Information processing method and device
By comparing the expected output information with the actual output information, identifying the missing information and using it to optimize the training data, the problem of low generation quality of machine learning models when there is insufficient data or lack of private information is solved, and the application and output of the model are improved quality.
Patent Information
- Application Number
- CN202410899627.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-05
- Publication Date
- 2024-11-26
AI Technical Summary
In the case of insufficient training data or lack of private information, existing machine learning models are difficult to generate results that users expect, resulting in low quality of model generation content, increasing subsequent workload and reducing work efficiency.
By comparing the expected output information with the actual output information, it is determined that the information contained in the expected output information but not contained in the actual output information is used as the second input information, and is used to optimize the training data of the machine learning model, thereby simulating private information or specific application information. Improve the training effect of the model.
By using information contained in the expected output information but not included in the actual output information to optimize the training data, the quality and application of the machine learning model are improved, making the content generated by the model more suitable for specific application scenarios, and improving the applicability of the model and output quality.
Smart Images

Figure CN119025910A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to an information processing method and device. Background Art
[0002] The training data for machine learning models often comes from public data, shared data, or publicly known data. For application scenarios that require the combination of supplementary information such as private information or specific application information to generate the final result, it is difficult to generate the expected results using machine learning models trained only with public data, shared data, or publicly known data.
[0003] However, due to privacy restrictions, the amount of supplementary information such as private information or specific application information that can be used to train machine learning models is often limited, making it difficult to reach the data level required for training the model, which greatly limits the training effect of the model and makes the content generated by the model differ greatly from the content expected by the user. The quality of the content generated by the model directly affects the subsequent workload and work efficiency of the operators. If the quality of the content generated by the model is high, the subsequent workload and work efficiency can be greatly reduced; if the quality of the content generated by the model is low, it will greatly increase the subsequent workload and work efficiency. Summary of the invention
[0004] The present application provides an information processing method and device.
[0005] According to a first aspect of an embodiment of the present application, there is provided an information processing method, including: obtaining first training data, the first training data including first input information and expected output information corresponding to the first input information; inputting the first input information into a machine learning model, performing model operations, and determining actual output information corresponding to the first input information; comparing the expected output information with the actual output information to determine second input information, the second input information including information included in the expected output information but not included in the actual output information; and determining second training data based on the first input information, the second input information, the actual output information, and the expected output information.
[0006] According to an embodiment of the present application, the expected output information and the actual output information are compared to determine the second input information, including: determining sub-information from the expected output information; judging whether the sub-information is included in the actual output information, and if not, determining the corresponding sub-information as the target sub-information; and determining the second input information according to the target sub-information.
[0007] According to an embodiment of the present application, determining sub-information from expected output information includes: dividing the expected output information into sentences, and determining each sentence included in the expected output information as sub-information.
[0008] According to an embodiment of the present application, after determining the sub-information from the expected output information, the process further includes: determining whether the sub-information complies with a predefined rule.
[0009] According to an embodiment of the present application, determining the second training data according to the first input information, the second input information, the actual output information and the expected output information includes: determining the second training data according to the first input information, the second input information and the expected output information.
[0010] According to an embodiment of the present application, determining the second training data according to the first input information, the second input information, the actual output information and the expected output information includes: determining the second training data according to the second input information, the actual output information and the expected output information.
[0011] According to one embodiment of the present application, a machine learning model is used to generate information of a specification in a patent application document, the first input information includes information representing a claim item, an abstract or a technical solution, and the expected output information or the actual output information includes specification information corresponding to the first input information.
[0012] According to a second aspect of an embodiment of the present application, there is provided an information processing device, comprising: a first training data acquisition module, used to acquire first training data, the first training data comprising first input information and expected output information corresponding to the first input information; an actual output information determination module, used to input the first input information into a machine learning model, perform model operations, and determine the actual output information corresponding to the first input information; a second input information determination module, used to compare the expected output information with the actual output information, and determine the second input information, the second input information comprising information included in the expected output information but not included in the actual output information; and a second training data determination module, used to determine the second training data based on the first input information, the second input information, the actual output information, and the expected output information.
