Artificial intelligence-based patent analysis method, system and electronic device

Abstract

The application discloses a patent analysis method and system based on artificial intelligence and electronic equipment, and particularly relates to the technical field of patent information processing, which comprises the following steps: performing semantic analysis on a to-be-applied patent text to generate a corresponding structured technical feature set; converting the structured technical feature set into a database query statement based on the field structure of a patent database, and retrieving a related patent set in the patent database; determining the technical field matching degree and semantic similarity between the to-be-applied patent text and each related patent, and screening out high-related patents according to the technical field matching degree and the semantic similarity; comparing the structured technical feature set with the structured technical feature set corresponding to each high-related patent respectively to obtain technical feature coincidence degree data; and generating a patent analysis result including the high-related patents, the coincident technical features and the different technical features according to the technical feature coincidence degree data. The method can realize related patent screening and technical feature comparison based on structured technical features.

Description

Technical Field

[0001] This invention relates to the field of patent information processing technology, and in particular to a patent analysis method, system and electronic device based on artificial intelligence. Background Technology

[0002] Currently, in the novelty search process before drafting patent application documents, personnel typically input keywords or search terms into patent databases to obtain relevant patents, and then read and compare the technical content of each patent. This method is problematic when faced with patent documents containing dense technical terminology and complex text structures. The search results often contain many irrelevant documents, the screening of relevant patents is time-consuming, and the efficiency of identifying technical overlaps and differences is low. Summary of the Invention

[0003] The purpose of this invention is to solve the technical problem that it is difficult to screen and compare related patents based on structured technical features in patent novelty search analysis, which makes it difficult to automatically determine overlapping and different technical features. The invention provides a patent analysis method, system and electronic device based on artificial intelligence.

[0004] The first aspect of this invention provides an artificial intelligence-based patent analysis method, comprising: performing semantic parsing on a patent application text to generate a structured technical feature set corresponding to the patent application text; converting the structured technical feature set into a database query statement based on the field structure of a patent database, and using the database query statement to retrieve a relevant patent set from the patent database; determining the technical field matching degree and semantic similarity between the patent application text and each relevant patent in the relevant patent set, and selecting highly relevant patents from the relevant patent set based on the technical field matching degree and the semantic similarity; comparing the structured technical feature set with the structured technical feature sets corresponding to each highly relevant patent to obtain technical feature overlap data; and generating a patent analysis result based on the technical feature overlap data, wherein the patent analysis result includes the highly relevant patents, overlapping technical features, and differing technical features.

[0005] In one embodiment of the present invention, the step of semantically parsing the text to be applied for a patent and generating a structured technical feature set corresponding to the text to be applied for a patent includes: dividing the text to be applied for a patent into paragraphs to obtain segmented text; extracting technical features from the segmented text to obtain multiple technical features; and generating the structured technical feature set based on the multiple technical features, wherein the technical features in the structured technical feature set include at least one of feature identifier, feature name, technical description, technical dimension, and corresponding claim.

[0006] In one embodiment of the present invention, the process of converting the structured technical feature set into a database query statement based on the field structure of the patent database includes: determining a target query field in the patent database for storing technical features according to the technical features in the structured technical feature set; generating query conditions corresponding to the target query field based on the technical features; and generating the database query statement according to the target query field and the query conditions.

[0007] In one embodiment of the present invention, generating query conditions corresponding to the target query field based on the technical features includes: extracting technical keywords from the structured technical feature set; generating inclusion relationship query conditions for the target query field based on the technical keywords; wherein the inclusion relationship query conditions include parallel query conditions and / or selective query conditions between at least two technical keywords.

[0008] In one embodiment of the present invention, determining the technical field matching degree and semantic similarity between the patent application text and each related patent in the related patent set includes: determining the technical field matching degree based on at least one of the technical field tags, classification numbers and technical keywords between the patent application text and each related patent; generating a first semantic vector corresponding to the patent application text and obtaining a second semantic vector corresponding to each related patent; calculating the vector similarity between the first semantic vector and each second semantic vector to obtain the semantic similarity.

[0009] In one embodiment of the present invention, the step of selecting highly relevant patents from the relevant patent set based on the technical field matching degree and the semantic similarity includes: determining the comprehensive relevance corresponding to each of the relevant patents based on the technical field matching degree and the semantic similarity; determining the relevant patents whose comprehensive relevance meets the preset relevance condition as the highly relevant patents; if the number of relevant patents that meet the preset relevance condition is less than the preset number, then supplementing the selection of highly relevant patents in descending order of comprehensive relevance.

[0010] In one embodiment of the present invention, the step of comparing the structured technical feature set with the structured technical feature sets corresponding to each of the highly relevant patents to obtain technical feature overlap data includes: determining the overlapping technical features between the text of the patent application and each of the highly relevant patents based on the comparison results between the structured technical feature set and the structured technical feature sets corresponding to each of the highly relevant patents; generating a paragraph group between the paragraph of the patent application and the paragraph of the highly relevant patent based on the text position of the overlapping technical features in the text of the patent application and the text position of the overlapping technical features in the highly relevant patents; sorting the overlapping technical features in the paragraph group according to the order of appearance of the overlapping technical features in the paragraph of the patent application; and determining the technical feature overlap data based on the sorted overlapping technical features.

