims the product 32.

[0121] The application text of product 32 includes the initial sequence 22 as claims of the application text. The initial sequence 22 is incorporated into the claims of the application text of the product using the text generation model 140.

[0122] The text generation model 140 generates a description of the advantages of the product 32 in the output sequence 30. The text generation model 140 generates at least the description of the advantages of the input sequence 22. The text generation model 140 generates the properties of the product 32 in the description of the advantages. When generating the description of the advantages of the product 32, the partial sequences of the input sequence 22 are at least partially adopted. When generating the description of the advantages of the product 32, the claims of the input sequence 22 are adopted. The text generation model 140 adopts the at least one independent claim of the input sequence 22 into the description of the advantages. The text generation model 140 adopts the characterizing part of the at least one dependent claim of the input sequence 22 into the description of the advantages.With the text generation model 140, the metadata 34 collected for the at least one independent claim of the input sequence 22 is generated in a text-based manner from the portion of the collected metadata 34 into the description of the advantages of the output sequence 30. With the text generation model 140, the metadata 34 collected for the at least one dependent claim of the input sequence 22 is generated in a text-based manner from the portion of the collected metadata 34 into the description of the advantages of the output sequence 30. With the text generation model 140, the metadata 34 linked to the respective portions 36 are generated in a text-based manner from the portion of the collected metadata 34 together with the corresponding portion, in accordance with the described portions 36 in the portions of the input sequence.

[0123] The output sequence 30 has at least one graphical representation of the product 32. The graphical representation of the output sequence 30 is adopted from the graphical input sequence 26 using the text generation model 140. The graphical representation of the output sequence 30 has reference symbols. The reference symbols are adopted from the graphical input sequence 22 using the text generation model 140. The graphical representation is in the form of a patent drawing. In at least one embodiment of the method 10 according to the invention, the graphical representation is generated by the text generation model 140. The graphical representation is generated based on the description of the product 32 in the input sequence 22. The graphical representation is a schematic structure of the product 32. Alternatively or additionally, a flow chart or functional diagram of the product 32 is generated using the text generation model 140.The flow chart or functional diagram is linked to the text-based part of the output sequence 30 via reference symbols.

[0124] The text generation model 140 generates at least one figure description of the product 32 in the output sequence 30. The figure description is generated based on the structure of the product 32. The figure description is generated based on the structure of the sub-elements 36 of the product 32. The figure description is constructed based on the recorded structure of the input sequence 22. The text generation model 140 describes the sub-elements 36 in the order in which the sub-elements 36 are related, in accordance with the recorded structure of the input sequence 22. When generating the figure description of the output sequence 30, the sub-sequences of the input sequence 22 are at least partially adopted. When generating the figure description, the claims are at least partially adopted as sub-sequences of the input sequence 22.For each feature of the input sequence 22, a sentence is generated in the character description using the text generation model 140. The text generation model 140 captures optional features of the input sequence 22 during the generation of the character description. The capture of the optional features of the input sequence 22 is carried out using the text generation model 14 with regular expressions for optional terms. The optional features of the input sequence 22 are each generated text-based as an independent sentence in the figure description. For subsequences of the input sequence 22 that have at least two features, at least one sentence is generated in the figure description for each feature using the text generation model 140. The portion of the collected metadata 34 as a property of the at least one subelement 36 is incorporated into the figure description using the text generation model 140.When generating the figure description, at least one subelement 36 is linked to reference symbols. The reference symbols are extracted from the input sequence 22 using the text generation model 140. Alternatively, a reference symbol list is entered. The reference symbol list includes an assignment of the subelements 36 of the product 32 to reference symbols. The text generation model 140 extracts the reference symbols for the at least one subelement 36 from the reference symbol list. The text generation model includes regular expressions with which the subelements 36 in the output sequence 30 are captured for linking to the respective reference symbols.

[0125] A prior art part of the output sequence 30 is generated using the text generation model 140. The prior art part describes the prior art for the product 32 of the input sequence 22. In the prior art part of the output sequence 30, a description of the prior art of the product 32 is generated in text-based form. For the prior art part of the output sequence 30, the collected metadata 40 from the at least one further input sequence 28 is transferred to the output sequence 30. In the prior art part, at least the publication number of the at least one further input sequence 28 is transferred to the output sequence 30. In the prior art part of the output sequence 30, at least one summary of the at least one further input sequence 28 is generated. A summary of the further product of the at least one further input sequence 28 is generated using the text generation model 140.Alternatively, the abstract of at least one further input sequence 28 is incorporated into the prior art portion of the output sequence 30. In the prior art portion of the output sequence 30, a task of the product 32 of the input sequence 22 is generated. The prior art portion describes the task of the product 32 of the input sequence 22 given the prior art. The task is determined by a most heavily weighted advantage of the collected metadata 76 during processing 16 of the input sequence 22. The weighting of the metadata 76 is described in Figure 4.

[0126] Using the text generation model 140, the terms of the input sequence 22 are transferred into the output sequence 30. Using the text generation model 140, the sentence structures of the input sequence 22 are at least partially transferred. The output sequence 30 is generated using the text generation model 140 according to text-based rules. The text-based rules vary depending on the part of the output sequence 30. The text-based rules define the sentence structure of the output sequence 30 to be generated. The text-based rules have variables. The variables are inserted using the text generation model 140. Using the text generation model 140, the variables are at least partially populated with the part of the collected metadata 34. The variables are at least partially populated with the input sequence 22.A text-based rule for generating the description of the advantages for an input subsequence as an output subsequence is, for example: "[Furthermore / In addition / Furthermore / In addition] it is proposed that [characterizing part of the claim]. [Properties of the subelements of the product as collected metadata], [Implementation of the subelements of the product as collected metadata], [Advantages of the subelements of the product as collected metadata]." With the text generation model 140, at least substantially the sentence structures of the part of the collected metadata 34 are adopted into the output sequence 30. In at least one embodiment of the method 10 according to the invention, the terms of the collected metadata 34 are adopted by the terms of the input sequence 22. With the technology recognition model.

[0127] 38, the terms of the part of the collected metadata 34 are replaced by the respective term of the sub-element 36 of the input sequence 22 to which the respective metadata 34 is linked.

[0128] The text generation model 140 generates general formulations. The general formulations are independent of the technology area of ​​the product 32. The general formulations of the description of the advantages are, for example: "The invention is based on a [preamble of independent claim 1]" and "It is proposed that [characterizing part of independent claim 1]" and "[Generic terms of the independent claims] are not intended to be limited to the application and embodiment described above. In particular, [genetic terms of the independent claims] can have a number of individual elements, components and units as well as method steps that differs from a number stated herein to fulfill a function described herein.In addition, in the value ranges specified in this disclosure, values ​​lying within the stated limits are also to be considered disclosed and can be used as desired." The general wording of the figure description is, for example: "Figure [number of the figure to be described] shows [description]" and "In the figures [number of figures for further embodiments of the product] a [number] further embodiment of the invention is shown. The following descriptions and the drawings are essentially limited to the differences between the embodiments, whereby with regard to components with the same designation, in particular with regard to components with the same reference numerals, reference can in principle also be made to the drawings and / or the description of the other embodiments, in particular Figures 1 to [last figure of the first embodiment].To distinguish the embodiments, the letter a is placed after the reference numerals of the embodiment in Figures 1 to [last figure of the first embodiment]. In the embodiments of the figures [number of the figures for further embodiments of the product], the letter a is replaced by the letters b to [highest BZ suffix].” The general formulations of the prior art section are, for example: “The invention relates to a [generic term of the independent claims] according to the preamble of the claim [numbering of the independent claims].” and “It has already been proposed that [brief description of the at least one further input sequence].” and “The object of the invention is in particular to provide a generic device with [characterizing advantage of the product].The object is achieved according to the invention by the features of claim 1, while advantageous embodiments and further developments of the invention can be found in the subclaims." The general formulations can be adapted and supplemented as desired. The general formulations can vary depending on the number of embodiments. During the generation 18 of the output sequence 30, the text generation model 140 takes the general formulation that belongs to the corresponding generic term. The text generation model 140 has a plurality of regular expressions for generating 18 the output sequence 30. The text-based rules are executed using the regular expressions. During the generation 18 of the output sequence 30, the parts of the output sequence 30 are put together using the regular expressions.With the regular expressions, the input sequence 22 is at least partially incorporated into the output sequence 30 during generation 18 of the output sequence 30. With the regular expressions, the variables are generated according to the given input sequence 22. With the regular expressions, the general formulations are generated according to the given input sequence 22.

[0129] In at least one embodiment of the invention, the algorithm 12 at least partially comprises a natural language model as a text generation model 140. The natural language model is a machine learning model for generating text sequences using natural language. The natural language model is used to generate a coherent text during the generation 18 of the output sequence 30. The input sequence 22 is an input variable of the natural language model. The concrete input sequence 24 is an input variable of the natural language model. The graphical input sequence 26 is an input variable of the natural language model. The at least one further input sequence 28 is an input variable of the natural language model. The part of the collected metadata 34 is an input variable of the natural language model. The natural language model uses the input variables to generate a coherent text as an output variable.The natural language model generates the coherent text based on the text-based rules of the output sequence 30. An example of an applicable natural language model is a privately hosted GPT model or LLaMA model, or similar. If the security criteria for generating a patent application for a patentable product 32 are met, an API interface to a natural language model is also conceivable.

[0130] The provision 20 of the output sequence 30 is carried out in at least one method step. For the provision 20, the output sequence 30 is output by the text generation model 140 to the algorithm 12. The output sequence 30 is provided with the algorithm 12 at least in the memory unit 128 of the computing device 70. The output sequence 30 is provided to the output means 132 of the computing device 70. The output sequence 30 is output to the user interface 126 via the output means 132 of the computing device 70. The output sequence 30 is stored in the sequence database 48 upon provision 20 of the output sequence 30. At least all output sequences 30 that are processed by the algorithm 12 are stored in the sequence database 48.When a new input sequence 22 is input 14, metadata 76 is collected from the output sequence 30 by the sequence database 48 during processing 16 of the new input sequence 22.

[0131] In at least one embodiment of the method 10 according to the invention, the output sequence

[0132] 30 with the concrete input sequence 24 using algorithm 12. The

[0133] Output sequence 30 is checked for completeness against the specific input sequence 24. Algorithm 12 checks whether all features of the specific input sequence 24 are present in the output sequence 30. Algorithm 12 compares the output sequence 30 with the specific input sequence 24 using regular expressions. In an alternative embodiment, algorithm 12 compares the output sequence 30 with the specific input sequence 24 using the technology recognition model 38.

[0134] In at least one further embodiment of the method 10 according to the invention, the portion of the collected metadata 34 in the generated output sequence 30 is marked during provision. The portion of the collected metadata 34 is marked for review by a user. The output sequence 30 with the markings is passed to the user interface via the output means of the computing device.

[0135] In at least one further embodiment of the method 10 according to the invention, a data construct is generated during the generation 20 of the output sequence 30. The data construct is a collection of text-based subsequences as the output sequence 30. The data construct has an assignment between the product 32 of the input sequence 22 and the part of the collected metadata 34. The data construct has an assignment of the subelements 36 of the product 32 and the part of the collected metadata 34. The data construct is provided to the user. The data construct is passed to the user interface via the output means of the computing device 70. The data construct is provided to the user for parallel creation of a coherent text. The data construct is provided to a user for creation of the registration text of the product 32.If a partial element 36 of the product 32 is entered via the user interface in text form, the part of the collected metadata 34 relating to the partial element 36 is output by the data construct as a supplement to the user interface.

[0136] Figure 4 shows a flowchart for processing an input sequence 22. Using algorithm 12, the input sequence 22 is input into the technology recognition model 38. The input sequence 22 is input into the technology recognition model 38 for processing 16. Using algorithm 12, the specific input sequence 24 is input into the technology recognition model 38. The specific input sequence 24 is input to the technology recognition model 38 for processing 16. Using algorithm 12, the graphical input sequence 26 is input into the technology recognition model 38. The graphical input sequence 26 is input to the technology recognition model 38 for processing 16. Using algorithm 12, the at least one further input sequence 28 is input from algorithm 12 into the technology recognition model 38.The at least one further input sequence 28 is input to the technology recognition model 38 for processing 16.

[0137] Metadata of the input sequence 22 is input into the algorithm 12. The metadata of the input sequence 22 determines a sequence type of the input sequence 22. The metadata of the input sequence 22 determines a file type of the input sequence 22. The sequence type of the input sequence 22 is text-based and in abstract form. The sequence type of the input sequence 22 is the claims of the input sequence 22. The metadata of the input sequence 22 is input for processing 16 of the input sequence 22. The metadata of the input sequence 22 is passed to the technology recognition model 38 for processing 16 of the input sequence 22. The metadata of the input sequence 22 is input into the algorithm 12 from the storage unit of the computing device 70. The metadata of the input sequence 22 is linked to the input sequence 22. The metadata of the input sequence 22 is at least partially a processor of the input sequence 22.The metadata of the input sequence 22 is at least partially a company of the product 32 of the input sequence 22. The metadata of the input sequence 22 is an inventor of the input sequence 22. The metadata of the input sequence 22 is a file title of the input sequence 22. In at least one embodiment of the method 10 according to the invention, the company of the product 32 is detected using the algorithm 12 based on the file title of the input sequence 22. The company is detected from the file title using regular expressions. In at least one further embodiment of the method 10 according to the invention, the metadata of the input sequence 22 is additionally entered by the user via the user interface 126.