[0013] According to a third aspect of an embodiment of the present application, there is provided an information processing method, comprising: obtaining input information; inputting the input information into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model comprises a machine learning model trained with third training data, and the third training data comprises second training data determined using the above method.
[0014] According to a fourth aspect of an embodiment of the present application, an information processing device is provided, including: an input information acquisition module, used to acquire input information; an output information determination module, used to input the input information into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model includes a machine learning model trained with third training data, and the third training data includes second training data determined using the above method.
[0015] According to a fifth aspect of an embodiment of the present application, a computer storage medium is provided, the storage medium comprising a set of computer executable instructions, which are used to execute any of the above-mentioned data forwarding methods when the instructions are executed.
[0016] The embodiment of the present application provides an information processing method and device. The method includes: obtaining first training data, the first training data includes first input information and expected output information corresponding to the first input information; inputting the first input information into a machine learning model, performing model operation, and determining the actual output information corresponding to the first input information; comparing the expected output information and the actual output information, and determining the second input information, the second input information includes information included in the expected output information but not included in the actual output information; determining the second training data according to the first input information, the second input information, the actual output information, and the expected output information. In this way, the second input information included in the expected output information but not included in the actual output information can be used to simulate supplementary information such as private information or specific application information, thereby obtaining second training data that can be used to train a machine learning model to receive supplementary information such as private information or specific application information and output the expected result. By using training data with private information input to optimize the model for training, a model output that is more suitable for a specific application scenario can be obtained, thereby greatly improving the applicability of the model and improving the quality of the content generated by the model.
[0017] It should be understood that the implementation of the present application does not require the realization of all the above beneficial effects, but a specific technical solution can achieve a specific technical effect, and other embodiments of the present application can also achieve beneficial effects not mentioned above. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become readily understood. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, wherein:
[0019] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.
[0020] Figure 1 A schematic diagram of a processing flow for implementing the information processing method of the present application in one embodiment of the present application;
[0021] Figure 2 A schematic diagram of a processing flow of implementing the information processing method of the present application in another embodiment of the present application;
[0022] Figure 3 A schematic diagram of a processing flow of implementing the information processing method of the present application in another embodiment of the present application;
[0023] Figure 4 A schematic diagram of a model framework used by another embodiment of the present application to implement the information processing method of the present application;
[0024] Figure 5 This is a schematic diagram of the structure of the information processing device of the present application embodiment;
[0025] Figure 6 A schematic diagram of a processing flow of another information processing method of the present application for implementing an embodiment of the present application;
[0026] Figure 7 This is a schematic diagram of the composition structure of another information processing device of the present application. DETAILED DESCRIPTION
[0027] In order to make the purpose, features, and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.
[0028] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0029] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0030] The information processing method of this application is mainly used to generate partial training data for a machine learning model. The machine learning model can be a traditional machine learning model or an artificial intelligence large model, such as Chat GPT, CPM-Bee large model, and Llama large model, or any computer model based on machine learning.
[0031] Training data is the initial data set used to train machine learning algorithms. Training data is also called training set or learning set. It is a set of samples used to fit the parameters of a machine learning model and correctly predict samples that do not appear in the training set.
[0032] In a broad sense, training data can be divided into two categories: labeled data and unlabeled data. The training data mentioned in this application mainly refers to labeled data, that is, a set of data samples annotated with one or more labels. Among them, the label marks the specific characteristics, attributes, categories and objects contained in the data. Labeled data is mainly used for supervised learning, which enables machine models to learn features associated with specific labels, which can be used to classify new data points.
[0033] Figure 1 The main process of implementing the information processing method of the present application in one embodiment of the present application is shown.
[0034] refer to Figure 1 , the method comprising:
[0035] Operation S110: acquiring first training data, where the first training data includes first input information and expected output information corresponding to the first input information.
[0036] The first training data refers to the labeled data intended to be used to train the machine learning model.
[0037] The first input information is one of the input information in a sample data in the first training data. The type and format of the first input information are not limited, and can be information of any data type such as a string type, a digital type, a graphic or a multimedia, or can be computer coded information obtained by encoding data of various digital types, or can be information of any other format recognizable by a computer.