[0011] In one embodiment of the present invention, generating patent analysis results based on the technical feature overlap data includes: selecting target paragraph groups that meet the overlap conditions from the paragraph groups; inputting the target paragraph groups, the sorted overlapping technical features, and the technical feature overlap data into a pre-trained analysis model to obtain improvement suggestions corresponding to the target paragraph groups; and generating the patent analysis results based on the target paragraph groups, the technical feature overlap data, and the improvement suggestions.

[0012] A second aspect of this invention provides an artificial intelligence-based patent analysis system, comprising: a semantic parsing module for performing semantic parsing on a patent application text to generate a set of structured technical features corresponding to the patent application text; a query generation module for converting the set of structured technical features into a database query statement based on the field structure of a patent database, and using the database query statement to retrieve a set of related patents from the patent database; a related patent filtering module for determining the technical field matching degree and semantic similarity between the patent application text and each related patent in the related patent set, and filtering highly relevant patents from the related patent set based on the technical field matching degree and the semantic similarity; a technical feature comparison module for comparing the set of structured technical features with the set of structured technical features corresponding to each of the highly relevant patents to obtain technical feature overlap data; and a result generation module for generating patent analysis results based on the technical feature overlap data, wherein the patent analysis results include the highly relevant patents, overlapping technical features, and differing technical features.

[0013] A third aspect of the present invention provides an electronic device, including a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the method described in the first aspect.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0015] Compared with the prior art, the technical effects achieved by the present invention are as follows:

[0016] This invention provides an AI-based patent analysis method that parses the text of a patent application into a structured set of technical features, enabling unstructured patent text to be converted into searchable and comparable data objects. This structured set of technical features generates database query statements based on the field structure of a patent database, ensuring that the search conditions correspond to the field information in the patent database, thus obtaining a set of patents related to the text of the patent application. Based on this set of relevant patents, highly relevant patents are selected using technical field matching and semantic similarity, limiting subsequent comparisons to patent texts that are related to the text of the patent application in both technical field and semantic content. Further comparison of the structured set of technical features corresponding to the text of the patent application with the structured set of technical features corresponding to highly relevant patents yields data on the degree of overlap in technical features, generating patent analysis results that include highly relevant patents, overlapping technical features, and differing technical features. This solves the technical problem in patent novelty search analysis where it is difficult to filter and compare relevant patents based on structured technical features, leading to the difficulty in automatically determining overlapping and differing technical features. Attached Figure Description

[0017] Figure 1 This is a flowchart of an artificial intelligence-based patent analysis method according to an embodiment of the present invention; Figure 2 This is a framework diagram of an artificial intelligence-based patent analysis system according to an embodiment of the present invention; Figure 3 This is a framework diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0019] The technical solution of the present invention is illustrated below through specific embodiments. It should be understood that the one or more steps mentioned in the present invention do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not for limiting the order of each method or limiting the scope of the present invention. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of the present invention.

[0020] In one embodiment of this application, an artificial intelligence-based patent analysis method is provided. This method can be deployed on a server, terminal device, or a computing environment formed by the collaboration of a server and terminal device, and can access a patent database and invoke model or algorithm components for text parsing, query statement generation, and text similarity processing. The patent text to be applied for may be a technical disclosure text, a draft patent application text, or the claims text to be analyzed; this application does not specifically limit this.

[0021] like Figure 1 As shown, the patent analysis method based on artificial intelligence in this application embodiment may specifically include the following steps: S101, perform semantic parsing on the text of the patent application to generate a set of structured technical features corresponding to the text of the patent application.

[0022] In this embodiment, after receiving the patent application text, semantic understanding is performed on its technical content, converting the technical description in natural language form into a data form that can be retrieved and compared subsequently. A structured set of technical features is used to characterize multiple technical features contained in the patent application text. Each technical feature can be organized as a text item, a structured field, or a computable data unit, enabling subsequent processing to perform retrieval, filtering, and comparison on a feature-by-feature basis.

[0023] For example, semantic parsing can be performed by a language model capable of understanding patent text. Based on the technical topics, technical objects, processing actions, and relationships between objects in the patent application text, technical features that characterize the technical solution of the patent application text are extracted, and these extracted technical features are organized into a structured set of technical features. This structured set of technical features is no longer used merely as ordinary text for retrieval, but rather serves as the data foundation for subsequently generating database query statements and performing technical feature comparisons.

[0024] Optionally, the language model used to perform semantic parsing can be a pre-trained patent domain language model. This patent domain language model can be obtained by efficiently fine-tuning its parameters based on the base language model using patent domain corpus. The patent domain corpus may include the specification text, claim text, examination document text, and structured sample data generated during patent search and analysis of granted patents. After fine-tuning using this type of patent domain corpus, the language model can identify technical objects, processing actions, functional relationships, and limiting relationships between claim items in the patent text.