[0138] The metadata of the concrete input sequence 24 is input into the algorithm 12. The metadata of the graphical input sequence 26 is input into the algorithm 12. The metadata of the at least one further input sequence 28 is input into the algorithm 12. The metadata of the concrete, graphical, and at least one further input sequence 24, 26, 26 correspond at least partially to the metadata of the input sequence 22. The metadata of the concrete, graphical, and at least one further input sequence 24, 26, 26 is transferred to the technology recognition model 38 for processing 16 of the input sequence 22.

[0139] Processing step 80 for text preparation

[0140] In at least one processing step 80, the input sequence 22 is prepared for collecting the metadata 76. In the processing step 80, the input sequence 22 is text-processed using the technology recognition model 38. The technology recognition model 38 performs the text-processing to detect the product 32. The technology recognition model 38 performs the text-processing to detect the sub-elements 36 of the product 32. The text-processing of the input sequence 22 converts the input sequence 22 into a machine-readable format.

[0141] In at least one text processing step, the input sequence 22 is tokenized. During tokenization, the input sequence 22 is segmented. The input sequence 22 is segmented into phrases. The input sequence 22 is segmented into trigrams. Segmentation into trigrams captures context information of the input sequence 22. Examples of trigrams of the input sequence 22 are "partially fireproof material" or "at least partially round."

[0142] Alternatively, the input sequence 22 is segmented into bigrams or words during tokenization.

[0143] A token is assigned to a segment of the segmented input sequence 22.

[0144] In at least one further text processing step, the input sequence 22 is normalized. The tokens of the input sequence 22 are brought into a uniform form. The words of the input sequence 22 are converted to lowercase letters. The words of the input sequence 22 are lemmatized. This allows consistency of the data of the input sequence 22 to be achieved.

[0145] In at least one further text processing step, the input sequence 22 is vectorized. The tokens of the input sequence 22 are converted into numeric vectors. The tokens are converted into vectors using a word embedding method. The tokens are converted into vectors using Word2Vec or GloVe. Alternatively, the tokens are converted into vectors using Bag of Words or TF-IDF or a similar method.

[0146] In at least one alternative embodiment of the method 10 according to the invention, subword-based tokenization is applied to the input sequence 22 in the text processing step in which the input sequence 22 is tokenized. With the subword-based tokenization, the input sequence 22 is segmented into words and subwords. The subword-based tokenization of the input sequence 22 is carried out, for example, using Byte Pair Encoding (BPE) or SentencePiece. With the subword-based tokenization, the context of the input sequence 22 can be understood. With the subword-based tokenization, the meaning of connecting words in the input sequence 22 can be captured. Connecting words are, for example, "with," "and," "or," "through," and "after." In the text processing step for vectorization, the tokens are vectorized by embedding them in a continuous vector space.The continuous vector space is defined by pre-trained embedding layers that directly convert the tokens into vectors.

[0147] In processing step 80, the concrete input sequence 24 is prepared. The concrete input sequence 24 is prepared according to the input sequence 22.

[0148] In processing step 80, the graphic input sequence 26 is prepared. The reference symbols of the graphic input sequence 26 are captured using OCR. The sub-elements 36 of the graphic input sequence 26 are captured using object recognition algorithms. The sub-elements 36 of the graphic input sequence 26 are segmented using image segmentation methods. Using the technology recognition model 38, the reference symbols from the OCR capture and the sub-elements 36 from the object recognition algorithms and the image segmentation are linked together in a structured representation. The structured representation is vectorized. Processing step 83 for capturing the technology areas of the product 32

[0149] In at least one further processing step 82, the product 32 of the input sequence 22 is determined. In the processing step 82, the product 32 is detected using the technology recognition model 38. In the processing step 82, the at least one sub-element 36 of the product 32 is detected using the technology recognition model 38. The technology recognition model 38 detects the product 32 in text-based form from the input sequence 22. The technology recognition model 38 detects the at least one sub-element 36 of the product 32 from the input sequence 22. The technology recognition model 38 detects the sub-elements 36 of the product 32 using subject recognition of the input sequence 22. The technology recognition model 38 performs a POS tagging process. The POS tagging process detects the subjects of the input sequence 22. The POS tagging process is applied to the lemmatized input sequence 22.The technology recognition model 38 filters all sub-elements 36 of the product 32 from the input sequence 22 using the POS tagging process. Since the product 32 is described in claim form in the claims of the input sequence 22, the sub-elements 36 of the product 32 can be detected by means of subject recognition.

[0150] In an alternative embodiment, the input sequence 22 is compared with the property database 44. The technology recognition model 38 compares the elements of the property database 44 with the input sequence 22. The technology recognition model 38 compares the elements of the property database 44 with the lemmatized tokens of the input sequence 22. The sub-elements 36 of the product 32 are recorded based on the comparison of the lemmatized token with the elements of the property database 44.

[0151] In processing step 82, at least one technology area of ​​the product 32 is determined. In processing step 82, at least one technology area of ​​the at least one sub-element 36 is determined. In processing step 82, at least one technology area is determined for each sub-element 36. The technology areas are determined using the technology recognition model 38. The technology areas are detected using the tokens of the input sequence 22. The technology area of ​​the at least one sub-element 36 is defined in abstract form using an abstract term of the sub-element 36 of the input sequence 22. For example, the technology recognition model 38 detects the term "operating element" in the input sequence 22 as sub-element 36. The abstract term "operating element" defines the technology area to be detected. The given technology area includes at least some "operable elements."

[0152] The technology area is captured based on the context information relating to the sub-element 36 of the product 32. The context information further defines the technology area of ​​the abstract concept of the sub-element 36. For example, in the case of a "partially fire-resistant operating element" from the input sequence 22, the technology area is an "operable element made of a fire-resistant material." Alternatively or additionally, at least one further technology area is defined based on the context information. In the given example, there is therefore the "operable element" as the first technology area and a "partially fire-resistant material" as the second technology area. The technology area is defined by the trigrams of the input sequence 22. Alternatively, the technology recognition model 38 has a text classifier. The text classifier uses the vectorized tokens to capture the technology areas of the input sequence 22.A Naive Bayes, Support Vector Machines or deep neural network is used as a text classifier.

[0153] The metadata 76 is collected from the technology areas of the sub-elements 36 of the product 32 of the input sequence 22. The metadata 46 is collected from the property database 44 using the technology areas of the sub-elements 36 of the product 32 of the input sequence 22. The metadata 60 is collected from the sequence database 48 using the technology areas of the sub-elements 36 of the product 32 of the input sequence 22.

[0154] In at least one embodiment of the method 10 according to the invention, parts of the input sequence 22 for which metadata 76 is collected are marked. The input sequence 22 is input into the algorithm 12 with the marked parts. The marked parts of the input sequence 22 are marked by a user via the user interface before the input sequence 22 is input 14. The technology recognition model 38 captures the marked parts of the input sequence 22. The marked parts define the technology areas of the input sequence 22.

[0155] In the further processing step 82, the concrete input sequence 24 is processed. In the further processing step 82, the properties and advantages of the product 32 of the concrete input sequence 24 are recorded. In the further processing step 82, the properties and advantages of the sub-elements 36 of the product 32 of the concrete input sequence 24 are recorded. The advantages and properties of the product 32 and the sub-elements 36 of the concrete input sequence 24 are transferred to a database processing 92 of the sequence database 48. The advantages and properties of the concrete input sequence 24 are transferred as property relationships to the database processing 92 of the sequence database 48. The advantages and properties of the concrete input sequence 24 are transferred as advantage relationships to the database processing 92 of the sequence database 48.The advantages and properties of the concrete input sequence 24 form the metadata of the concrete input sequence 24.

[0156] Processing step 84 for capturing the hierarchy structure of the input sequence 22

[0157] In at least one further processing step 84, a structure of the input sequence 22 is detected based on structural features. The input sequence 22 has the structural features. The structural features are detected using the technology recognition model 38. The technology recognition model 38 detects the structural features from the input sequence 22. The technology recognition model 38 uses the structural features to detect the structure of the product 32 of the input sequence 22. At least one of the structural features is at least one format-specific structural feature. The format-specific structural feature determines a text-based structure of the input sequence 22. At least one of the structural features is at least one text-specific structural feature. The text-specific structural feature determines a content-related structure of the input sequence 22.The input sequence 22 normally has a variety of format-specific structural features and format-specific structural features.

[0158] The technology recognition model 38 uses the structural features to capture a structural structure of the input sequence 22. The subsequences of the input sequence 22 are uniquely separated from one another by format-specific structural features. The format-specific structural features are given by a structural arrangement of the subsequences of the input sequence 22. The format-specific structural features are at least partially control characters. The format-specific structural features are at least partially special characters. The format-specific structural features are a number as unique features of the format-specific structure of the input sequence. As format-specific structural features, the subsequences of the input sequence 22 have numbering. The input sequence 22, as claims of the product 32, is numbered consecutively. Alternatively, punctuation of the input sequence 22 is captured as format-specific structural features.Punctuation as format-specific structural features can be applied because each claim consists of one sentence.

[0159] The subsequences of the input sequence 22 each have at least one text-specific structural feature. Based on the text-specific structural features, a unique assignment of the subsequences of the input sequence 22 to one another is determined. The assignment of the subsequences of the input sequence 22 is determined with the text-specific structural features by reference to at least one of the format-specific structural features of the subsequences of the input sequence 22. The assignment of the subsequences of the input sequence 22 is determined with the text-specific structural features by reference to the number of the consecutively numbered subsequences of the input sequence 22. For example, the text-specific structural feature is "according to claim [number]", or "according to claims [number] and [number]", or "according to one of the preceding claims", or a feature more comparable to a person skilled in the art.The subsequences of the input sequence 22 are structured hierarchically based on the recorded assignment of the subsequences. With the assignment of the subsequences, an evaluation value is assigned to the subsequences of the input sequence 22. For example, the hierarchical structure of the subsequences of the input sequence 22 is represented by an adjacency matrix. Each subsequence of the input sequence 22 represents a node of the adjacency matrix. The assignments of the subsequences of the input sequence 22 represent the connections of the adjacency matrix. The adjacency matrix is ​​an evaluation value for the hierarchical structure of the subsequences of the input sequence 22. Each hierarchical structure has an adjacency matrix. Each independent claim of the input sequence 22 has its own hierarchical structure. The adjacency matrix uniquely determines a hierarchical position of each claim of the input sequence 22.The hierarchical position is determined by the dependence of the claims of the input sequence 22.

[0160] Based on the text-specific structural features, a unique assignment of the sub-elements 36 to one another is determined. The technology recognition model 38 records a relationship between the sub-elements 36 of the product 32 in the input sequence 22. The technology recognition model 38 records an arrangement of the at least one sub-element 36 in the product 32 in the input sequence 22. The relationship between the sub-elements 36 of the product 32 is recorded using the text-specific structural features in the input sequence 22. The relationship between the sub-elements 36 of the product 32 is recorded for each sub-sequence of the input sequence 22. The relationship between the sub-elements 36 is recorded across sub-sequences using the recorded structural layout of the input sequence 22. The input sequence 22 has assignment terms between the sub-elements 36 of the product 32 as text-specific structural features.The relationship of the sub-elements 36 of the product 32 in the input sequence 22 is determined using the assignment terms. Using the assignment terms, an assignment of one sub-element 36 to another sub-element 36 can be unambiguously determined. Using the assignment terms, a property of a sub-element 36 can be unambiguously assigned to the sub-element 36. For example, an assignment term is “characterized by” and / or “characterized in that” and / or “with” and / or “have” and / or “wherein” and / or “which” and / or “in a [number] method step” and / or another assignment term that appears meaningful to a person skilled in the art. The assignment term “characterized in that,” for example, defines an already introduced sub-element 36 of the product 32 or a feature in further detail. The assignment term “characterized by,” for example, introduces a new sub-element 36 of the product 32.The assignment term "which," for example, introduces a property of sub-element 36 or describes a relationship to another sub-element 36 of product 32. The technology recognition model 38 detects a link between several text-specific structural features. The technology recognition model 38 detects a link between several assignment terms. The link further details the meaning of the assignment term. The link determines a probability for a meaning of the assignment term for the relationship of the sub-elements 36 in product 32. For example, in the following description in input sequence 22: "characterized by an operating element which has at least one locking element," the following assignment terms are detected: "characterized by," "which," and "has." Furthermore, a link between the assignment terms "which" and "has" is detected.This makes it possible to determine that the "locking element" is part of the "operating element". Each assignment term has at least one rule. Each assignment term has a predefined list with at least one rule. The assignment term is linked to the sub-element 36 of the product 32 according to the rule. The rule defines the text-specific position of the sub-element 36 in relation to the assignment term. The rule determines whether the assignment term is linked to another assignment term. The rule determines whether the assignment term links the sub-element 36 to another subsequence of the input sequence 22. The rule defines whether the assignment term defines a relationship between the sub-elements 36 connected to it or a link between a property and the sub-element 36.

[0161] The sub-elements 36 of the product 32 are hierarchically structured using the structural features of the input sequence 22. The sub-elements 36 are hierarchically structured in the sub-sequence of the input sequence 22. The hierarchical structuring of the sub-elements 36 defines the relationships between the sub-elements 36 of the input sequence 22. The at least one sub-element 36 is evaluated using the technology recognition model 38 based on the structural features. The technology recognition model 38 assigns an evaluation value to the at least one sub-element 36. The at least one sub-element 36 of the product 32 is assigned an evaluation value using the assignment concept. The evaluation value determines the hierarchical structuring of the sub-elements 36 in a sub-sequence of the input sequence 22.