[0038] The expected output information corresponding to the first input information mainly refers to the correct output information expected to be obtained after the first input information is input into the machine learning model and the model is operated, that is, the label or annotation of the sample data where the first input information is located. Similar to the input information, the type and format of the expected output information are not limited.
[0039] Operation S120: input the first input information into the machine learning model, perform model operation, and determine actual output information corresponding to the first input information.
[0040] Among them, different from the expected output information, the actual output information corresponding to the first input information is the output information actually obtained after the first input information is input into the machine learning model and the model is operated. Usually, for the convenience of comparison, the data type and format of the actual output information and the expected output information are the same. The value of the actual output information may be the same as or different from the value of the expected output information. The smaller the difference between the value of the actual output information and the value of the expected output information, the higher the accuracy and the better the effect of the machine learning model.
[0041] The machine learning model in operation S120 mainly refers to a machine learning model that has been trained with a certain amount of public data, shared data or publicly known data and has reached a certain accuracy through multiple iterations. However, due to the lack of participation of supplementary information such as private information or specific application information during the training process, the actual output information of the machine learning model is still very different from the expected output information, and the ideal model effect cannot be achieved.
[0042] Operation S130: compare the expected output information with the actual output information to determine second input information, where the second input information includes information included in the expected output information but not included in the actual output information.
[0043] Comparing the expected output information with the actual output information mainly refers to a computer operation of comparing the expected output information with the actual output information to find the same information and different information between the two. Comparing the expected output information with the actual output information can be achieved using a third-party tool or by self-programming.
[0044] Information included in the expected output information but not included in the actual output information generally refers to information with the same or close values that exists in the expected output information but does not exist in the expected output information.
[0045] Through continuous practice and in-depth research, the inventors of the present application have discovered that due to the lack of supplementary information such as private information or specific application information, even after multiple iterations and various optimizations such as parameter tuning, the actual output information output by the machine learning model used in operation S120 always has certain differences from the expected output information. These differences, especially those information included in the expected output information but not in the actual output information, are often precisely the supplementary information such as private information or specific application information that is lacking in the training process.
[0046] Therefore, the inventor of the present application broke with conventional thinking and did not seek more data with supplementary information such as private information or specific information of actual usage scenarios generated in an actual production environment to generate subsequent training data. Instead, he went the other way and creatively thought of: by comparing the expected output information with the actual output information, the information included in the expected output information but not included in the actual output information can be obtained, and the information included in the expected output information but not included in the actual output information can be used to simulate supplementary information such as private information or specific application information to obtain second input information, and the second input information can be involved in the subsequent training of the machine learning model.
[0047] Operation S140: determining second training data according to the first input information, the second input information, the actual output information, and the expected output information.
[0048] Among them, the second training data mainly refers to the subsequent training data used for further optimization training of the machine learning model.
[0049] In some embodiments, the subsequent training data may still include the first input information, and the expected output information corresponding to the first input information may still be used as the labeled expected output information.
[0050] Different from the first training data, the second training data will add second input information for simulating supplementary information such as private information or specific application information to improve the accuracy and effect of the model.
[0051] Among them, the second input information can be input as part of the second training data, or can be used as a plug-in or plug-in data that can be used during the model operation process, and participate in the subsequent training process of the machine learning model in various ways.
[0052] In an embodiment of the present application, first, first input information and expected output information corresponding to the first input information are obtained through operation S110; then, actual output information is obtained through operation S120; next, the expected output information and the actual output information are compared through operation S130, and information included in the expected output information but not included in the actual output information is determined as second input information; then, through operation S140, the second input information is used to simulate supplementary information such as private information or specific application information that the machine learning model lacks in the previous training process, and participates in the subsequent training of the machine learning model.
[0053] In this way, we can make up for the lack of private information in the previous training process, further narrow the gap between the expected output information and the actual output information, and greatly improve the model effect of the machine learning model; we can also overcome a series of problems such as the difficulty in obtaining private information produced in the actual production environment, or even if it is obtained, it cannot reach the ideal data level, and the cost of manual labeling is high.