[0025] In one feasible approach, a low-rank adaptation method can be used to efficiently fine-tune the parameters of the base language model. During fine-tuning, the training samples in the patent domain are constructed as instruction data formats. Each training sample includes instruction information, input information, output information, and system prompts. The instruction information defines the type of task the model will perform, the input information carries the patent text or technical solution text to be processed, the output information records the structured set of technical features, technical keywords, or parsing results corresponding to the input information, and the system prompts specify the output results according to the patent text parsing rules.

[0026] For example, when the input information of the training sample is a claim text, the output information may include the technical field, technical features, and corresponding text source location extracted from the claim text. Thus, after receiving the patent application text, the fine-tuned patent domain language model can output a set of structured technical features according to the parsing format formed in the training samples. This fine-tuning process can be completed before the patent analysis method is executed; during the execution of the patent analysis method, the trained patent domain language model is directly invoked to perform semantic parsing of the patent application text.

[0027] In this embodiment, the structured technical feature set can be stored in the form of field-based data, key-value pair data, or tabular data. For each technical feature, at least one of the following can be recorded: feature identifier, feature name, technical description, technical dimension, and corresponding claim. The feature identifier is used to distinguish different technical features; the feature name is used to summarize the technical object or processing action corresponding to the technical feature; the technical description is used to retain the semantic content of the technical feature in the patent application text; the technical dimension is used to characterize the technical field, technical means, or technical effect to which the technical feature belongs; and the corresponding claim is used to characterize the source position of the technical feature in the patent application text.

[0028] In one example, a technical feature in the structured technical feature set can be represented as follows: feature identifier F001, feature name "structured technical feature extraction", technical description "semantic parsing of the patent application text and extraction of technical features", belonging to the technical dimension "text parsing", and corresponding claim "claim 1". In another example, multiple technical features can also be organized according to their corresponding claim, technical dimension, or order of appearance in the text, so that subsequent database query generation and technical feature comparison can call the structured technical feature set according to a unified data format.

[0029] S102, based on the field structure of the patent database, the structured technical feature set is converted into a database query statement, and the relevant patent set is retrieved from the patent database using the database query statement.

[0030] Before generating database query statements, the field structure of the patent database can be pre-configured. The patent database stores structured patent information corresponding to published patent data, and includes at least a patent information table, a claim information table, a specification technical feature table, and an abstract information table. The patent information table stores patent bibliographic information, while the claim information table, specification technical feature table, and abstract information table store the paragraph content, technical features, and feature codes of the corresponding text sections, respectively.

[0031] In one example, the patent information table can be named "patent," and its fields include: id, publication_id (patent document number), title (invention title), type (patent type), application_date, publication_data (publication date), applicants, inventor, class_id (patent agency), and agency. Specifically, the id field is a universally unique identifier (UUID) used to record the application number and serves as the primary key of the patent information table; the publication_id field is a UUID used to record the publication or announcement number; the title field is a text field used to record the patent name; the type field is a text field used to record the patent type; the application_date field is a date field used to record the application date; the publication_data field is a date field used to record the publication or announcement date; the applicants field is a text field used to record the applicant; the inventor field is a text field used to record the inventor; the class_id field is a text field used to record the classification number; and the agency field is a text field used to record the agency.

[0032] The claim information table can be named "claim," and its fields include id, segment_id, tenant_id, position, content, technical_feature, and technical_id. Specifically, the id field is of UUID data type and records the application number, serving as a foreign key field associated with the id field in the patent information table; the segment_id field is of UUID data type and records the segment number of the claim text; the tenant_id field is of UUID data type and records the main segment number; the position field is of int4 (4-byte integer) data type and records the segment position; the content field is of text data type and records the segment content; the technical_feature field is of json data type and records the technical feature extracted from the segment content; and the technical_id field is of json data type and records the corresponding feature code.

[0033] The technical feature table in the specification can be named "description," and its fields include id, segment_id, position, content, technology_feature, and technology_id. The id field is of UUID data type and records the application number, and is associated with the id field in the patent information table. The segment_id field is of UUID data type and records the paragraph number of the specification text. The position field is of int4 data type and records the paragraph position. The content field is of text data type and records the paragraph content. The technology_feature field is of json data type and records the technical features extracted from the paragraph content. The technology_id field is of json data type and records the corresponding feature code.

[0034] The abstract information table can be named `abstract`, and its fields include `id`, `segment_id`, `position`, `content`, `technic_feature`, and `technic_id`. The `id` field is of data type `UUID` and records the application number, which is associated with the `id` field in the patent information table. The `segment_id` field is of data type `UUID` and records the paragraph number of the abstract text. The `position` field is of data type `int4` (4-byte integer) and records the paragraph position. The `content` field is of data type `text` and records the abstract content. The `technic_feature` field is of data type `json` and records the technical features extracted from the abstract content. The `technic_id` field is of data type `json` and records the corresponding feature code.