[0162] The structure of the product 32 is determined by a combination of the recorded structural structure of the input sequence 22 and the recorded relationship of the sub-elements 36 in the input sequence 22. The structure of the product 32 is determined by a combination of the evaluation variable of the structural structure of the input sequence 22 with the evaluation variables of the relationship of the sub-elements 36 for each sub-sequence of the input sequence 22. For example, the combination is determined using an extended adjacency matrix. The extended adjacency matrix describes the relationship of the sub-elements 36 while taking into account the hierarchy structure of the sub-sequences of the input sequence 22. The extended adjacency matrix describes the structure of the product 32 of the input sequence 22. A graph 88 is created using the structure of the product 32. The graph 88 is created using the extended adjacency matrix.The subelements 36 of the product 32 of the input sequence 22 form the nodes of the graph 88. The technology areas of the subelements 36 of the product 32 of the input sequence 22 form the nodes of the graph 88. The relationships of the subelements 36 form the edges of the graph 88. With the graph 88, complex and interconnected subelements 36 of the product 32 can be efficiently mapped as data. The graph 88 is provided for collecting the metadata 76.

[0163] In at least one embodiment of the method 10 according to the invention, the relationships of the subelements 36 are determined using a Kl model of the technology recognition model 38. The Kl model of the technology recognition model 38 uses the text-specific features to capture the relationship between the subelements 36. The Kl model determines at least one rule for the assignment term. The Kl model calculates a probability for the rule of the assignment term. This allows for increased flexibility compared to the rule as a predefined list, since even rare combinations of assignment terms can be considered.

[0164] In at least one embodiment of the method 10 according to the invention, the relationships of the sub-elements 36 of the product 32 are further determined using the graphical input sequence 26. The structure of the product 32 of the graphical input sequence 26 is compared with the structure of the product 32 according to the structural features of the input sequence 22. Using the graphical input sequence 26, relationships of the sub-elements 36 that are missing from the input sequence 22 are supplemented. Using the graphical input sequence 26, the graph 88 is further supplemented. Using the graphical input sequence 26, further sub-elements 36 of the product 32 that are missing from the input sequence 22 are supplemented in relation to the sub-elements 36 of the input sequence 22. The further sub-elements 36 of the product 32 of the graphical input sequence 26 are supplemented as nodes to the graph 88.The relationship of the further sub-elements 36 of the product 32 of the graphic input sequence 26 are added to the graph 88 as edges.

[0165] In at least one further embodiment of the method 10 according to the invention, the relationships of the subelements 36 of the product 32 are supplemented with the subelements 36 of the specific input sequence 26. The subelements 36 of the specific input sequence 26 are assigned to the subelements 36 of the input sequence 22. The graph 88 is supplemented with the subelements 36 of the specific input sequence 26. The graph 88 is supplemented with the advantages of the specific input sequence 26. The graph 88 is supplemented with the properties of the specific input sequence 26.

[0166] Processing step 86 to collect the metadata 76

[0167] In at least one further processing step 86, metadata 76 is collected for the input sequence 22 using the technology recognition model 38. The metadata 76 is collected for the preprocessed input sequence. The metadata 76 is collected for the technology areas of the input sequence 22. The technology recognition model 38 links the metadata 76 to the sub-elements 36 of the product 32. The metadata 76 is collected for each sub-element 36 of the product 32 of the input sequence 22. The metadata 76 is collected for each sub-element 36 of the product 32 from the graphical input sequence 26. The metadata 76 is collected for the properties of the sub-elements 36. The metadata 76 is collected for the graph 88. The metadata 76 is collected for each node of the graph 88.

[0168] The technology recognition model 38 is used to derive at least the implementation 58 of the product 32. The technology recognition model 38 is used to derive at least the implementation 58 of at least one sub-element 36. The implementation 58 is derived based on the abstract form of the product 32. The technology recognition model 38 is used to derive the implementation 58 using the metadata 76. The technology recognition model 38 is used to derive a plurality of implementations 58 for the product 32 using the metadata 76. The implementations 58 are at least a subset of the collected metadata 76 for a generation 18 of the output sequence 30. The implementations 58 are derived based on the technology areas of the product 32. The implementations 58 are derived based on the technology areas of the sub-elements 36 of the product 32 of the input sequence 22. The implementations 58 are embodiments of the product 32.The implementations 58 are embodiments of the sub-elements 36 of the product 32. The implementations 58 of the collected metadata 76 at least partially have the same technology area as the product 32 of the input sequence 22. The respective implementations 58 of the sub-elements 36 of the product 32 at least partially have the technology area of ​​the respective sub-element 36 of the product 32. The implementations 58 of the product 32 are at least one level of abstraction deeper than the abstract form of the product 32 of the input sequence 22. The implementations 58 of the sub-elements 36 of the product 32 are at least one level of abstraction deeper than the abstract form of the sub-elements 36 of the product 32 of the input sequence 22. The implementations 58 are a concretization of the abstract form of the product 32 of the input sequence 22. The implementations 58 are an alternative embodiment of the product 32 of the input sequence 22.The implementations 58 are a concretization of the abstract form of the sub-elements 36 of the product 32 of the input sequence 22. The implementations 58 are an alternative embodiment of the sub-elements 36 of the product 32 of the input sequence 22.

[0169] The implementations 58 are derived using property relationships and / or advantage relationships of the metadata 76. A property relationship is a relationship between at least two sub-elements that at least partially have the same property. An advantage relationship is a relationship between at least two sub-elements that at least partially have the same advantage. The implementations 58 are derived based on the properties and / or advantages of the product 32 of the input sequence 22. The implementations 58 are derived based on the properties and / or advantages of the sub-elements 36 of the product 32 of the input sequence 22. Based on the properties and / or advantages of the metadata 76 relating to the sub-elements 36 of the product 32, at least one concretization of the sub-elements 36 is derived as an implementation 58.With the sub-elements 36 of the product 32, an alternative design of the sub-elements 36 is derived as an implementation 58 with the property and / or the advantage.

[0170] In processing step 86, the technology recognition model 38 performs a weighting 106 of the collected metadata 76. Using the weighting 106, the portion of the collected metadata 34 for generating 18 the output sequence 30 is determined from the collected metadata 76. After processing 16 the input sequence 22, depending on the weighting 106 of the collected metadata 76, the portion of the collected metadata 34 is output to the algorithm 12 for generating 18 the output sequence 30. After processing 16 the input sequence 22, a remainder of the portion of the collected metadata 34 is removed depending on the weighting 106.

[0171] The weighting 106 is calculated for each sub-element 36 of the product 32. The weighting 106 is calculated for the collected metadata 76 for each sub-element 36. The weighting 106 is composed of weighting variables. One weighting variable is calculated for the collected metadata 46 of the property database 44. Another weighting variable is calculated for the collected metadata 52 of the sequence database 48. Another weighting variable is calculated for the collected metadata 40 of the at least one further input sequence 28. The weighting variables are added to calculate the weighting 106. The weighting variables are multiplied by a weighting factor. The weighting factor is a predefined constant. The weighting factor defines a quality of the data source for the collected metadata 46, 60.The weighting factor is different for the property database 44, the sequence database 48, and the at least one further input sequence 28. The weighting 106 is calculated, for example, using equation (1).

[0172] (1 ) G = a ■ A + ß ■ B + Y ■ C

[0173] In equation (1), G is the weighting 106 for the collected metadata 46, 60 for each sub-element 36, a is the weighting factor and A is the weighting value of the property database 44, ß is the weighting factor and B is the weighting value of the sequence database 48, and y is the weighting factor and C is the weighting value of the at least one further input sequence 28. The collected metadata of the concrete input sequence 24 is output to the algorithm 12 without a weighting 106 as part of the collected metadata 34 for generation 18 of the output sequence 30. The calculation of the weighting values ​​will be described in further detail after a detailed description for collecting the metadata 76 from the respective databases 44, 48, 62, with an explanation of the weighting 106 of the collected metadata 76.

[0174] Collecting metadata 46 from the property database 44

[0175] In at least one database processing 90, the metadata 46 is collected from the properties database 44. The metadata 46 of the properties database 44 is collected by the algorithm 12. A connection to the properties database 44 is created using the algorithm 12. The product 32 of the input sequence 22 is detected using the technology recognition model 38 and passed to the algorithm 12. The sub-elements 36 of the product 32 of the input sequence 22 are detected using the technology recognition model 38 and passed to the algorithm 12. The technology areas of the product 32 are passed to the algorithm 12 for collecting the metadata 46 from the properties database 44. The technology areas of the sub-elements 36 of the product 32 are passed to the algorithm 12 for collecting the metadata 46 from the properties database 44.The sub-elements 36 of the product 32 are passed to the algorithm 12 as a graph 88 for collecting the metadata 46 from the property database 44. The algorithm 12 searches the property entries 74 of the property database 44 to compare the sub-elements 36 of the product 32. The algorithm 12 compares the sub-elements 36 of the product 32 with the elements of the property entries 74 of the property database 44. The algorithm 12 compares the technology areas of the sub-elements 36 with technology areas defined by the elements of the property entries 74 of the property database 44. The technology areas of the elements of the property entries 74 of the property database 44 are defined by an abstract concept of the elements of the property database 44. The algorithm 12 collects at least some of the matching property entries 74 of the property database 44 as metadata 46.Algorithm 12 collects the metadata 46 for which the sub-element 36 matches the element of the property entry 74. Algorithm 12 collects the metadata 46 for which the technology area of ​​the sub-elements 36 of the product 32 matches the technology areas of the elements of the property database 44. The metadata 46 of the property database 44 is passed by algorithm 12 to the technology recognition model 38 for further processing of the metadata 46.

[0176] Each property entry 74 in the property database 44 has at least one definition of the element. The definition describes the element of the property database 44. The definition at least partially describes a structure of the element of the property database 44. The definition at least partially defines a function of the element of the property database 44. The definition of each element of the property database 44, which is captured by the algorithm 12 starting from the input sequence 22, is collected as metadata 46 by the algorithm 12. Each property entry 74 in the property database 44 has at least one property of the element. The property of the element of the property database 44 is, for example, at least one material of the element. The property is a link to another property entry 74 of the property database 44 with another element or technology area.The property of the element of the property database 44 is, for example, that a storage medium can also be configured as a cloud. The properties of the property database 44 are transferred to the technology recognition model 38 as collected metadata 46. The technology recognition model 38 links the properties of the property database 44 with the respective sub-element 36 of the product 32 of the input sequence 22. Each property entry 74 of the property database 44 has at least one advantage of the element. Each element of the property database 44 has at least one advantageous property. An advantageous property or advantage of the property database 44 is, for example, that an element is "fireproof." The advantage of the property database 44 is a link to another property entry 74 of the property database 44 with another element or another technology area.The technology recognition model 38 links the advantages of the properties database 44 with the respective sub-element 36 of the product 32 of the input sequence 22. The advantages of the elements of the properties database 44 are transferred to the technology recognition model 38 as collected metadata 46. The properties of the elements of the properties database 44 are used to link to further property entries 74 of the properties database 44 with further elements. The advantages of the elements of the properties database 44 are used to link to further property entries 74 of the properties database 44 with further elements. The link to a further technology area is a link to a further element of the properties database 44. The further element of the properties database 44 is transferred to the technology recognition model 38 as collected metadata 46.The additional element of the property database 44 is one link level lower than the element that is directly recorded for the sub-element 36 of the input sequence 22. A further element is recorded for the additional element of the property database 44 based on the advantages or properties of the additional element of the property database 44. The additional elements of the property database 44 are passed to the technology recognition model as collected metadata 46 for further processing. Each property entry 74 of the property database 44 has at least one given evaluation factor of the element. The given evaluation factor influences the weighting 106 of the element when determining the part of the metadata 34. The given evaluation factor is stored by the user in the property database 44 for each property entry 74.

[0177] The definitions of the property database 44 of the sub-elements 36 of the input sequence 22 are output directly as part of the part of the collected metadata 34. The advantages and properties of the matching property entries 74 of the property database 44 are passed to the database processing 92 of the sequence database 48. The advantages and properties of the matching property entries 74 of the property database 44 are passed as property relationships to the database processing 92 of the sequence database 48. The advantages of the matching property entries 74 of the property database 44 are passed as advantage relationships to the database processing 92 of the sequence database 48. The additional elements of the property database 44 captured via the link are passed to the technology recognition model 38 for the weighting 106 of the collected metadata 46 of the property database 44.Advantages and / or properties captured with the additional elements are transferred to the technology recognition model 38 as collected metadata 46 of the property database 44. The additional elements, as collected metadata 46 of the property database 44, at least partially constitute the implementations 58 of the sub-elements 36 of the product 32 of the input sequence 22.

[0178] In at least one embodiment, the property database 44 has a plurality of definitions for each element of the property database 44. The property database 44 has a different company-specific definition for each element of the property database 44 for a plurality of companies. The algorithm 12 collects as metadata 46 only the company-specific definition that matches the company in the metadata of the input sequence 22. The property database 44 has at least one general definition. The general definition is collected as metadata 46 if the company is missing as metadata of the input sequence 22. The general definition is collected as metadata 46 if a company-specific definition for the company in the metadata of the input sequence 22 is missing. Alternatively or additionally, the property database 44 has a plurality of company-specific properties for each element.The property database 44 contains several company-specific advantages for each element. The algorithm 12 collects, as metadata 46 of the property database 44, only the company-specific property of the element that matches the company of the metadata of the input sequence 22. The algorithm 12 collects, as metadata 46 of the property database 44, only the company-specific advantage of the element that matches the company of the metadata of the input sequence 22.