[0054] It should be added that if the training data that has participated in the previous training of the machine learning model is used and then input into the machine learning model, it can often match the previously learned knowledge, and make the actual output information very close to the expected output information, so that the second input information is empty or less, and the meaning of supplementary information such as simulated private information or specific application information is lost. Therefore, it is recommended to obtain training data that has not participated in the previous training of the machine learning model as the first training data in operation S110. The second input information obtained in this way can better represent supplementary information such as simulated private information or specific application information, and is more efficient and effective.
[0055] Figure 2 The figure shows the processing flow of another embodiment of the present invention to implement the information processing method of the present invention. Figure 2 In this embodiment, the method includes:
[0056] In operation S210, first training data is acquired, where the first training data includes first input information and expected output information corresponding to the first input information.
[0057] Operation S220: input the first input information into the machine learning model, perform model operation, and determine actual output information corresponding to the first input information.
[0058] Operation S230: determining sub-information from the expected output information to obtain a sub-information set.
[0059] Wherein, sub-information refers to part of the expected output information. For example, if the expected output information is a piece of text information, the sub-information may be a sentence included in the piece of text information; if the expected output information is a data set, the sub-information may be a subset of the data set.
[0060] The sub-information can be independent of each other or may have some overlap.
[0061] Operation S240: acquiring a sub-information from the sub-information set.
[0062] Operation S250, determining whether the sub-information is included in the actual output information, if not, proceeding to operation S260, if yes, returning to operation S240.
[0063] The “include” mentioned therein, for example, mainly refers to the existence of information in the actual output information that is semantically similar to the sub-information (e.g., a sentence) for text information; and mainly refers to the fact that the actual output information is a superset of the sub-information (e.g., a subset of the data set) for a data set.
[0064] Operation S260: determine the sub-information as target sub-information and add it to the target sub-information subset.
[0065] Operation S270: determining second input information according to the target sub-information in the target sub-information subset.
[0066] Exemplarily, for text information, the target sub-information in the target sub-information subset can be combined to form a new text information as the second input information; for data sets, the target sub-information in the target sub-information subset can be merged to obtain a new data set as the second input information.
[0067] Operation S280: determining second training data according to the first input information, the second input information, and the expected output information.
[0068] In this embodiment, by determining the sub-information in the expected output information, the target sub-information included in the expected output information but not included in the actual output information is found one by one, and then the target sub-information is merged or connected to obtain the second input information. This implementation method can simplify the complex and break the whole into parts, greatly simplify the calculation difficulty, and is simple and easy to implement.
[0069] In this embodiment, the first input information and the second input information are used as inputs of the machine learning model, and the original expected output information is retained as the output information annotated for the first input information and the second input information to form the second training data. In this way, in actual applications, the machine learning model trained with the second training data can not only continue to receive the first input information, but also receive supplementary information corresponding to the first input information, so as to further improve the output effect of the machine learning model in combination with the supplementary information, so as to make the output information richer and more suitable for the user's specific application scenario.
[0070] In addition, in other embodiments of the present application, corresponding rules can be pre-defined according to the needs of the actual application scenario, and in operation S240, after obtaining a sub-information from the sub-information set, it is determined whether the sub-information meets the pre-defined rules to determine whether to perform subsequent processing, thereby realizing the screening of sub-information and only processing the sub-information that is most likely to become the target to save computing power.
[0071] Figure 3 Another embodiment of the present application shows the specific implementation steps of applying the information processing method of the present application to the application scenario of patent application document generation.
[0072] In this embodiment, the machine learning model is a large language generation model, which is used to generate information about the description in the patent application document; the first input information is the text information representing the claim item obtained from the disclosed relevant documents; the expected output information is the text information of the description corresponding to the claim item obtained from the disclosed relevant documents, and is used as the output of the claim item annotation, and is jointly constructed with the claim text information as the first training data.
[0073] In this embodiment, the inventor of the present application found that the large language generation model trained with some existing public related documents can generate the corresponding specification text information according to the claim text information input by the user. However, compared with the annotated specification text information, that is, the expected output information, the specification text information always lacks some technical details of the technical solution in the specific implementation process, such as "entering the kernel through the syscall instruction, the entry address is the KiSystemCall64 function" and other information. The above information is often obtained through technical briefings or technical communication with patent agents, and is not included in the claim text information. For the large language generation model, it is a blank knowledge point that cannot be obtained through learning.