[0035] Based on the above field structure, the `id` field in the patent information table serves as the primary key field for the application number dimension. The `id` fields in the claim information table, specification technical feature table, and abstract information table are respectively used as association fields, establishing a connection with the `id` field in the patent information table. The `publication_id` field is used to identify published or announced text in the search results. The `segment_id` and `position` fields are used to determine the text segment and paragraph position where the technical feature is located. The `content` field stores the original paragraph content, the `technic_feature` field stores the structured technical feature, and the `technic_id` field stores the feature code. Therefore, when converting the set of structured technical features into a database query statement, the `technic_feature` field in `claim`, `description`, and `abstract` can be used as the target query field. Alternatively, the `content`, `technic_id`, `segment_id`, and `position` fields can be combined to generate query conditions, and the corresponding patent identifier, paragraph content, paragraph position, and technical feature information can be returned in the search results.

[0036] In this embodiment, the field structure available for retrieval in the patent database is read, and a database query statement adapted to the field structure is generated based on the technical features in the structured technical feature set. The database query statement is used to search for patent data related to the technical features of the patent application text in the patent database, and the search results form a relevant patent set.

[0037] In one example, after the database query is executed, the patent database returns patent records that meet the query conditions. The relevant patent set may include the corresponding application number, publication or announcement number, paragraph content, paragraph position, technical features, and feature codes.

[0038] S103, determine the technical field matching degree and semantic similarity between the patent text to be applied for and each relevant patent in the relevant patent set, and select highly relevant patents from the relevant patent set based on the technical field matching degree and semantic similarity.

[0039] In this embodiment, a relevance assessment is performed on each relevant patent in the relevant patent set. Technical field matching degree is used to represent the degree of similarity between the patent application text and the relevant patents in terms of technical field; semantic similarity degree is used to represent the degree of similarity between the patent application text and the relevant patents in terms of technical content expression.

[0040] After obtaining the technical field matching degree and semantic similarity of each relevant patent, highly relevant patents are selected from the relevant patent set based on these two factors. Highly relevant patents are those that meet the preset relevance requirements in both technical field and semantic content to the text of the patent application. Through this selection process, subsequent technical feature comparisons do not need to be performed on all patents in the relevant patent set, but only on the selected highly relevant patents.

[0041] S104, compare the structured technical feature set with the structured technical feature set corresponding to each highly relevant patent to obtain technical feature overlap data.

[0042] In this embodiment, a set of structured technical features corresponding to each highly relevant patent is obtained, and the set of structured technical features corresponding to the text of the patent application is compared with the set of structured technical features corresponding to each highly relevant patent. The comparison process is performed on a feature-by-feature basis to determine the technical features that correspond to each highly relevant patent and the technical features that do not correspond to each other.

[0043] Technical feature overlap data is used to characterize the degree of overlap in technical features between the patent application text and highly relevant patents. This data may include the degree of overlap between each highly relevant patent and the patent application text, as well as information on overlapping and non-overlapping technical features used in the calculation. For different highly relevant patents, corresponding technical feature overlap data is obtained separately to generate subsequent patent analysis results.

[0044] S105, Generate patent analysis results based on the overlap data of technical features.

[0045] The patent analysis results in this embodiment include highly relevant patents, overlapping technical features, and differing technical features. Based on the overlap data of technical features corresponding to each highly relevant patent, the technical relationship between the patent application text and the highly relevant patents is organized to generate patent analysis results.

[0046] In this embodiment, highly relevant patents in the patent analysis results are used to indicate existing patents that are highly relevant to the text of the patent application; overlapping technical features are used to indicate technical content that has a corresponding relationship with the text of the patent application and highly relevant patents; and different technical features are used to indicate technical content that does not have a corresponding relationship with the text of the patent application relative to highly relevant patents.

[0047] In one example, patent analysis results can be output in the form of tables, text reports, or structured data. For each highly relevant patent, the patent analysis results can record the patent identifier of the highly relevant patent, the technical features that overlap with the text of the patent application, the technical features that differ from the text of the patent application, and the corresponding technical feature overlap data.

[0048] This embodiment parses the patent application text into a structured set of technical features, enabling unstructured patent text to be converted into searchable and comparable data objects. This structured set of technical features generates database query statements based on the field structure of the patent database, ensuring that the search conditions correspond to the field information in the patent database, thus obtaining a set of patents related to the patent application text. Based on this set of relevant patents, highly relevant patents are selected using technical field matching and semantic similarity, limiting subsequent comparisons to patent texts that are related to the patent application text in both technical field and semantic content. Further comparison of the structured set of technical features corresponding to the patent application text with the structured set of technical features corresponding to highly relevant patents yields data on technical feature overlap, generating patent analysis results that include highly relevant patents, overlapping technical features, and differing technical features. This solves the technical problem in patent novelty search analysis where it is difficult to filter and compare relevant patents based on structured technical features, leading to difficulties in automatically determining overlapping and differing technical features.

[0049] In an optional embodiment of this application, when performing semantic parsing on the patent application text, the patent application text can be divided into paragraphs to obtain segmented text.