[0179] The properties database 44 has at least one checklist 72 for at least one technology area. The properties database 44 has at least one checklist 72 for the technology area of ​​the product 32. The properties database 44 has a plurality of checklists 72 for different technology areas. The properties database 44 has at least one checklist 72 for a complex technology. The properties database 44 has at least one checklist 72 for a software technology. The properties database 44 has at least one checklist 72 for a sustainable technology. The properties database 44 has at least one checklist 72 for the technology area of ​​at least one sub-element 36. The algorithm 12 compares the technology area of ​​the product 32 of the input sequence 22 with the technology area of ​​the checklists 72.The checklist 72 defines rules for the advantages and / or properties of the technology area. If a technology area of ​​the product 32 of the input sequence 22 matches the technology area of ​​one of the checklists 72, the rules of the checklist 72 are applied. If further elements, advantages and / or properties are listed in the checklist 72, these are collected as metadata 46. Based on the rules of the matching checklist 72, further property entries 74 of the property database 44 are collected as metadata 46. Based on the rules of the checklist 72, further elements of the property database 44 are collected as metadata 46. Based on the rules of the checklist 72, further advantages and / or properties of the property database 44 are collected. Based on the rules of the checklist 72, the further elements for the generation 18 of the output sequence 30 are passed to the algorithm 12.

[0180] Collecting metadata 60 from the sequence database 48

[0181] In at least one further database processing 92, the metadata 60 is collected from the sequence database 48. The metadata 60 of the sequence database 48 is collected by the algorithm 12. A connection to the sequence database 48 is created using the algorithm 12. The entire metadata 60 of the sequence database 48 is formed at least from the metadata 46 of the sequences 50 of the sequence database 48, the metadata 40 of the at least one further input sequence 28 stored in the sequence database 48, the trend properties 56 of the sequence database 48, and at least one relationship matrix 110 of the sequence database 48.

[0182] The sequence database 48 is at least partially divided into classes 54. The sequences 50 in the sequence database 48 are divided into classes 54. The sequences 50 of the sequence database 48 are captured using algorithm 12. The sequences 50 of the sequence database 48 are passed to the technology recognition model 38 for further processing using algorithm 12. The sequences 50 of the sequence database 48 are passed to the technology recognition model 38 for collecting the metadata 52 relating to the input sequence 22. The technology recognition model 38 collects the metadata 52 of the sequence database 48 relating to the input sequence 22.

[0183] Classes 54 of the sequence database 48 are defined by the respective assigned company. Each class 54 of the sequence database 48 contains sequences 50 of the company. Each class 54 of the sequence database 48 contains the company's intellectual property rights as sequences 50. Each class 54 of the sequence database 48 contains intellectual property rights relevant to the company as sequences 50. Each class 54 of the sequence database 48 contains prior art documents for the company as sequences 50.

[0184] The input sequence 22 is assigned to one of the classes 54 based on metadata of the input sequence 22. The metadata of the input sequence 22 for assignment to a class 54 is the company of the input sequence 22. The input sequence 22 is assigned to class 54 of the sequence database 48 where the company of class 54 matches the company of the input sequence 22. The sequences 50 of the assigned class 54 are at least partially transferred to the technology recognition model 38 for collecting the metadata 52 of the sequence database 48. The metadata 52 of the product 32 is taken from the assigned class 54 of the sequence database 48. During processing 16 of the input sequence 22, only metadata 52 is collected from the sequences 50 of the sequence database 48 that correspond to class 54 of the input sequence 22.When weighting the advantages and / or properties of the sequence database 48, only those sequences 50 of the sequence database 48 that correspond to class 54 of the input sequence 22 are used. The output sequence 30 is stored in class 54, which corresponds to class 54 of the sequence database 48, upon provision 20 of the output sequence 30. The sequences 50 of the assigned class 54 of the sequence database 48 are transferred to the technology recognition model 38 for processing. The algorithm 12 collects all sequences 50 of the assigned class 54 of the sequence database 48 and transfers sequences 50 to the technology recognition model 38 for collecting the metadata 52. Alternatively, a preselection of the sequences 50 of the assigned class 54 is carried out based on the technology area of ​​the product 32 of the input sequence 22.The preselection of the sequences 50 is made by comparing the technology area of ​​the product 32 with the technology areas of the sequences 50 of the assigned class 54 of the sequence database 48. The sequences 50 that at least partially have the same technology area as the product 32 of the input sequence 22 are passed to the technology recognition model 38 for collecting the metadata 52. The technology areas of the sequences 50 of the sequence database 48 are defined by the other products of the respective sequences 50 of the sequence database 48. Alternatively, the technology areas of the sequences 50 of the sequence database 48 are IPC or CPC classes. In at least one further embodiment, the technology areas of the sequences 50 of the respective classes 54 are defined by a user. Alternatively, the technology areas of the sequences 50 of the sequence database 48 are defined by the company of the respective class 54.

[0185] In at least one embodiment of the method 10 according to the invention, the sequences 50 of the sequence database 48 are evaluated with regard to their relevance to the company. The evaluation of the relevance of the sequences 50 of the sequence database 48 is performed by the user. Alternatively, the evaluation of the relevance of the sequences 50 of the sequence database 48 is performed by the company of the respective class 54. The evaluation of the relevance of the sequences 50 is taken into account when weighting the collected metadata 52 of the respective sequence 50.

[0186] In a further embodiment of the method 10 according to the invention, the sequence database 48 is divided into classes 54 according to the company in only a portion of the sequence database 48. In at least one further portion of the sequence database 48, the sequence database 48 is divided into technology classes according to technology areas. The sequence database 48 has further sequences that are divided into technology classes. The further sequences of the technology classes are missing from the classes 54 that are divided according to the company. The metadata of the input sequence 22 for assignment to one of the technology classes is the technology area of ​​the input sequence 22 and the product 32.

[0187] Sequence analysis98 for processing the sequences 50

[0188] A sequence analysis 98 is performed using the technology recognition model 38. In the sequence analysis 98, the sequences 50 of the sequence database 48 that the algorithm 12 passes to the technology recognition model 38 are processed. In the sequence analysis 98, the metadata 52 of the sequences 50 of the sequence database 48 are collected using the technology recognition model 38.

[0189] In at least one step, the technology recognition model 38 is used to identify the sequences 50 of the

[0190] Sequence database 48 is prepared for collecting metadata 52. The sequences 50 of the

[0191] Sequence database 48 is prepared according to the input sequence 22 in processing step 80. The sequences 50 of the sequence database 48 are tokenized, normalized, and vectorized. In at least one further step, the technology areas of the sequences 50 of the sequence database 48 are determined using the technology recognition model 38. The further products of the sequences 50 are determined using the technology recognition model 38. The further sub-elements of the further products of the sequences 50 are determined using the technology recognition model 38. The technology areas of the further products of the sequences 50 are determined using the technology recognition model 38. The technology areas of the further sub-elements of the further products of the sequences 50 are determined using the technology recognition model 38. The advantages of the further products of the sequences 50 of the sequence database 48 are determined using the technology recognition model 38.The technology recognition model 38 is used to determine properties of the additional products of the sequences 50 of the sequence database 48. The technology recognition model 38 is used to determine advantages of the additional sub-elements of the additional products. The technology recognition model 38 is used to determine properties of the additional sub-elements of the additional products. The technology areas, the advantages, and the properties of the sequences 50 of the sequence database 48 are determined according to processing step 82 when determining the technology areas of the input sequence 22.

[0192] The additional sub-elements of the additional products of the sequences 50 of the sequence database 48 are extracted as a sub-element data pool from the sequences 50 of the sequence database 48. The advantages, properties, and technology areas of the additional sub-elements are linked to the respective sub-elements of the sub-element data pool. The sub-element data pool only contains the additional sub-elements of the additional products and the respective properties, advantages, and technology areas of the additional sub-elements of the sequences 50 of the sequence database 48.

[0193] To collect the metadata 52 from the sequences 50, a portion of the further subelements is determined from the subelement data pool using the technology recognition model 38. Using the technology recognition model 38, the portion of the further subelements from the subelement data pool is determined based on the abstract form of the product 32 of the input sequence 22. The portion of the further subelements from the subelement data pool are implementations 58 of at least a portion of the product 32 of the input sequence 22. The implementations 58 are derived from the subelement data pool of the sequences 50 of the sequence database 48 as part of the further subelements. The portion of the further subelements from the subelement data pool is determined using the property relationships of the specific input sequence 24. The portion of the further subelements from the subelement data pool is determined using the advantage relationships of the specific input sequence 24.The portion of the further sub-elements from the sub-element data pool is determined using the property relationships of the property database 44. The portion of the further sub-elements from the sub-element data pool is determined using the advantage relationships of the property database 44. Using the technology recognition model 38 and the graph 88, the portion of the further sub-elements from the sub-element data pool is at least partially determined. The portion of the further sub-elements from the sub-element data pool is determined based on the structure of the product 32 of the input sequence 22. Using the technology recognition model 38, the portion of the further sub-elements from the sub-element data pool is determined based on the recorded hierarchy structure of the sub-elements 36 of the product 32 of the input sequence 22. Based on the technology areas of the sub-elements 36 of the input sequence 22 on the same hierarchy level, the alternative design is derived as implementation 58.Based on the technology areas, the concretization is derived as an implementation 58. The technology recognition model 38 takes into account at least the structural features within the input sequence 22 to derive the implementation 58 from the sub-element data pool. According to the graph 88 for the structure of the product 32 of the input sequence 22, an implementation 58 is determined as a further sub-element from the sub-element data pool for at least one sub-element 36 of the product 32. Using the technology recognition model 38, an implementation 58 is determined as a further sub-element from the sub-element data pool based on a property and advantages of the sub-element 36 of the product 32 of the input sequence 22.Using the technology recognition model 38, the implementation 58 is derived as a further subelement from the subelement data pool based on the property relationships and advantage relationships of the specific input sequence 24, starting from one of the subelements 36 of the product 32 of the input sequence 22. Using the technology recognition model 38, the implementation 58 is derived as a further subelement from the subelement data pool based on the property relationships and advantage relationships of the property database 44, starting from one of the subelements 36 of the product 32 of the input sequence 22.

[0194] Using the technology recognition model 38, further conversions 58 of the sub-elements 36 of the product 32 of the input sequence 22 are recorded from the sub-element data pool across various hierarchy levels, starting from the graph 88. The conversions 58 to the sub-elements 36 of the product 32 are determined according to the graph 88, starting from the respective sub-elements 36 of the product 32 of the input sequence 22 at a higher hierarchy level. The sub-elements 36 of the product 32 of the input sequence 22 that are one hierarchy level higher each have or form the corresponding sub-elements 36 of the product 32 of the input sequence 22 at a lower hierarchy level, according to the graph 88. The conversions 58 starting from the graph 88 form the further conversions 58 of the sub-elements 36 of the lower hierarchy level, starting from the sub-elements 36 of the higher hierarchy level.The further implementations 58 of the sub-elements 36 of lower hierarchy levels are determined as part of the further sub-elements from the sub-element data pool. The part of the further sub-elements from the sub-element data pool forms the collected metadata 52 of the sequences 50 of the sequence database 48. The part of the further sub-elements from the sub-element data pool forms the collected metadata 52 of the sequences 50 of the sequence database 48 for the sub-elements 36 of the product 32 of the input sequence 22. The part of the further sub-elements from the sub-element data pool is linked with a factor for a link level. The link level of the part of the further sub-elements from the sub-element data pool depends on the link level of the element of the property database 44, by means of whose property relationship and / or advantage relationship the further sub-element from the sub-element data pool is determined.Alternatively or additionally, the technology recognition model 38 defines linking levels based on the graph 88 according to property relationships and / or advantage relationships within the sequences 50 of the sequence database 48. The linking levels are determined based on the subelements 36 of the product 32 of the input sequence 22 according to the graph 88 according to the property relationships and / or advantage relationships of the sequences 50 of the sequence database 48.

[0195] Using the portion of the additional sub-elements from the sub-element data pool of the sequences 50 of the sequence database 48 as collected metadata 52, a relationship matrix 52 is generated with the technology recognition model 38. Using the technology recognition model 38, the relationship matrix 78 is generated based on the property relationship and / or advantage relationships of the metadata 52 of the sub-elements 36. The relationship matrix 78 is generated with the technology recognition model 38 based on the portion of the additional sub-elements from the sub-element data pool. The relationship matrix 78 creates a link between at least the sub-elements 36 of the product 32. The relationship matrix 78 creates a link between the sub-elements of the sequences 50 of the sequence database 48.The linking of the relationship matrix 78 is generated by the property relationships and advantage relationships between the sub-elements 36 of the input sequence 22 and the part of the further sub-elements from the sub-element data pool of the sequences 50 of the sequence database 48. The collected metadata 52 of the sequence database 48 as part of the further sub-elements from the sub-element data pool of the sequences 50 of the sequence database 48 at least partially form the relationship matrix 78. The relationship matrix 78 forms the collected metadata 52 of the sequences 50 of the sequence database 48 linked to the sub-elements 36 of the input sequence 22. The relationship matrix 78 forms the collected metadata 52 of the sequences 50 of the sequence database 48 with the properties and advantages of the sub-elements 36 of the input sequence 22 resulting from the property relationships and advantage relationships.The relationship matrix 78 has a reduced amount of data compared to the sequences 50 of the sequence database 48. The relationship matrix 78 includes metadata 52 of the sequences 50 of the sequence database 48 that is at least partially relevant for generating 18 the output sequence 30. The relationship matrix 78 is provided as collected metadata 52 of the sequences 50 of the sequence database 48 for weighting.