[0074] To this end, the inventor of this application has come up with an improvement measure, which is to provide the briefing provided by the technician or the communication record with the patent agent as supplementary information to the big language generation model for training. In this way, in actual application, the big language generation can receive the corresponding supplementary information provided by the user and generate the specification content together with the text information of the claim items, thereby filling the above-mentioned blank knowledge points.
[0075] However, this creates another problem: briefing documents or communication records with patent agents are generally private information of patent applicants and are not disclosed to the public. Therefore, it is difficult to obtain briefing documents or communication records with patent agents corresponding to published patent documents to construct training data for large language models.
[0076] Therefore, the inventor of the present application Figure 3 The following steps shown in the figure use the generated second input information to simulate the key information in the briefing document or the communication record with the patent agent to construct subsequent training data for optimizing the large language generation model, namely, the second training data:
[0077] Step S3010, obtaining claim text information in the first training data and specification text information annotated with the claim text information.
[0078] Assume that the text information of the claim item is: "A control method, the method comprising: in response to an application calling a system kernel, determining that a far jump performance monitoring condition is satisfied;..."
[0079] The specification text information annotated with respect to the text information of the claim item is: "This embodiment of the application provides a control method, such as Figure 1 As shown, the method includes: Step S110, in response to the application calling the system kernel, determining that the far jump performance monitoring condition is met; In the implementation process, when the application calls the application programming interface (Application Programming Interface, API), the system call instruction (syscall) can be used to enter the system kernel layer from the system application layer. Here, starting from Intel's 6th generation Core processor, the performance monitoring unit has added a Unit Mask for recording far jumps, in which the syscall instruction belongs to far jumps. In the implementation process, the performance monitoring function can be turned on first to meet the far jump performance monitoring condition. ..."
[0080] In this embodiment, the above information is obtained by reading the "input" and "output" fields from the training data in Json format using a python program.
[0081] In actual applications, implementers can obtain it through any applicable programs, tools or other technical means.
[0082] Step S3020, input the claim text information into the large language generation model to obtain the actual output specification text information.
[0083] Assume that in this embodiment, the instruction text information output by the large language generation model is: "The embodiment of the present application provides a control method, such as Figure 1 As shown, it is a flow chart of an application control method provided by an embodiment of the present application, and the method includes the following steps:\nS101, in response to the application calling the system kernel, determining that the far jump performance monitoring condition is met. \nThe application may include but is not limited to the main application, the low-code platform and other applications. "
[0084] Step S3030: divide the annotated instruction manual text information into sentences to obtain a sentence set.
[0085] Among them, a sentence is a broader sentence, representing a language unit with certain semantics, including words, short sentences, clauses, etc.
[0086] In the embodiment of the present application, sentence division is achieved by segmenting the specification text information through punctuation marks, such as ";", ".", ", ", etc.
[0087] In addition, in other embodiments of the present application, words or short sentences with certain semantics can be extracted from the instruction manual text information by means of semantic recognition.
[0088] Step S3040, obtaining a statement from the statement set.
[0089] For example, "the system kernel layer can be entered from the system application layer through the system call instruction (syscall)".
[0090] Operation S3050: determining whether the sentence is a common cliché in the relevant file.
[0091] In the annotated specification, there are often common clichés in related documents, such as "embodiments of the present application", "in the implementation process", "in some embodiments", "wherein", etc. These clichés are often irrelevant to the technical solution itself and have no specific meaning.
[0092] In the embodiment of the present application, it will be ignored and no further processing will be performed to avoid introducing interference information or wasting computing power.
[0093] “The system kernel layer can be entered from the system application layer through the system call instruction (syscall)” is a common phrase that does not exist in the relevant files, so the step S3060 can be continued.
[0094] Step S3060, determine whether the statement is included in the manual information actually output, if not, continue to step S3070, if so, return to step S3040.
[0095] Step S3070: determine the sentence as a target sentence and add it to the target sentence subset.
[0096] Step S3080, determine whether all statements have been processed, if so, continue to step S3090, if not, return to step S3040.