[0050] Specifically, paragraph segmentation is used to break down text content with different semantic functions in the patent application text into relatively independent processing units, so that subsequent technical feature extraction can be carried out within a smaller text scope. The segmented text can be text fragments obtained by dividing the text according to natural paragraphs, chapter identifiers, claim numbers, period separation positions, or semantic boundaries; when the patent application text is a technical disclosure text, the segmented text can also correspond to content fragments such as technical problems, technical solutions, and implementation methods.

[0051] Technical features were extracted from the segmented text to obtain multiple technical features.

[0052] Specifically, this technical feature extraction is used to identify objects, actions, processing relationships, data objects, or functional units that constitute a technical solution from segmented text. For the same segmented text, one or more technical features can be extracted; for semantically identical or similar expressions in different segmented texts, they can be merged or normalized during the extraction process to reduce the impact of duplicate features on subsequent retrieval and comparison.

[0053] After obtaining multiple technical features, a structured set of technical features is generated based on these features. This structured set of technical features can be organized according to the semantic relationships between the technical features, their corresponding text positions, or the execution order in the technical solution. This ensures that each technical feature no longer exists in isolation as a paragraph of ordinary natural language, but rather forms a data set that can be used in database query generation, related patent screening, and technical feature comparison processes.

[0054] In one optional embodiment of this application, based on the field structure of the patent database, the structured technical feature set is converted into a database query statement, including but not limited to: determining the target query field in the patent database for storing technical features according to the technical features in the structured technical feature set; generating query conditions corresponding to the target query field based on the technical features; and generating a database query statement according to the target query field and the query conditions.

[0055] In this embodiment, the target query field refers to a field in the patent database that can carry technical feature content and can be invoked by a query statement. It can correspond to a technical feature field in the claim text, or to a technical feature field in the specification text, abstract text, or other structured patent information. For the same technical feature, when there are multiple searchable fields in the patent database, one or more target query fields can be determined based on the text source or technical dimension to which the technical feature belongs.

[0056] Query conditions corresponding to the target query field are generated based on technical features. This process converts the technical features in the structured technical feature set into conditional expressions that can be executed by the database, allowing the technical features to participate in the retrieval as query values ​​for the target query field. Query conditions can be used to limit whether the target query field contains the corresponding technical feature, or to limit the similarity or equivalence relationship between the target query field and the corresponding technical feature. For multiple technical features, corresponding query conditions can be generated separately, and the combination method of each query condition can be determined based on the relationship between the multiple technical features.

[0057] Based on the target query fields and query conditions, a database query statement is generated. This statement includes the data tables to be accessed, the target query fields, and the corresponding query conditions, enabling the structured technical feature set to be transformed into an executable retrieval expression for the patent database. The generated database query statement is used to retrieve patent data related to the technical features of the patent application text from the patent database. The search results are then used as a set of relevant patents for subsequent filtering.

[0058] In one feasible approach, when converting a set of structured technical features into a database query, a text-to-structured query language model can be used to generate the database query. The information input to this model may include a field structure description text of the patent database, a set of structured technical features corresponding to the text of the patent application, technical keywords extracted from the set of structured technical features, and example query pairs for example guidance.

[0059] The field structure description text provides the model with the searchable data tables, field names, and field meanings in the patent database, enabling the model to determine the data tables and fields invoked by the database query. The structured technical feature set provides the technical content to be retrieved from the patent application text, technical keywords define the core search objects in the query conditions, and query example pairs provide conversion styles between technical features and database query statements. Query example pairs may include historical technical keywords, historical structured technical feature sets, and their corresponding historical database query statements.

[0060] For example, a model suggestion term can be formed by concatenating the field structure description text, the structured technical feature set, technical keywords, and query examples. This model suggestion term is then input into a text-to-structured query language model. The text-to-structured query language model generates candidate database query statements based on the model suggestion term. The candidate database query statements include the data table to be queried, the target query field, and the query conditions corresponding to the technical features or technical keywords.

[0061] After generating candidate database query statements, syntax and database structure consistency checks can be performed. Syntax check verifies whether the candidate query statement conforms to the syntax rules of the Structured Query Language (SCL), while database structure consistency check verifies whether the data tables and fields called in the candidate query statement exist in the field structure of the patent database. When a candidate query statement fails the check, the check error message, the candidate query statement, and the field structure description text are fed back to the text-to-structured query language model to regenerate a corrected database query statement. When a candidate query statement passes the check, it is used as the database query statement for retrieving relevant patent sets.

[0062] Optionally, query example pairs can be derived from historical retrieval tasks, retrieval requests entered by testers, or validated database query statements. Each query example pair may include technical keywords, a structured set of technical features, and their corresponding database query statement.

[0063] For example, when the technical keywords include "Boolean search" and "semantic search," these two technical keywords can be used as query values ​​for the target query field to generate inclusion relationship query conditions covering the technical feature fields in the claim data, specification data, and / or abstract data, in order to retrieve relevant patent records containing the aforementioned technical keywords. Thus, the query example provides a style for converting technical keywords into database query statements for a text-to-structured query language model.