[0196] Sequence analysis 100 for processing the at least one further input sequence 28

[0197] A sequence analysis 100 of the at least one further input sequence 28 is carried out using the technology recognition model 38. In the sequence analysis 100 of the at least one further input sequence 28, the at least one further input sequence 28 is processed using the technology recognition model 38. In the sequence analysis 100 of the at least one further input sequence 28, the metadata 40 of the at least one further input sequence 28 is at least partially collected using the technology recognition model 38. In at least one step, the at least one further input sequence 28 is prepared using the technology recognition model 38 for collecting the metadata 40. The at least one further input sequence 28 is prepared according to the sequences 50 of the sequence database 48. The at least one further input sequence 28 is prepared according to the input sequence 22 in the processing step 80.In at least one further step, the technology areas of the at least one further input sequence 28 are determined using the technology recognition model 38. The properties of the at least one further input sequence 28 are determined using the technology recognition model 38. The advantages of the at least one further input sequence 28 are determined using the technology recognition model 38. The technology areas, properties, and advantages of the at least one further input sequence 28 are determined using the technology recognition model 38 according to the sequences 50 of the sequence database 48.

[0198] The additional sub-elements of the additional product of the at least one additional input sequence 28 are removed from the sequences 50 as a sub-element data pool. The advantages and properties of the additional sub-elements are removed from the at least one additional input sequence 28 as a sub-element data pool. The sub-element data pool contains only the additional sub-elements of the additional products and the respective properties and advantages of the additional sub-elements of the at least one additional input sequence 28. The properties and advantages of the sub-element of the at least one additional input sequence 28 are stored for each additional sub-element of the additional products. The sub-element data pool of the at least one additional input sequence 28 forms the collected metadata 40 of the at least one additional input sequence 28.The partial element data pool of the at least one further input sequence 28 forms the collected metadata 40 of the at least one further input sequence 28 for the partial elements 36 of the input sequence 22. The partial elements of the partial element data pool of the at least one further input sequence 28 are each linked with a factor for a linking level. Starting from the partial elements 36 of the product 32 of the input sequence 22 according to the graph 88, the linking levels are defined as linking levels via property relationships and / or advantage relationships of the further partial elements of the further product of the at least one further input sequence 28.

[0199] A relationship matrix 42 is generated using the partial element data pool of the at least one further input sequence 28 as collected metadata 40. For the at least one further input sequence 28, the relationship matrix 42 is generated using the entire partial element data pool of the at least one further input sequence 28. The relationship matrix 42 has the property relationships and advantage relationships of the partial elements of the input sequence 22 as collected metadata 40 from the partial element data pool. The relationship matrix 42 is generated using the technology recognition model 38 from the partial element data pool of the further input sequence 28. The relationship matrix 42 has a link between at least the partial elements 36 of the product 32 of the input sequence 22 and the partial elements from the partial element data pool of the at least one further input sequence 28.The linking of the relationship matrix 42 is generated by the property relationships and advantage relationships between the sub-elements 36 of the input sequence 22 and the sub-elements from the sub-element data pool of the at least one further input sequence 28. The collected metadata 40 of the at least one further input sequence 28 as a sub-element data pool at least partially forms the relationship matrix 42. The relationship matrix 42 forms the collected metadata 40 of the at least one further input sequence 28 linked to the sub-elements 36 of the input sequence 32. The relationship matrix 42 forms the collected metadata 40 of the at least one further input sequence 28 with the properties and advantages of the sub-elements 36 of the input sequence 22 resulting from the property relationships and advantage relationships. The relationship matrix 42 has a reduced amount of data compared to the at least one further input sequence 28.The relationship matrix 42 contains all metadata 40 of the at least one further input sequence 28 relevant for generating 18 the output sequence 30. The relationship matrix 42 is provided as collected metadata 40 of the at least one further input sequence 28 for weighting.

[0200] Recording the relationship matrices 66 of the sequence database 48

[0201] During database processing 92 of the sequence database 48, a relationship matrix acquisition 102 is carried out. A plurality of relationship matrices 66 are stored in the sequence database 48. The relationship matrices 66 each have a plurality of elements. The relationship matrices 66 each have a plurality of advantages of the elements. The relationship matrices 66 each have a plurality of properties of the elements. The elements in the relationship matrices 66 are each linked to one another. The elements in the relationship matrices 66 are each linked to one another via property relationships and / or advantage relationships. The sequence database 48 has at least one relationship matrix 66 starting from a product. The sequence database 48 has at least one relationship matrix 66 starting from an element. The relationship matrices 66 each have further linked elements starting from the product or element.The relationship matrices 66 each comprise several different technology areas based on the product or element. At least some of the relationship matrices 66 are from past processing of past input sequences. At least some of the relationship matrices 66 are at least partially created using the additional sequence database 62. The creation of the relationship matrices 66 from the additional sequence database 62 is described in further detail after a trend detection 96 and a merging of the collected metadata 46, 60 during a data set generation 112. The relationship matrices 66 are at least partially assigned to the respective classes 54 of the sequence database 62. The relationship matrices 66 are indexed for assignment to the respective class 54. In the relationship matrix detection 102, only relationship matrices 66 from the class 54 that matches the class 54 of the input sequence 22 are recorded.Starting from the input sequence 22, a plurality of relationship matrices 66 are selected. At least one of the relationship matrices 66 is selected based on the technology area of ​​the product 32 of the input sequence 22. At least one of the relationship matrices 66 is selected based on the technology area of ​​the sub-elements 36. Using the technology recognition model 38, one of the relationship matrices 66 is selected for each sub-element 36 of the product 32 of the input sequence 22. The same relationship matrix 66 can also be selected for multiple sub-elements 36 of the product 32 of the input sequence 22.

[0202] At least the relationship matrix 66 is selected which has at least the sub-element 36 of the product 32 of the input sequence 22 as an element of the relationship matrix 66. If multiple relationship matrices 66 have the sub-element 36 of the product 32 of the input sequence 32 as an element of the relationship matrices 66, at least the relationship matrix 66 is selected which has at least one further sub-element 36 of the product 32 of the input sequence 22 as an element of the relationship matrix 66. Alternatively, the relationship matrix 66 is selected in which the element from which the relationship matrix 66 originates corresponds to the sub-element 36 of the product 32 of the input sequence 22. Alternatively, multiple relationship matrices 66 which have the sub-element 36 of the product 32 of the input sequence 22 as an element of the relationship matrices 66 are combined to form a relationship matrix 110.

[0203] The selected relationship matrices 66 are combined into a relationship matrix 110 for provision. The combined relationship matrix 110 has the subelements 36 of the product 32 of the input sequence 22 as elements. The combined relationship matrix 110 has properties of the subelements 36 of the product 32 of the input sequence 22. The combined relationship matrix 110 has advantages of the subelements 36 of the input sequence 22. The combined relationship matrix 110 has the implementations 58 of the subelements 36 of the input sequence 22 as elements. The combined relationship matrix 110 is provided for weighting.

[0204] Trend detection 96 of trend properties 54

[0205] When collecting the metadata 60 from the sequence database, a trend detection 96 is performed to detect trend properties 56. The trend properties 56 for the respective classes 54 are stored in the sequence database 48. The trend properties 56 are transferred to the technology recognition model 38 for processing 16 of the input sequence 22 with the algorithm 12. The trend properties 54 are technology areas that can be combined with a product 32 across technologies. The trend properties 56 are properties that can be combined with a technical product across technologies. The trend properties 56 are properties and / or technology areas that are the focus of the assigned class 54 of the sequence database. The trend properties 54 are properties and / or technology areas that are the focus of the company of the assigned class 54.Trend characteristics 56 are characteristics and / or technologies that are future-relevant for the company assigned to class 54. Trend characteristics 56 are promising technology areas and / or characteristics. Trend characteristics 56 are sustainable technology areas and / or characteristics.

[0206] The trend characteristics 56 of the assigned class 54 are included in the calculation of the weighting 106 of the metadata 76. The metadata 76 that has at least one technology area, property, and / or benefit that at least partially matches the technology areas, properties, and / or benefits of the trend characteristics 56 are positively weighted in the weighting 106 of the collected metadata 76.

[0207] The trend properties 56 are defined by a user in the sequence database 48 for the respective class 54. Alternatively or additionally, the trend properties 56 are defined by the company of the respective classes 54. Alternatively or additionally, the trend properties 56 are at least partially determined by the input 14 of past input sequences 22, which repeatedly exhibit the same novel or complex properties.

[0208] Merging 104 of the metadata 60 of the sequence database 48

[0209] In at least one merge 104, the collected metadata 52, 40 of the sequences 50 of the sequence database 48 and the at least one further input sequence 28 are merged using the technology recognition model 38. During the merge, the relationship matrix 78 of the sequences 50 of the sequence database 48, the relationship matrix 42 of the at least one further input sequence 28, and the combined relationship matrix 110 of the sequence database 48 are merged. After the merge 104 of the relationship matrices 42, 76, 110 of the sequence database 48, formed at least partially from the metadata 52, 40, and the at least one further input sequence 28, a merged relationship matrix 94 is provided. The merged relationship matrix 94 forms the entire collected metadata 60 of the sequence database 48.

[0210] The merged relationship matrix 94 has all sub-elements 36 of the product 32 of the input sequence 22 as elements. The merged relationship matrix 94 has the implementations 58 of the sub-elements 36 of the product 32 of the input sequence 22 as elements. The merged relationship matrix 94 has the advantages of the elements. The merged relationship matrix 94 has the properties of the elements. The elements of the merged relationship matrix 94 are at least partially linked to one another. The elements of the merged relationship matrix 94 at least partially have a link that at least partially links the elements in the merged relationship matrix 94 to one another. The linking of the elements of the merged relationship matrix 94 defines the relationship of the respective elements to one another.During the merging 104, identical elements of the respective relationship matrices 42, 78, 110 from the sequences 50 of the sequence database 48, the at least one further input sequence 28, and / or the combined relationship matrix 110 are combined to form one element of the merged relationship matrix 94. During the merging 104, different elements of the respective relationship matrix 42, 78, 110 from the sequences 50 of the sequence database 48, the at least one further input sequence 28, and / or the combined relationship matrix 110, which have the same property relationship and / or advantage relationship, are at least partially linked to one another in the merged relationship matrix 94.

[0211] Each element of the merged relationship matrix 94 is indexed with respect to its origin. Each element of the merged relationship matrix 94 is indexed with respect to the origin of the relationship matrix 78 of the sequences 50 of the sequence database 48, the relationship matrix 42 of the at least one further input sequence 28, and the combined relationship matrix 110. Each element of the merged relationship matrix 94 is indexed for weighting. Each element of the merged relationship matrix 94 is indexed with respect to the presence of a trend characteristic 56. The merged relationship matrix 94 is provided for further processing.

[0212] Merging 114 to the final relationship matrix 116

[0213] In at least one further merge 114, the collected metadata 46 of the property database 44 and the merged relationship matrix 94 are merged with the technology recognition model 38. During the further merge 114, the merged relationship matrix 94 is supplemented with the further elements of the property database 44 as collected metadata 46. During the further merge 114, a final relationship matrix 116 is created. During the further merge 114, identical elements of the merged relationship matrix 94 and the collected metadata 46 of the property database 44 are combined in the final relationship matrix 116. The elements of the final relationship matrix 116 at least partially have a link 118, which at least partially links the elements in the final relationship matrix 116 to one another.During the further merging 114, various elements of the merged relationship matrix 94 and the collected metadata 46 of the property database 44, which have the same property relationship and / or benefit relationship, are at least partially linked to one another in the final relationship matrix 116. The property relationships and / or benefit relationships arise from the benefits and / or properties of the elements of the collected metadata 46 of the property database 44. The links 118 of the elements of the final relationship matrix 116 each connect the elements of the final relationship matrix 116 that form the sub-elements 36 of the product 32 of the input sequence 22 with the elements that form the implementation 58 of the sub-elements 36 of the product 32 of the input sequence 22.The links 118 of the final relationship matrix 116 connect the elements of the final relationship matrix 116 with the corresponding advantages and / or properties of the respective elements.

[0214] Each element of the collected metadata 46 of the property database 44 is indexed in the final relationship matrix 116 with respect to its origin. Each element of the final relationship matrix 116 is indexed with respect to the origin of the property database 44, the sequence database 48, and / or the at least one further input sequence 28.

[0215] Data query 108 for data set generation of additional elements

[0216] After the further merging 114, a data query 108 checks whether any further sub-elements 36 of the product 32, which are described at least in the input sequence 22, are missing in the final relationship matrix 116. The sub-elements 36 are included in the final relationship matrix 116 using the technology recognition model 38.

[0217] During the data query 108, a check is performed to determine whether one of the links 118 to a further element of the final relationship matrix 116 is missing for at least one further sub-element 36 of the product 32 described in the input sequence 22. The technology recognition model 38 is used to check whether a specific minimum number of links 118 per sub-element 36 of the product 32 of the input sequence 22 has been reached in the final relationship matrix 116. The minimum number of links 118 per sub-element is at least one link 118. Alternatively, the required minimum number of links 118 per sub-element 36 of the product 32 of the input sequence 22 is specified by a user.