[0097] Step S3090, connecting the sentences in the target sentence subset to form second input information.
[0098] Assume that the target sentence subset obtained is:
[0099] In response to an application calling the system kernel
[0100] Determine if far jump performance monitoring conditions are met
[0101] When an application calls the Application Programming Interface (API)
[0102] The system kernel layer can be entered from the system application layer through the system call instruction (syscall).
[0103] Starting with Intel 6th Generation Core Processors
[0104] The performance monitoring unit has added a Unit Mask for recording far jumps
[0105] Among them, the syscall instruction is a far jump.
[0106] You can enable the performance monitoring function first
[0107] To meet the far jump performance monitoring conditions.
[0108] Then you can use a space "" to connect the above clauses to get the following second input information:
[0109] In response to the application calling the system kernel to determine that the far jump performance monitoring condition is met, the application can enter the system kernel layer from the system application layer through the system call instruction (syscall) when calling the application programming interface (Application Programming Interface API). Starting from the 6th generation Intel Core processor, the performance monitoring unit has added a unit mask for recording far jumps, of which the syscall instruction belongs to far jumps. The performance monitoring function can be turned on first to meet the far jump performance monitoring condition.
[0110] Step S3100, determining subsequent training data based on the actually output specification text information, the second input information, and the specification text information annotated with respect to the claim text information.
[0111] and Figure 2 The embodiments shown are different. Figure 3 The embodiment shown does not determine the second training data based on the first input information, the second input information and the expected output information, but determines the second training data based on the second input information, the actual output information and the expected output information.
[0112] Through this implementation method, in the actual application process, a more complete model output can be obtained through two steps: the first step is to input the text information representing the claim item to obtain the first specification text information generated by the large language model; if the first specification text information can meet the user's needs, the execution can be terminated; if it cannot meet the needs, the second step is to input the first specification text information and supplementary information to obtain a more complete second specification text information; if it still does not meet the needs, the second specification text information and new supplementary information can be continuously input as needed until satisfactory specification text information is obtained.
[0113] In addition, in other implementations of this embodiment, the first input information may also be abstract information of a patent application document or description information of a solution, etc. The implementer may flexibly apply it according to actual needs.
[0114] Similarly, in other implementations of this embodiment, knowledge base reserve information and technical information identification technology can be used to identify statements representing technical details, parameter information or specifications of specific solutions to replace the operations of steps S3030 to S3039 and directly obtain the second input information.
[0115] Similarly, in other implementations of this embodiment, the second training data may continue to be cleaned and processed to select high-quality training data and train the model to further improve the accuracy and precision of the model.
[0116] For example, language fluency and / or content relevance scoring is performed, a reasonable score threshold is specified, training data with scores below the score threshold is deleted, and only training data with scores above the score threshold is retained.
[0117] For example, the training data whose second input information is empty or whose length is less than the length threshold is deleted, and only the training data whose second input information is not empty or whose length is less than the length threshold is retained.
[0118] Similarly, in other implementations of this embodiment, the second training data may be further processed to obtain training data that is more in line with application habits and closer to real information.
[0119] For example, the second input information in the second training data or the annotated instruction manual text information is subjected to secondary processing such as language polishing, sentence processing or text translation through a language processing tool.
[0120] Through the above implementation, this embodiment can use multiple calculations of the large language generation model to obtain more accurate instruction manual generation results, so that the flexibility and accuracy of the large language generation model in application can be improved.
[0121] Figure 4 Another embodiment of the present application is shown in Figure 3 The model architecture used by the information processing method shown in Figure 1 is shown in Figure 2. Figure 4 As shown, this embodiment adopts a dual-sub-model architecture, that is, the large language generation model 40 includes a first sub-model 401 and a second sub-model 402. In actual applications, the two sub-models collaborate to complete the instruction manual generation task.
[0122] Among them, the first sub-model 401 is used to generate text information 4022 representing the initial specification content based on the text information 4011 representing the claim item; the second sub-model 402 is used to generate text information 4023 representing the improved specification content based on the text information 4022 representing the initial specification content actually output by the first sub-model 401 and the supplementary information 4021 input by the user.