[0064] In one optional embodiment of this application, the query conditions corresponding to the target query field are generated based on technical features, including but not limited to: extracting technical keywords from a structured set of technical features; generating inclusion relationship query conditions for the target query field based on the technical keywords; wherein, the inclusion relationship query conditions include parallel query conditions and / or selective query conditions between at least two technical keywords.

[0065] In this embodiment, technical keywords are used to characterize the core objects, processing actions, data types, or functional relationships within a technical feature. These keywords can be directly derived from the technical feature text, or obtained by filtering the technical feature text through synonym unification, word form correction, or redundant word filtering. For a single technical feature, one technical keyword can be extracted, or multiple technical keywords with combined retrieval significance can be extracted.

[0066] Generate inclusion relationship query conditions for the target query field based on technical keywords. These inclusion relationship query conditions are used to restrict the content of the target query field to include the corresponding technical keywords, enabling database retrieval to be performed around the core technical content of the structured technical feature set. For example, when the target query field is a field used to store technical features, the technical keywords can be used as the query value for that target query field to generate query conditions where the field content includes the technical keywords; when there are multiple target query fields, corresponding inclusion relationship query conditions can be generated for each target query field separately.

[0067] In one example, when multiple technical keywords jointly define the same technical feature, at least two parallel query conditions can be generated, ensuring that the search results simultaneously satisfy the query requirements corresponding to multiple technical keywords. When multiple technical keywords represent different expressions of the same technical meaning, at least two alternative query conditions can be generated, ensuring that the search results satisfy the query requirements corresponding to any one of the technical keywords. Parallel and alternative query conditions can be combined based on the semantic relationships between technical keywords to form a database query statement targeting the target query field.

[0068] In one optional embodiment of this application, the technical field matching degree and semantic similarity between the patent application text and each related patent in the related patent set are determined, including but not limited to: determining the technical field matching degree based on at least one of the technical field tags, classification numbers and technical keywords between the patent application text and each related patent; generating a first semantic vector corresponding to the patent application text and obtaining a second semantic vector corresponding to each related patent; calculating the vector similarity between the first semantic vector and each second semantic vector to obtain the semantic similarity.

[0069] In this embodiment, the technical field label can be used to characterize the technical branch to which the patent text belongs, the classification number can be the classification information stored in the patent database, and the technical keywords can be derived from the core technical expressions extracted from the patent application text and related patents. For each related patent, its corresponding technical field label, classification number, or technical keywords are compared with the information corresponding to the patent application text to obtain the technical field matching degree of the related patent.

[0070] When determining semantic similarity, a first semantic vector corresponding to the patent application text can be generated, and second semantic vectors corresponding to each related patent can be obtained. The first semantic vector is used to represent the semantic content of the patent application text, and the second semantic vector is used to represent the semantic content of the related patents. The second semantic vectors corresponding to the related patents can be pre-stored in a patent database or in vector data associated with the patent database, or they can be generated based on the related patent text when analyzing related patents.

[0071] Furthermore, the vector similarity between the first semantic vector and each of the second semantic vectors is calculated to obtain the semantic similarity. For each relevant patent, the similarity between its corresponding second semantic vector and the first semantic vector is calculated to obtain the semantic similarity between the relevant patent and the text of the patent application. Thus, the technical field matching degree reflects the degree of closeness between the text of the patent application and the relevant patent at the technical field level, while the semantic similarity reflects the degree of closeness between the two at the textual semantic level.

[0072] In one optional embodiment of this application, highly relevant patents are selected from the relevant patent set based on technical field matching degree and semantic similarity, including but not limited to: determining the comprehensive relevance of each relevant patent based on technical field matching degree and semantic similarity; determining relevant patents whose comprehensive relevance meets the preset relevance condition as highly relevant patents; if the number of relevant patents that meet the preset relevance condition is less than the preset number, then highly relevant patents are selected in order of comprehensive relevance from high to low.

[0073] In this embodiment, the comprehensive relevance is used to comprehensively characterize the degree of domain proximity and semantic proximity between the relevant patent and the patent application text. It can be obtained by fusing technical field matching degree and semantic similarity. For each relevant patent in the relevant patent set, the technical field matching degree and semantic similarity corresponding to the relevant patent are obtained respectively, and the comprehensive relevance corresponding to the relevant patent is generated according to the preset fusion rules.

[0074] In one example, the preset fusion rule can be a weighted fusion rule, where the technical field matching degree and semantic similarity each correspond to different weight parameters. The weight parameters can be set according to the focus of the patent analysis task; when focusing more on existing patents within the same technical field, the weight corresponding to the technical field matching degree can be increased; when focusing more on the similarity of the textual technical content, the weight corresponding to semantic similarity can be increased. Alternatively, the technical field matching degree and semantic similarity can be normalized first, and then the overall relevance can be determined based on the normalized values.

[0075] In a specific example, the technical field matching degree and semantic similarity can be normalized to a numerical range of 0 to 1, and the comprehensive relevance R corresponding to the i-th relevant patent can be determined according to the following formula. i : (1) Among them, F i S represents the degree of technical matching between the text of the patent application and the i-th related patent. i This represents the semantic similarity between the text of the patent application and the i-th related patent, where a and b are the corresponding weight coefficients, and a + b = 1. For example, when the analysis task focuses more on the closeness of semantic content, a can be 0.4 and b can be 0.6; when the analysis task focuses more on patents within the same technical field, a can be 0.6 and b can be 0.4. The weight coefficients a and b can be set based on practical experience, and are not limited in this embodiment.