[0218] In at least one embodiment of the method 10, a minimum number of links 118 varies for each sub-element 36 of the product 32 of the input sequence 22. The minimum number of links 118 depends on the created graph 88 of the input sequence 22. The minimum number of links 118 depends on the structure of the input sequence 22. The minimum number of links 118 differs depending on the sub-sequence of the input sequence 22 as a claim in which the sub-element 36 is described, or depending on the position of the sub-element 36 in the input sequence 22. Alternatively, the minimum number of links 118 depends on the technology area of ​​the sub-element 36 of the product 32 of the input sequence 22. Alternatively, the minimum number of links 118 depends on a relevance of the respective sub-element 36. The relevance of the sub-elements 36 is at least partially determined based on the specific input sequence 24.The relevance of the sub-elements 36 of the product 32 of the input sequence 22 is determined based on the properties and / or advantages of the specific input sequence 24. If a property and / or advantage of the sub-element 36 of the product 32 of the input sequence 22 in the final relationship matrix 116 matches a property and / or advantage of the specific input sequence 24, the sub-element 36 is rated as having high relevance. Sub-elements 36 of the product 32 of the input sequence 22 that are rated as having high relevance have a higher minimum number of links 118 than other sub-elements 36 of the product 32 of the input sequence 22. Alternatively or additionally, the minimum number varies depending on the origin of the sub-element 36 of the product 32 from the input sequence 22 or the graphical input sequence 26.The minimum number is different for the sub-elements 36 described in the input sequence 22 and the sub-elements 36 depicted in the graphical input sequence 26 and missing from the input sequence 22. The sub-elements 36 of the product 32 from the input sequence 22 have a higher minimum number than the sub-elements 36 of the product 32 from the graphical input sequence 26 that are missing from the input sequence 22.

[0219] If, during the data query 108, the minimum number of links 118 in the final relationship matrix 116 is recorded for all sub-elements 36 of the product 32 of the input sequence 22, the final relationship matrix 116 is provided for a weighting 106 of the elements and links 118 of the final relationship matrix 116. If at least one sub-element 36 of the product 32 described in the input sequence 22 lacks the minimum number of links 118 to at least one further element of the final relationship matrix 116, a data set generation 112 is executed. The sub-element data set is generated for the sub-element 36 that is missing and / or insufficiently present in the sequence database 48 and / or the property database 44. The sub-element data set is generated for the sub-element 36 that is missing from the final relationship matrix 116.

[0220] The technology recognition model 38 transfers the sub-elements 36 of the product 32 of the input sequence 22, for which the sub-element data set is generated, to the algorithm 12. The sub-element data set is generated using a comprehensive technology term of the sub-element 36. The algorithm 12 generates the sub-element data set based on the product 32. The novel and patentable core of the product 32 of the input sequence 22 is removed with the comprehensive technology term during the generation of the sub-element data set. The sub-element data set is, for example, an IPC class and / or CPC class and / or another comprehensive technology term of the product 32 or the sub-element 36.

[0221] The partial element data set is used to determine a subset of the plurality of sequences 64 in the additional sequence database 62. The partial element data set is used to transfer the subset of the plurality of sequences 64 to the sequence database 48 for generating an additional relationship matrix 66. The subset of the plurality of sequences 64 comprises all sequences or all intellectual property rights of the IPC class and / or CPC class of the additional sequence database 62. The subset of the plurality of sequences 64 in the additional sequence database 62 is transferred to the sequence database 48 for generating the additional relationship matrix 66. The subset of the plurality of sequences 64 in the additional sequence database 62 is processed from the sequence database 48 using the technology recognition model 38. The processing of the subset of the plurality of sequences 64 of the further sequence database 62 is according to the sequence analysis 98 of the sequences 50 of the sequence database 48.The further relationship matrix 66 is at least substantially in accordance with the relationship matrices 66. The further relationship matrix 66 comprises the products and subelements of the subset of the plurality of sequences 64 of the further sequence database 62. The further relationship matrix 66 comprises the links between the products and subelements of the subset of the plurality of sequences 64. The further relationship matrix 66 comprises the subelements 36 of the product 32 of the input sequence 22, to which at least one link is missing in the final relationship matrix 116. The further relationship matrix 66 comprises links to the subelements 36 of the product 32 of the input sequence 22, to which at least one link 118 is missing in the final relationship matrix 116.

[0222] After generating the further relationship matrix 66, the transferred subset of the plurality of sequences 64 of the further sequence database 62 is removed from the sequence database 48. The further relationship matrix 66 is incorporated into the final relationship matrix 116 with the technology recognition model 38.

[0223] The elements, properties, and / or advantages of the final relationship matrix 116 that lack text preparation are at least tokenized and vectorized. The final relationship matrix 116 forms all collected metadata 76 from the property database 44, the sequence database 48, the at least one further input sequence 28, and the further sequence database 62. The final relationship matrix 116 is provided for weighting 106. The portion of the collected metadata 34 for the generation 18 of the output sequence 30 is determined from the final relationship matrix 116.

[0224] Weighting 106 of the links 118 of the final relationship matrix 116

[0225] During weighting 106, each element, benefit, and / or property of the final relationship matrix 116 is weighted. The weighting 106 determines the portion of the collected metadata 34 for the generation 18 of the output sequence 30 from the final relationship matrix 116.

[0226] During weighting 106, the links 118 of the final relationship matrix 116 are weighted with the technology recognition model 38. The links 118 of the relationship matrix 116 are weighted. During weighting 106, each link 118 of the elements, benefits, and / or properties of the final relationship matrix 116 is weighted. The links 118 of the relationship matrix 116 are weighted at least based on the metadata 76 of the input sequence 22. The links 118 of the at least one implementation 58 to the product 32 of the input sequence 22 are weighted as respective elements in the final relationship matrix 116. The link 118 of the at least one implementation 58 to the sub-element 36 of the product 32 of the input sequence 22 is weighted in the final relationship matrix 116. The links 118 of the implementations 58 are weighted according to the advantages and / or properties of the product 32 of the input sequence 22.The links 118 of the implementations 58 are weighted based on the advantages and / or properties of the sub-element 36 of the product 32 of the input sequence 22. The links 118 are weighted based on the trend properties 56, the sequences 50 and relationship matrices 66 of the sequence database 48, the at least one further input sequence 28, the properties and / or advantages of the properties database 44 and / or the at least one checklist 72 for the technology area of ​​the product 32 of the input sequence 22. The trend properties 56, the sequences 50 and relationship matrices 66 of the sequence database 48, the at least one further input sequence 28, the properties and / or advantages of the properties database 44 are included as influencing factors in the weighting 106. Each influencing factor is included in the weighting 106 to a varying extent.

[0227] The selection of the elements of the final relationship matrix 116 for the portion of the collected metadata 34 results from the weighting 106 of the links 118 of the elements of the relationship matrix 116 to the respective sub-element 36 of the product 32 of the input sequence 22. The weighting 106 of the elements of the final relationship matrix 116 results from the weighting of the link 118 of the final relationship matrix 116. The weighting 106 of the elements of the final relationship matrix 116 results from the calculation of the weighting 106 of the link 106 of the respectively connected elements. The weighting 106 of the link 118 is calculated using the indexed elements of the final relationship matrix 118. The weighting of the link 118 to the sub-elements 36 of the product 32 of the input sequence 22 to the respective elements of the final relationship matrix 116 is assigned to each element.For each element of the final relationship matrix 116, as an implementation 58, property, and / or benefit of each sub-element 36 of the product 32 of the input sequence 22, a value is calculated as the strength of the weighting 106. The value as the strength of the weighting 106 determines, in relation to other values ​​of the weighting for other elements of the final relationship matrix 116 to the respective sub-element 36 of the product 32 of the input sequence 22, whether the element is part of the part of the collected metadata 34 for the generation 18 of the output sequence 30.

[0228] The elements of the final relationship matrix 116 for which the link 118 to the respective sub-element 36 of the product 32 of the input sequence 22 has the strongest weighting 106 relative to links 118 to other elements of the final relationship matrix 116 are determined as part of the collected metadata 34. The implementations 58 of the sub-elements 36 that have the strongest weighting 106 relative to other implementations 58 from the final relationship matrix 116 are determined as part of the collected metadata 34. The part of the collected metadata 34 to be determined depends on the minimum number of links 118 to each sub-element 36 of the product 32 of the input sequence 22.If a number of links 118 per sub-element 36 of the product 32 of the input sequence 22 in the final relationship matrix 116 is identical to the minimum number for the respective sub-element 36, all linked elements of the final relationship matrix 116 are determined as part of the collected metadata 34.

[0229] For weighting 106, the technology recognition model 38 calculates the weighting variables of the property database 44, the weighting variables of the sequence database 48, and the weighting variable of the at least one further input sequence 28. The calculated weighting variables are each multiplied by the weighting factors according to equation (1). The value for the strength of the weighting is calculated by adding the weighting variables multiplied by the respective weighting factors. The calculation of the respective weighting factors depends on the indexing of the origin of the respective elements of the final relationship matrix 116.

[0230] The weighting value of the property database 44 for links 118 of elements of the final relationship matrix 116 is calculated according to equation (2).

[0231] (2) A = a • r) • n A • T A

[0232] In equation (2), a is a factor for the occurrence of the element in the property database 44 according to the indexing of the origin. The factor a is a Boolean value. If an element of the final relationship matrix 116 has no indexing of the property database 44, the factor a is zero and the weighting variable A of the property database 44 is ineffective. In equation (2), / / is the given evaluation factor of the element from the property database 44. In equation (2), n is A a factor for the link level in the property database 44. The factor n A is lower for elements of the property database 44, the deeper the linking level of the element. In equation (2), T A a Kl ​​factor of the element of the property database 44, which is determined using the technology recognition model 38 as a machine learning system.

[0233] The weighting size of the sequence database 48 for links 118 of elements of the final relationship matrix 116 is calculated according to equation (3).

[0234] (3) B = b • A • r • n B • T B

[0235] In equation (3), b is a factor for the occurrence of the element in the sequence database 48 according to the indexing of the origin. The factor b is a Boolean value. If an element of the final relationship matrix 116 has no indexing of the sequence database 48, the factor b is zero and the weighting value B of the sequence database 48 is ineffective. In equation (3), A is a factor linked to the trend property 56, which determines the extent to which the element of the final relationship matrix 116 corresponds to at least one of the trend properties 56 determined by the trend detection 96. In equation (3), r is a factor determined from the evaluation of the relevance of the sequences 50 of the sequence database 48.

[0236] For example, the value r corresponding to the evaluation of the relevance of the sequences 50 is a value between zero and one. In equation (3), n Ba factor for the linkage level of the element in the sequences 50 of the sequence database 48. The factor n B is lower for elements of sequences 50 of the sequence database 48, the deeper the linking level of the element. In equation (3), T B a Kl ​​factor of the element of the sequence database 48, which is determined using the technology recognition model 38 as a machine learning system.

[0237] The weighting value of the at least one further input sequence 28 for links 118 of elements of the final relationship matrix 116 is calculated according to equation (4).

[0238] (4) C = c • A • n c • T c

[0239] In equation (4), c is a factor for the occurrence of the element in the at least one further input sequence 28 according to the indexing of the origin. The factor c is a Boolean value. If an element of the final relationship matrix 116 has no indexing of the at least one further input sequence 28, the factor c is zero and the weighting value C of the at least one further input sequence 28 is ineffective. In equation (4), A is a factor linked to the trend property 56, which determines the extent to which the element of the final relationship matrix 116 corresponds to at least one of the trend properties 56 determined by the trend detection 96. In equation (4), n is c a factor for the linking level of the element in the at least one further input sequence 28. The factor n cis lower for elements of the at least one further input sequence 28, the deeper the linking level of the element. In equation (4), T c a Kl ​​factor of the element of the at least one further input sequence 28, which is determined with the technology recognition model 38 as a machine learning system.

[0240] In at least one embodiment of the method 10 according to the invention, the user enters a difference between the input sequence 22 and the at least one further input sequence 28. The difference relates to various sub-elements 36, properties and / or advantages of the respective products 32 of the input sequence 22 and the at least one further input sequence 28. The difference between the input sequence 22 and the at least one further input sequence 28 forms, at least in part, the novel and patentable core of the input sequence 22. During weighting 106, elements of the final relationship matrix 116 that have links 118 to sub-elements 36 of the product 32 of the input sequence 22 that form the difference to the at least one further input sequence 28 are given a higher weighting.The weighting value C of the at least one further input sequence 28 has a further factor that defines whether respective elements in the final relationship matrix 116 are different from the at least one further input sequence 28 with an index according to the user's input of the difference. This allows the difference of the input sequence 22 from the current state of the art to be better represented.

[0241] The Kl factor is calculated using a Kl module of the technology recognition model 38. The Kl module is a machine learning module. The Kl module calculates the Kl factor based on a comparison of the element of the final relationship matrix 116 with one of the sub-elements 36 of the product 32 of the input sequence 22, to which the element is linked via the link 118. The Kl module calculates a similarity or relevance of the elements to the respective sub-elements 36 of the product 32 of the input sequence 22. The Kl module calculates the similarity or relevance of the properties, advantages, and / or implementations 58 of the final relationship matrix 116 of the respective sub-elements 36 of the product 32 of the input sequence 22.

[0242] In at least one embodiment of the method 10 according to the invention, the Kl module of the technology recognition model 38 is a neural network. The Kl module is a multi-input neural network that compares the sub-element 36 of the product 32 of the input sequence 22 with the respectively linked element of the final relationship matrix 116. The multi-input neural network is a Siamese network. For example, the known Siamese network described in the following publication can be used: W. Lei and Z. Meng, "Text similarity calculation method of Siamese network based on ALBERT," 2022 International Conference on Machine Learning and Knowledge Engineering (MLKE), Guilin, China, 2022, pp. 251-255, doi: 10.1109 / MLKE55170.2022.00055.