[0123] The first sub-model 401 is trained by using some existing public related documents, while the second sub-model 402 is trained by using the actual output text information representing the content of the initial specification, the second input information and the annotated specification text as training data.
[0124] In addition, in other implementations of this embodiment, the annotation information of the training data of the second sub-model 402 can be the instruction text obtained by fusing the text information representing the initial instruction content with the second input information. The instruction content obtained in this way includes both the supplementary information and all the information generated by the first sub-model 401, so that the generated instruction content is richer and more complete.
[0125] In actual applications, the first sub-model 401 and the second sub-model 402 are usually invisible to the front-end user, and the user only needs to input the text information representing the claim item and the supplementary information respectively.
[0126] In the implementation of this embodiment, the independence of the first sub-model 401 and the second sub-model 402 is fully retained, so that the two sub-models can be trained separately first, and after achieving the best effect, they are jointly trained to obtain a large language generation model 40, so that the accuracy and precision are higher, the quality of the generated results is also higher, and a more satisfactory instruction manual is obtained.
[0127] It should be noted that Figures 1 to 4 The embodiments of the present application shown are only exemplary descriptions of the information processing method of the present application, and are not intended to limit the application scenarios and implementation methods of the embodiments of the present application. The implementer may make any combination and expansion according to the actual implementation conditions and requirements to achieve the best implementation effect.
[0128] For example, in other embodiments of the present application, the first input information, the expected output information, and the actual output information may also be data sets; when comparing the expected output information and the actual output information to determine the second input information, the expected output information and the actual output information may be intersected first, and then the intersection may be subtracted from the data set of the expected output information to obtain information included in the expected output information but not included in the actual output information.
[0129] Furthermore, the present application also provides an information processing device. Figure 5As shown, the device 50 includes: a first training data acquisition module 501, used to acquire first training data, the first training data includes first input information and expected output information corresponding to the first input information; an actual output information determination module 502, used to input the first input information into a machine learning model, perform model operations, and determine the actual output information corresponding to the first input information; a second input information determination module 503, used to compare the expected output information with the actual output information, and determine the second input information, the second input information includes information included in the expected output information but not included in the actual output information; a second training data determination module 504, used to determine the second training data based on the first input information, the second input information, the actual output information and the expected output information.
[0130] According to one embodiment of the present application, the second input information determination module 503 includes: a sub-information determination sub-module, used to determine the sub-information from the expected output information; a target sub-information determination sub-module, used to determine whether the sub-information is included in the actual output information, and if not, the corresponding sub-information is determined as the target sub-information; the second input information determination sub-module, used to determine the second input information based on the target sub-information.
[0131] According to an embodiment of the present application, the sub-information determination submodule is specifically used to divide the expected output information into sentences, and determine each sentence included in the expected output information as sub-information.
[0132] According to an embodiment of the present application, the sub-information determination submodule further includes: a sub-information screening unit, which is used to determine whether the sub-information complies with a predefined rule.
[0133] According to an embodiment of the present application, the second training data determination module 504 is specifically configured to determine the second training data according to the first input information, the second input information and the expected output information.
[0134] According to an embodiment of the present application, the second training data determination module 504 is specifically configured to determine the second training data according to the second input information, the actual output information and the expected output information.
[0135] In addition, the present application also provides an information processing method, which is used in practical application scenarios of machine learning models, such as Figure 6 The embodiment of the present application shown implements the main process of the information processing method of the present application, and the method includes:
[0136] Operation S610, obtaining input information;
[0137] The input information refers to any input information that can be received by the machine learning model. Combined with the related file generation example given above, the input information can be any combination of the following input information:
[0138] Claim text information;
[0139] Claim text information and supplementary information;
[0140] Previously generated instruction manual text information and supplementary information.
[0141] Operation S620, inputting the input information into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model includes a machine learning model trained with third training data, and the third training data includes the second training data determined using the above method.
[0142] The output information refers to the information actually output by the machine learning model after receiving the input information and performing model operations. Combined with the related file generation example given above, the output information is the instruction text information corresponding to the input information.