[0076] When screening for highly relevant patents, the overall relevance R can be used. i Compare with a preset relevance threshold Tr. If R i If the relevance threshold is greater than or equal to Tr, then the i-th relevant patent is identified as a highly relevant patent. The preset relevance threshold Tr can be set to any value between 0.5 and 0.75, for example, 0.65. When R is satisfied... i When the number of relevant patents ≥Tr is less than the preset number M, the comprehensive relevance R can be used as the criterion. i In descending order, select highly relevant patents from the remaining relevant patents until the number of highly relevant patents reaches M, or until there are no unselected relevant patents in the relevant patent set.

[0077] After obtaining the overall relevance of each relevant patent, patents whose overall relevance meets the preset relevance criteria are identified as highly relevant patents. The preset relevance criteria can be that the overall relevance is greater than or equal to a preset threshold, or that the overall relevance ranks among the top few in the sorting results. By filtering the relevant patent set using the preset relevance criteria, the number of patents entering the subsequent technical feature comparison is controlled.

[0078] When the number of relevant patents meeting the preset relevance criteria is less than the preset number, highly relevant patents can be selected from those that did not meet the preset relevance criteria, in descending order of overall relevance, until the number of highly relevant patents reaches the preset number, or there are no more relevant patents to be selected from the relevant patent set. Therefore, even when the number of relevant patents is small or the relevance distribution is scattered, a certain number of candidate patents can still be reserved for subsequent technical feature comparisons.

[0079] Optionally, semantic similarity, overall relevance, and technical feature overlap can be compared with corresponding thresholds to determine whether related patents, highly relevant patents, or paragraph groups meet the conditions for subsequent processing. These thresholds can be pre-determined based on authorized patent samples from the same or similar technical fields in the patent database, or adjusted according to different analysis tasks. For example, several groups of patent texts with similar technical fields can be selected first, their semantic similarity and technical feature overlap can be calculated, and then the threshold range applicable to the current patent analysis task can be determined based on the calculation results.

[0080] In one example, the semantic similarity threshold and the technical feature overlap threshold can be set between 0.5 and 0.75. When the semantic similarity between the patent application text and related patents reaches the semantic similarity threshold, they can be considered to meet the similarity requirement at the textual semantic level; when the technical feature overlap between the patent application text and highly related patents reaches the technical feature overlap threshold, they can be considered to meet the overlap condition at the technical feature level. These thresholds can also be adjusted according to the technical field of the patent text, the scope of the search, or the distribution of database samples.

[0081] When determining whether technical features overlap, both literal consistency and semantic similarity can be used as criteria for establishing an overlap relationship. Literal consistency means that the technical features in the patent application text are identical or substantially identical in wording to the technical features in a highly relevant patent; semantic similarity means that even if the wording differs, they still point to the same or similar technical objects, processing actions, data relationships, or functional relationships. For semantically similar technical features, whether an overlap relationship exists can be determined through semantic vector similarity, correspondence of technical keywords, or model judgment results.

[0082] In one example, the technical feature in the patent application text is "converting a structured set of technical features into a database query statement," while the technical feature in a highly relevant patent is "generating SQL query conditions based on technical features." Although the wording is not entirely identical, both involve the process of generating executable database query content based on technical features, and can be judged as semantically similar overlapping technical features. Technical features that cannot achieve literal consistency or semantic similarity can be included in the patent analysis results as differing technical features.

[0083] In an optional embodiment of this application, when comparing the structured technical feature set with the structured technical feature sets corresponding to each highly relevant patent, the overlapping technical features between the patent application text and each highly relevant patent can be determined based on the comparison results between the structured technical feature set and the structured technical feature sets corresponding to each highly relevant patent. Overlapping technical features refer to technical features that can form a corresponding relationship between the patent application text and the highly relevant patent. This correspondence can be determined based on the textual content, semantic meaning, or the combination relationship between the technical object and the processing action of the technical feature.

[0084] After identifying overlapping technical features, paragraph groups are generated between the paragraphs of the patent application and the highly relevant patents, based on the text positions of the overlapping technical features in the text of the patent application and in the text of highly relevant patents. Text positions can indicate the paragraph, claim, specification paragraph, or other text unit where the overlapping technical feature is located. For the same overlapping technical feature, a correspondence is established between its paragraph in the patent application and its paragraph in the highly relevant patent, thus forming a paragraph group. A paragraph group may include one paragraph of the patent application and one or more corresponding paragraphs of highly relevant patents.

[0085] Furthermore, the overlapping technical features in the paragraph group are sorted according to their order of appearance in the paragraphs of the patent application. For multiple overlapping technical features within the same paragraph group, their order in the paragraphs of the patent application can be used as the sorting basis, so that the sorted overlapping technical features can reflect the descriptive order of the technical solutions in the patent application text.