[0243] In at least one further embodiment of the method 10 according to the invention, the Kl module of the technology recognition model is a deep graph network. Using the deep graph network, the final relationship matrix 116 is mapped as a graph, with the links 118 forming the edges and the elements forming the nodes of the graph. The similarity or relevance between one of the subelements 36 of the product 32 of the input sequence 22 and the respective linked element is determined by comparing their graph embeddings. For example, the known deep graph network described in the following publication can be used: Mohebbi, M., Razavi, SN & Baiafar, MA Computing semantic similarity of texts based on deep graph learning with ability to use semantic role label information. Sci Rep 12, 14777 (2022). https: / / doi.org / 10.1038 / S41598-022-19259-5.

[0244] However, additional Kl modules of the technology recognition model 38 are also conceivable, which calculate the similarity or relevance of the elements of the final relationship matrix 116 to the respective subelements 36 of the product 32 of the input sequence 22. Alternatively, Naive Bayes classifiers, Support Vector Machines (SVM), Latent Dirichlet Allocation (LDA), or cross-encoders could be applied. For example, one of the known Kl modules described in the following publications can be used in the computer-implemented method 10: a) Z. Liu, X. Lv, K. Liu and S. Shi, "Study on SVM Compared with the other Text Classification Methods," 2010 Second International Workshop on Education Technology and Computer Science, Wuhan, China, 2010, pp. 219-222, doi: 10.1109 / ETCS.2010.248; b) Zhang, Haiyi and Di Li. “Naive Bayes Text Classifier.” 2007 IEEE International Conference on Granular Computing (GRC 2007) (2007): 708-708; c) J. Mazarura and A.de Waal, "A comparison of the performance of latent Dirichlet allocation and the Dirichlet multinomial mixture model on short text," 2016 Pattern Recognition Association of South Africa and Robotics and Mechatronics International Conference (PRASA-RobMech), Stellenbosch, South Africa, 2016, pp. 1 -6, doi: 10.1109 / RoboMech.2016.7813155.

[0245] In at least one further embodiment of the method 10 according to the invention, the weighting 106 of the elements of the final relationship matrix 116 is calculated solely based on the Kl module of the technology recognition model 38. The individual factors of the property database 44, sequence database 48, and the at least one further input sequence 28 each form input variables of the Kl module. The calculated Kl factor as an output variable forms the value for the strength of the weighting 106. In at least one further embodiment of the method 10 according to the invention, no Kl factor is calculated. The weighting 106 is calculated solely based on given parameters and defined rules of the collected metadata 76.

[0246] Using the calculated weighting variables, the technology recognition model 38 determines the portion of the collected metadata 34. The portion of the collected metadata 34 is determined based on the strength of the weighting 106 of the links 118 of the elements of the final relationship matrix 116. Based on the strength of the weighting 106 of the links 118 of the elements of the final relationship matrix 116, at least the implementation 58 of the product 32 of the input sequence 22 is determined as part of the collected metadata 34. Based on the weighting 106 of the links 118, at least the implementation 58 of the product 32 is integrated into the generation 18 of the output sequence 30. Based on the strength of the weighting 106 of the links 118 of the elements of the final relationship matrix 116, at least the implementation 58 of the at least one sub-element 36 of the product 32 of the input sequence 22 is determined as part of the collected metadata 34.Based on the weighting 106 of the links 118, at least the implementation 58 of at least one sub-element 36 is integrated during the generation 18 of the output sequence 30. The portion of the collected metadata 34 includes the implementations 58 of the product 32 of the input sequence 22. The portion of the collected metadata 34 includes the properties and / or advantages of the sub-elements 36 of the product 32 of the input sequence 22.

[0247] Issue 120 of the collected metadata part 34

[0248] After weighting 106 the final relationship matrix 116 and determining the portion of the collected metadata 34, an output 120 of the portion of the collected metadata 34 is executed. The output 120 of the portion of the collected metadata 34 is executed after processing step 86. The output 120 is executed with the technology recognition model 38 to generate 18 the output sequence 30 based on the portion of the collected metadata 34. The output 120 is executed with the technology recognition model 38 to the algorithm 12. The algorithm 12 passes the portion of the collected metadata 34 to the text generation model 38.

[0249] After processing step 86, the final relationship matrix 116 is stored in the sequence database 48. Upon repeated execution of the method 10 according to the invention, the final relationship matrix 116 forms a relationship matrix 66 of the relationship matrix acquisition 102.

[0250] Figure 5 shows a computer-implemented method 200 for continuously training the algorithm 12. The machine learning algorithm 12 is continuously trained. At least the input sequence 22 for continuously training the algorithm 12 is input. The input sequence 22 describes the product 32. The product 32 is at least partially novel. The input sequence 22 is processed. With the algorithm 12, metadata 76 for the product 32 is collected for continuously training the algorithm 12. The output sequence 30 is generated with at least a portion of the collected metadata 34 of the algorithm 12 for continuously training the algorithm 12. The output sequence 30 is at least partially text-based. The output sequence 30 is provided.The computer-implemented method 200 is at least substantially identical to the described computer-implemented method 10 up to the provision 20 of the output sequence 30. The computer-implemented method 200 comprises all method steps 14, 16, 18, 20 of the described computer-implemented method 10. The computer-implemented method 200 comprises the computer-implemented method 10. The input 14, processing 16 of the input sequence 22, and the generation 18 and provision 22 of the output sequence 30 of the computer-implemented method 200 are at least substantially identical to the computer-implemented method 10.

[0251] In at least one further method step of the computer-implemented method 200, an evaluation 202 of the output sequence 30 is performed. The output sequence 30 is output for the evaluation 202. The output sequence 30 is output to the user. The evaluation 202 of the output sequence 30 is at least partially performed by the user. The part of the collected metadata 34 described in the provided output sequence 30 is at least partially evaluated. The advantages, properties, and / or implementations 58 of the product 32 are at least partially evaluated as parts of the output sequence 30. The evaluation 202 is performed for each sub-element 36 of the product 32 and the associated metadata 34 in the provided output sequence 30. The evaluation 202 is enabled via evaluation parameters 206.During the evaluation 202, a negative or positive evaluation parameter 206 is assigned to the metadata 34 in the provided output sequence 30. The evaluation parameters 206 are Boolean values ​​that form the positive or negative evaluation parameters 206. Alternatively, a scale is provided to the user as the evaluation parameter 206 when evaluating the metadata 34. The scale is, for example, a value between 0 and 1. A value below 0.5 is a negative evaluation, and a value above 0.5 is a positive evaluation.

[0252] The evaluation parameters 206 define the quality of the parts of the output sequence 30. The evaluation parameters 206 define the extent to which the metadata 34 linked to the respective sub-elements 36 of the product 32 matches the respective sub-element 36. The evaluation parameters 206 define the extent to which metadata 34 for the respective sub-element 36 is technically correct, applicable, and combinable. The evaluation parameters 206 mark incorrect or inappropriate metadata 34 for the respective sub-elements 36 of the product 32 in the provided output sequence 30. The evaluation parameters 206 mark appropriate metadata 34 for the respective sub-elements 36 of the product 32 in the provided output sequence 30.

[0253] In at least one further method step, at least one adjustment 204 of parameters of algorithm 12 is performed. The parameters of algorithm 12 are adjusted based on evaluation 202. The parameters of algorithm 12 are adjusted using evaluation parameters 206 of evaluation 202. The parameters of technology recognition model 38 are adjusted based on evaluation 202.

[0254] During the parameter adjustment, weighting parameters of the Kl model of the technology recognition model 38 are adjusted. Alternatively, the evaluation parameters 206 are included as a further layer of the Kl model as a neural network as parameters of algorithm 12. Alternatively, the evaluation parameters 206 are included as a further weighting variable for calculating the weighting as parameters of algorithm 12. The evaluation parameters 206 are included as a further weighting variable in equation (1). The evaluation parameters 206 as a weighting variable are multiplied by the further weighting variables of the property database 44, the sequence database 48, and the at least one further input sequence 28. The parameters of algorithm 12 are continuously adjusted during each run of the computer-implemented method 200 by the evaluation 202 of the provided output sequence 30.The parameters of the algorithm 12 are further optimized by means of each re-evaluation 202 of each additional generated and provided output sequence 30.

[0255] In at least one embodiment of the method 200 according to the invention, metadata 76 collected with the technology recognition model 38 according to the Boolean value as evaluation parameter 206 is removed or determined as part of the collected metadata 34 upon repeated execution of the computer-implemented method 200. If a newly input and processed input sequence 22 with a further new and patentable product 32 at least partially comprises subelements 36 for which evaluation parameters 206 are present for the correspondingly collected metadata 76, the collected metadata 76 is directly removed or added to the part of the collected metadata 34 according to the Boolean values ​​as evaluation parameters 206.

[0256] With the ongoing continuous adjustment of the parameters of the algorithm 12, incorrect or inappropriate metadata 76 are removed by means of the evaluation 202 when determining the portion of the collected metadata 34 with the technology recognition model 38. With the ongoing continuous adjustment of the parameters of the algorithm 12, preferred metadata 76 are included when determining the portion of the collected metadata 34 with the technology recognition model 38.

[0257] In at least the embodiment of the method 10 according to the invention in which the output sequence 30 is provided to the user as a metadata construct in the processing module, the evaluation parameters 206 are continuously calculated in the computer-implemented method 200 based on the user input. A Boolean value is assigned to the metadata construct based on the user input. The Boolean value is used as the evaluation parameter 206 to adjust the algorithm parameters.

[0258] Alternatively and / or additionally, the parameters of the text generation model 140 of algorithm 12 are adjusted based on the evaluation 202. The parameters of the text generation model 140 are adjusted based on structural evaluations 202 of the output sequence 30.

[0259] The system 68 is provided for executing the computer-implemented method 200 for continuously training the algorithm 12. The system 68 is provided for inputting 14 and processing 16 the input sequence 22 and for generating 18 and providing 20 the output sequence 30 for continuously training the algorithm 12. The system 68 is provided for evaluating 202 the output sequence 30. The system 68 is provided for adapting 204 the parameters of the algorithm 12. The system 68 adapts the parameters of the algorithm 12 based on the evaluation 202 of the provided output sequence 30. The user interface 126 of the system 68 has an evaluation module for evaluating 202 the provided output sequence 30. The evaluation module is provided for inputting the evaluation parameters 206 for the evaluation 202 of the provided output sequence 30. The user enters the evaluation parameters 206.The evaluation module is provided for inputting the evaluation parameters 206 for the evaluation 202 of the portion of the collected metadata 34 of the provided output sequence 30. The evaluation module transfers the evaluation parameters 206 to the input means 130 of the computing device 70. The computing device 70 is provided for adapting 204 the algorithm 12 based on the evaluation parameters 206 of the evaluation 202.

[0260] In a further embodiment of the invention, the output sequence 30 is generated sequentially. When the output sequence 30 is generated sequentially, at least the method steps generation 18 and provision 20 are repeated sequentially until the output sequence 30 is completely generated. In a further embodiment of the invention, the input sequence 22 is processed sequentially. When the input sequence 22 is processed sequentially, at least the method steps processing 16, generation 18, and provision 20 are repeated sequentially until the input sequence 22 is completely processed and the output sequence 30 is completely generated.

[0261] Figure 6 shows a computer-implemented method 300 for training algorithm 12. The computer-implemented method 300 is used to train the machine learning algorithm 12. The computer-implemented method 300 for training the algorithm 12 is executed before the computer-implemented method 10 is applied to provide 20 the output sequence 30.

[0262] In one method step, an input 314 of at least one input sequence 322 is executed into the algorithm 12. The at least one input sequence 322 describes a product. The product of the at least one input input sequence 322 is known. The product of the at least one input input sequence 322 is at least partially accessible to the algorithm 12. The product of the at least one input input sequence 322 is at least partially part of the prior art. The at least one input sequence 322 is entered into the algorithm 12 in text-based, abstract form. The at least one input sequence 322 is entered in the form of claims of the product. The at least one input sequence 322 is entered from the sequence database 48. During input, a plurality of input sequences 322 of the sequence database 48 are entered into the algorithm 12 for training the algorithm 12.The plurality of input sequences 322 describes a plurality of different products. The plurality of input sequences 322 are from one of the classes 54 of the sequence database 48. The algorithm 12 is trained at least substantially separately for each class 54 of the sequence database 48.

[0263] The sequences 50 of the sequence database 48 have a validation part 330 for training the algorithm 12. Each input sequence 322 of the sequence database 48 has an associated validation part 330. The validation part 330 is a detailed description of the product associated with the input sequence 322. The validation part 330 is a description of the claims and / or a figure description of the product of the associated input sequence 322. The structure of the validation part 330 of the sequences 50 is at least approximately identical to a structure of the output sequence 30 of a trained algorithm 12 provided in the computer-implemented method 10. The validation part 330 of the sequences 50 has relevant metadata of the associated input sequence 322.The validation part 330 comprises advantages, properties, and / or implementations of the product and the subelements of the product of the inputted associated input sequence 322. The validation part 330 is input together with the associated input sequence 322. The input sequences 322 are divided into smaller subsets as training data. The subsets are input sequentially into algorithm 12.