[0143] The process of training the machine learning model with the third training data (including the second training data) may have been completed before the information processing method of the present application. The subject who completes the above training process may be the application party of the machine learning model, or the provider or distributor of the machine learning model.
[0144] Since the machine learning model includes a process of training with the third training data (including the second training data), it can receive supplementary information to the original input information, making the output results more complete and closer to the user's usage scenarios, which can greatly improve the applicability and practicality of the machine learning model.
[0145] and Figure 6 Corresponding to the above-mentioned information processing method of the present application, the embodiment of the present application also provides an information processing device, such as Figure 7 As shown, the device 70 includes: an input information acquisition module 701, used to obtain input information; an output information determination module 702, used to input the input information into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model includes a machine learning model trained with third training data, and the third training data includes second training data determined using the above method.
[0146] According to a fifth aspect of the embodiments of the present application, a computer storage medium is provided, the storage medium comprising a set of computer executable instructions, which are used to execute any of the above-mentioned information processing methods when the instructions are executed.
[0147] It should be pointed out here that the above description of the information processing device embodiment and the above description of the computer storage medium embodiment are similar to the description of the aforementioned method embodiment and have similar beneficial effects as the aforementioned method embodiment, so they are not described in detail. For the technical details that have not been disclosed in the description of the information processing device embodiment and the description of the computer storage medium embodiment of this application, please refer to the description of the aforementioned method embodiment of this application for understanding, and in order to save space, they are not described in detail.
[0148] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0149] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another device, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.
[0150] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0151] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.
[0152] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage medium, read-only memory (ROM), disk or optical disk, etc. Various media that can store program codes.
[0153] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage media, ROM, magnetic disks or optical disks.
[0154] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. An information processing method, comprising: Acquire first training data, where the first training data includes first input information and expected output information corresponding to the first input information; Inputting the first input information into a machine learning model, performing model calculations, and determining actual output information corresponding to the first input information; Comparing the expected output information with the actual output information to determine second input information, where the second input information includes information included in the expected output information but not included in the actual output information; Second training data is determined according to the first input information, the second input information, the actual output information and the expected output information.
2. The method according to claim 1, wherein comparing the expected output information with the actual output information to determine the second input information comprises: Determining sub-information from the expected output information; Determining whether the sub-information is included in the actual output information, and if not, determining the corresponding sub-information as the target sub-information; The second input information is determined according to the target sub-information.
3. The method according to claim 2, wherein determining sub-information from the expected output information comprises: The expected output information is divided into sentences, and each sentence included in the expected output information is determined as sub-information.
4. The method according to claim 2, after determining the sub-information from the expected output information, further comprising: Determine whether the sub-information complies with a predefined rule.
5. The method according to claim 1, wherein determining the second training data according to the first input information, the second input information, the actual output information and the expected output information comprises: Second training data is determined according to the first input information, the second input information and the expected output information.
6. The method according to claim 1, determining second training data according to the first input information, the second input information, the actual output information and the expected output information, comprising: Second training data is determined according to the second input information, the actual output information and the expected output information.
7. A method according to any one of claims 1 to 6, The machine learning model is used to generate information about the specification in the patent application document, The first input information includes information representing a claim, an abstract or a technical solution, The expected output information or the actual output information includes specification information corresponding to the first input information.
8. An information processing device, comprising: A first training data acquisition module, used to acquire first training data, where the first training data includes first input information and expected output information corresponding to the first input information; an actual output information determination module, configured to input the first input information into a machine learning model, perform model calculations, and determine actual output information corresponding to the first input information; A second input information determination module, configured to compare the expected output information with the actual output information to determine second input information, wherein the second input information includes information included in the expected output information but not included in the actual output information; The second training data determination module is used to determine second training data according to the first input information, the second input information, the actual output information and the expected output information.
9. An information processing method, comprising: Get input information; The input information is input into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model includes a machine learning model trained with third training data, and the third training data includes second training data determined using the method described in claim 1.
10. An information processing device, comprising: An input information acquisition module, used to acquire input information; An output information determination module is used to input the input information into a machine learning model to obtain output information corresponding to the input information, wherein the machine learning model includes a machine learning model trained with third training data, and the third training data includes second training data determined using the method described in claim 1.