[0264] In a further method step, processing 316 of the input sequence 322 is carried out. Metadata 76 is collected for the product using the algorithm 12. Metadata 76 is collected based on predefined parameters of the algorithm 12. The metadata 76 is collected at least partially from the properties database 44. The metadata 76 is collected at least partially from the sequence database 48. The metadata 76 is collected at least partially from the further sequence database 62. At least a portion of the collected metadata 34 is determined from the collected metadata 76. The algorithm 12 has the technology recognition model 38 for processing 316 the input sequence 322. Based on the predefined parameters, the input sequence 322 is processed using the technology recognition model 38. The technology recognition model 38 determines the portion of the collected metadata 34 for the respective input input sequence 322.A processing sequence 316 of the input sequence 322 with the technology recognition model 38 is at least substantially identical to a processing sequence 16 of the input sequence 22 of the computer-implemented method 10 with the trained algorithm 12. The portion of the collected metadata 34 is output for validation and adaptation 318 of the technology recognition model 38.

[0265] During processing 316, the validation part 330 associated with the input sequences 322 is additionally processed. The technology recognition model 38 captures metadata of the validation part 330 associated with the input sequences 322. The technology recognition model 38 captures properties, advantages, and / or implementations of the product described in the validation part 330, described in the validation part 330, for respective sub-elements of the product as metadata of the validation part 330.

[0266] In a further method step, the adaptation 318 of algorithm 12 is carried out. The parameters of algorithm 12 are adapted. The parameters are adapted based on at least a portion of the collected metadata 34. The parameters are adapted based on the specific portion of the collected metadata 34. The validation part 330 is used to validate the portion of the collected metadata 34 of the product. Based on the validation of the portion of the collected metadata 34 of the associated input sequence 322, the parameters of algorithm 12 are adapted using the respective validation part 330 of the sequences 50.

[0267] The technology recognition model 38 of algorithm 12 is trained. The K1 model for calculating the K1 factor of the technology recognition model 38 is trained. The parameters of the technology recognition model 38 of algorithm 12 are adjusted. The parameters of the technology recognition model 38 are adjusted by validating and determining the portion of the collected metadata 34 of the technology recognition model 38 relating to the product of the input sequence 322. The validation is performed by calculating a difference between the portion of the collected metadata 34 of the technology recognition model 38 relating to the input sequence 322 and the validation portion 330 associated with the input sequence 322. The difference between the properties, advantages, and / or implementations of the portion of the collected metadata 34 and the properties, advantages, and / or implementations of the validation portion 330 is calculated.

[0268] During parameter adjustment 318, a loss function is calculated for the portion of collected metadata 34 of algorithm 12. The loss function specifies the difference between the portion of collected metadata 34 and the metadata of the validation portion 330 associated with the respective input sequence 322. The loss function determines the quality of the portion of collected metadata 34 in comparison to the respective associated validation portion 330.

[0269] During parameter adjustment 318, a gradient of a loss is calculated with respect to each parameter of the technology detection model 38. The gradient of the loss is calculated using the derivative of the loss function.

[0270] During parameter adjustment 318, an optimization algorithm is applied to adjust the parameters of the technology detection model 38 based on the loss gradients. Examples of optimization algorithms are Stochastic Gradient Descent, Adam, or RMSprop.

[0271] The steps of inputting 314, processing 316, and adapting 318 for training algorithm 12 are repeated cyclically. Steps 314, 316, 318 are repeated until a predetermined number of cycles is reached or an adjustment of the parameters of algorithm 12 reaches a local or global optimum. Steps 314, 316, 318 are repeated until, at least approximately, no improvement in performance is recorded on the input sequences 322 as a training data set. Based on the adjustment 318 of the parameters of the technology recognition model 38, a different portion of collected metadata 34 is determined from the collected metadata 76 during a further processing 316 of the input sequences 322.When improving the performance on the input sequences 322 as a training data set, the specific part of the collected metadata 34 approaches at least informationally the validation part 330 for the respective associated input sequences 322.

[0272] In a further embodiment of method 300, the text generation model 140 is trained during adaptation 318 of the parameters of algorithm 12. To train text generation model 140, an output sequence is generated using the portion of collected metadata 34 with text generation model 38. Based on a comparison of the formulation and / or structure of the validation portion 330 associated with the respective input sequences 322, the parameters of text generation model 140 are adapted using the output sequence.

[0273] The system 68 is provided to execute the computer-implemented method 300 for training the algorithm 12. The algorithm 12 is trained by the system 68 prior to applying the machine learning algorithm 12 with the system 68. The system 68 is provided to train the algorithm 12 to the input 314 of the input sequence 322. The input sequence 322 is input from the sequence database 48 into the computing device 70. The system 68 is provided to train the algorithm 12 to process 316 the input sequence 322. During processing 316, the metadata 76 from the property database 44, sequence database 48, and / or the further sequence database 62 of the system 68 is collected with the algorithm 12. The system 68 is provided to adapt 318 parameters of the algorithm 12. The algorithm 12 is trained with the computing device 70.Alternatively, the computer-implemented method 300 is trained on another system having a connection to a property database, sequence database, and / or another sequence database.

[0274] Reference symbol

[0275] 10 Computer-implemented method

[0276] 12 Algorithm

[0277] 14 Input

[0278] 16 Processing

[0279] 18 Generation

[0280] 20 Provision

[0281] 22 Input sequence

[0282] 24 Concrete introductory sequence

[0283] 26 Graphic input sequence

[0284] 28 Further entrance sequence

[0285] 30 Output sequence

[0286] 32 Product

[0287] 34 metadata

[0288] 36 sub-elements

[0289] 38 Technology recognition model

[0290] 40 metadata

[0291] 42 Relationship matrix

[0292] 44 Property database

[0293] 46 metadata

[0294] 48 sequence database

[0295] 50 sequence

[0296] 52 metadata

[0297] 54 Class

[0298] 56 Trend property

[0299] 58 Implementation

[0300] 60 metadata

[0301] 62 sequence database

[0302] 64 sequence

[0303] 66 Further relationship matrix

[0304] 68 System

[0305] 70 computing device

[0306] 72 Checklist

[0307] 74 Property entry

[0308] 76 metadata

[0309] 78 Relationship matrix

[0310] 80 processing steps

[0311] 82 Processing step Processing step

[0312] Processing step

[0313] graph

[0314] Database processing

[0315] Database processing

[0316] Relationship matrix

[0317] Trend detection

[0318] Sequence analysis

[0319] Sequence analysis

[0320] Relationship matrix recording

[0321] Merge

[0322] Weighting

[0323] Data query

[0324] Relationship matrix

[0325] Data set generation

[0326] Merger

[0327] Relationship matrix

[0328] Link

[0329] output

[0330] handover

[0331] Complete database

[0332] User interface

[0333] storage unit

[0334] Input devices

[0335] Output medium

[0336] Text generation model

[0337] Computer-implemented procedure

[0338] Evaluation

[0339] Adjustment

[0340] Evaluation parameters

[0341] Computer-implemented procedure

[0342] input

[0343] processing

[0344] Adjustment

[0345] Entrance sequence

[0346] Validation part

Claims

Claims 1 . Computer-implemented method (10) with an algorithm (12), in particular machine learning, comprising the following steps: Input (14) of at least one input sequence (22) into the algorithm (12) which describes a product (32), system and / or method which is novel, in particular at least for the algorithm (12), and preferably patentable; - processing (16) the input sequence (22) by collecting metadata (76) about the product (32), system and / or method using the algorithm (12); - generation (18) of an output sequence (30) based on at least part of the collected metadata (34), in particular at least partially text-based, by means of the algorithm (12); and - Providing (20) the output sequence (30).

2. Computer-implemented method (10) according to claim 1, characterized in that the input sequence (22) describes at least one sub-element (36) and / or method step and / or a combination of at least two sub-elements (36) of the product (32), system and / or method.

3. Computer-implemented method (10) according to claim 1 or 2, characterized in that the algorithm (12) has at least one technology recognition model (38) with which the product (32), system and / or method, in particular the at least one sub-element (36) and / or the method step, is detected.

4. Computer-implemented method (10) according to claim 2 and 3, characterized in that the at least one sub-element (36) is evaluated with the technology recognition model (38) on the basis of structural features within the input sequence (22).

5. Computer-implemented method (10) according to one of the preceding claims, in particular according to claim 2, characterized in that the product (32), method and / or system, in particular the at least one sub-element (36), is described in abstract form in the input sequence (22).

6. Computer-implemented method (10) according to one of the preceding claims, characterized in that at least one concrete input sequence (24) is input which describes the product (32), system and / or method in concrete form.

7. Computer-implemented method (10) according to claim 2, characterized in that at least one graphic input sequence (26) is input which graphically represents the product (32), the system and / or the method and / or the at least one sub-element (36) of the product (32), system and / or method, in particular in the form of a technical drawing.

8. Computer-implemented method (10) according to claim 7, characterized in that the input sequence (22) is linked to the graphical input sequence (26).

9. Computer-implemented method (10) according to one of the preceding claims, characterized in that at least one further input sequence (28) is input, which describes a further, in particular obvious, product, system and / or method.

10. Computer-implemented method (10) according to one of the preceding claims, characterized in that, in particular with the algorithm, at least one, in particular private, property database (44) is accessed, which stores metadata (46), in particular information on, preferably technical, properties and / or advantages, for at least one element.

11. Computer-implemented method (10) according to one of the preceding claims, characterized in that, in particular with the algorithm, at least one, in particular private, sequence database (48) is accessed, with a plurality of, in particular text-based, sequences (50), which each describe a further product, system and / or method, in particular with metadata (52) for the further product, system and / or method.

12. Computer-implemented method (10) according to claim 11, characterized in that the sequences (50) in the sequence database (48) are divided into classes (54) and the input sequence (22) is assigned to one of the classes (54) on the basis of metadata of the input sequence (22).

13. Computer-implemented method (10) according to claim 12, characterized in that the metadata (52) of the product (32), system and / or method are at least partially taken from the assigned class (54) of the sequence database (48).

14. Computer-implemented method (10) according to claim 12, characterized in that trend properties (56) are stored in the sequence database (48) for the respective classes (54), which are transferred to the technology recognition model (38) for processing (16) of the input sequence (22) with the algorithm (12).

15. Computer-implemented method (10) according to claim 3, characterized in that with the technology recognition model (38) at least one implementation (58) of the product (32), system and / or method and / or of the at least one sub-element (36) is derived based on the abstract form of the product (32), system and / or method with the metadata (76).

16. Computer-implemented method (10) according to claim 2 and 3, characterized in that the technology recognition model (38) links the metadata (76) with the sub-elements (36) of the product (32), system and / or method and the implementation (58) is derived by means of property relationships and / or advantage relationships of the metadata (76).

17. Computer-implemented method (10) according to claim 16 and in particular claim 11, characterized in that with the technology recognition model (38) on the basis of the property relationship and / or advantage relationships of the metadata (76) of the sub-elements (36) a relationship matrix (116) is generated, which creates a link between at least the sub-elements (36) of the product (32), system and / or method, in particular the sub-elements of the sequences (50) of the sequence database (48).

18. Computer-implemented method (10) according to claim 17, characterized in that the links (118) of the relationship matrix (116) are weighted at least on the basis of the metadata (76) of the input sequence (22).

19. Computer-implemented method (10) according to claim 18, characterized in that based on the weighting (106) of the links (118), at least the implementation (58) of the product (32), system and / or method, in particular of the at least one sub-element (36), is integrated in the generation of the output sequence (30).

20. Computer-implemented method (10) according to one of the preceding claims, characterized in that, in particular with the algorithm (12), at least one further, in particular public, sequence database (62), preferably a patent database, is accessed, which has a plurality of sequences (64).

21. Computer-implemented method (10) according to claim 17 and 20, characterized in that the algorithm (12) generates a partial element data set on the basis of the product (32), system and / or method, by means of which a subset of the plurality of sequences (64) is transferred into the, in particular private, sequence database (48) for generating a further relationship matrix (66).

22. Computer-implemented method (300) for training a machine learning algorithm (12), in particular according to one of claims 1 to 21, comprising the following steps: - inputting (314) at least one input sequence (322) into the algorithm (12) which describes a product, system and / or method; - processing (316) the input sequence (322) by collecting metadata (76) relating to the product, system and / or method using the algorithm (12); and - adapting parameters of the algorithm (12) based on at least part of the collected metadata (34).

23. Computer-implemented (200) method for continuously training a machine learning algorithm (12), in particular according to one of claims 1 to 21, comprising the following steps: Input (14) of at least one input sequence (22) into the algorithm (12) which describes a product (32), system and / or method which is novel, in particular at least for the algorithm (12), and preferably patentable; - processing (16) the input sequence (22) by collecting metadata (76) about the product (32), system and / or method using the algorithm (12); - generating (18) an output sequence (30) based on at least part of the collected metadata (34), in particular at least partially text-based, by means of the algorithm (12); - providing (20) the output sequence (30); - evaluation (202) of the output sequence (30); and - Adjustment (204) of the parameters of the algorithm (12) based on the evaluation.

24. System (68) for inputting (14) and processing (16) an input sequence (22) and for generating (18) and providing (20) an output sequence (30), in particular an at least partially text-based output sequence, and / or for teaching a machine learning algorithm (12) for inputting (14) and processing (16) an input sequence (22) and for adapting the parameters of the algorithm (12) and / or for continuously teaching a machine learning algorithm (12) for inputting (14) and processing (16) an input sequence (22) and for generating (18), providing (20) and evaluating an output sequence (30), in particular an at least partially text-based output sequence, and for adapting the parameters of the algorithm, comprising: a computing device (70) which is configured to carry out at least one of the methods according to one of claims 1 to 21, 22 and / or 23.

25. A computer program with program code, comprising instructions which, when the program code is executed by a computer, cause the computer to carry out at least one of the methods according to one of claims 1 to 21, 22 and / or 23.

26. Computer-readable storage medium on which the computer program according to claim 25 is stored

Images