[0086] Based on the sorted overlapping technical features, the overlap degree of technical features is determined. In a specific example, for a group of paragraphs formed by a paragraph of a patent application and a highly relevant patent paragraph, all technical features contained in the paragraph of the patent application can be identified first, and feature weights can be assigned to each technical feature. The feature weights can be determined according to the technical dimension to which the technical feature belongs, or they can be uniformly set to 1.

[0087] Suppose that the patent application paragraph contains n technical features, denoted as f1 to f2. n The weights corresponding to each technical feature are w1 to w1 respectively. n Among them, the technical features that overlap with highly relevant patent paragraphs form a set of overlapping technical features C. The degree of overlap of the technical features corresponding to this paragraph group is O. g It can be determined using the following formula: (2) in, This represents the sum of the weights corresponding to each overlapping technical feature in the set C of overlapping technical features. This represents the sum of the weights of all technical features in the paragraph to be applied for, where k corresponds to the technical features in the set C of overlapping technical features, and j corresponds to all technical features in the paragraph to be applied for. If the weight of each technical feature is set to 1, then the degree of overlap of technical features is O. g It is the ratio between the number of overlapping technical features and the total number of technical features in the paragraph of the patent application.

[0088] When a highly relevant patent has multiple paragraph groups, the overlap of technical features corresponding to each paragraph group can be calculated separately, or the overlap of technical features corresponding to each paragraph group can be weighted according to the paragraph weight to obtain the overlap data of technical features of the highly relevant patent relative to the text of the patent application.

[0089] Optionally, when generating patent analysis results based on technical feature overlap data, target paragraph groups that meet the overlap criteria can be selected from the paragraph groups. The overlap criteria can be determined based on the technical feature overlap data. For example, paragraph groups with a technical feature overlap reaching a preset threshold can be identified as target paragraph groups, or a preset number of paragraph groups can be selected as target paragraph groups in descending order of technical feature overlap. Target paragraph groups represent combinations of text segments in the patent application text that have a high degree of technical feature overlap with highly relevant patents.

[0090] If the overlap of technical features in a paragraph group reaches the overlap threshold, the paragraph group is identified as the target paragraph group. If no paragraph group reaches the overlap threshold, the improvement suggestion generation process is skipped, and patent analysis results are generated based on existing highly relevant patents, overlapping technical features, different technical features, and technical feature overlap data.

[0091] For a target paragraph group, the target paragraph group, the sorted overlapping technical features, and the technical feature overlap data can be input into a pre-trained analysis model to obtain improvement suggestions for the target paragraph group. The data input into the analysis model can include the text of the patent application paragraph, the text of highly relevant patent paragraphs, the sorted overlapping technical features, the differing technical features, and the technical feature overlap degree corresponding to the target paragraph group. The sorted overlapping technical features are used to provide the analysis model with the arrangement relationship of technical features in the patent application paragraphs, and the technical feature overlap degree data is used to provide the analysis model with the degree of overlap between the target paragraph group and highly relevant patents.

[0092] In one example, the output of the analysis model can be in a structured data format, including paragraph group identifiers, related patent identifiers, overlapping technical features, differing technical features, overlap degree values, and improvement suggestion fields. The improvement suggestion field records text adjustment information for the target paragraph group, which may include the text position to be adjusted, suggested adjustments, and corresponding differing technical features. The improvement suggestions do not change the basic order of technical features in the patent application text; rather, they provide adjustment information to differentiate expressions or highlight differing technical features for text content with a high degree of overlap with highly relevant patents.

[0093] Patent analysis results are generated based on target paragraph groups, technical feature overlap data, and improvement suggestions. The patent analysis results can record the target paragraph groups, highly relevant patents, overlapping technical features, differing technical features, technical feature overlap data, and improvement suggestions.

[0094] Patent analysis results are generated based on target paragraph groups, technical feature overlap data, and improvement suggestions. The patent analysis results can record the target paragraph groups, highly relevant patents, overlapping technical features, technical feature overlap data, and improvement suggestions, enabling the output to simultaneously reflect the overlap relationship between the patent application text and highly relevant patents, as well as corresponding text processing suggestions.

[0095] Another embodiment of the present invention relates to an artificial intelligence-based patent analysis system, such as... Figure 2 As shown, the system may include: Semantic parsing module 20 is used to perform semantic parsing on the text of the patent application and generate a set of structured technical features corresponding to the text of the patent application. The query generation module 21 is used to convert the structured technical feature set into a database query statement based on the field structure of the patent database, and use the database query statement to retrieve the relevant patent set in the patent database. The relevant patent screening module 22 is used to determine the technical field matching degree and semantic similarity between the patent application text and each relevant patent in the relevant patent set, and to screen out highly relevant patents from the relevant patent set based on the technical field matching degree and the semantic similarity. The technical feature comparison module 23 is used to compare the structured technical feature set with the structured technical feature set corresponding to each of the highly relevant patents to obtain technical feature overlap data. The result generation module 24 is used to generate patent analysis results based on the technical feature overlap data. The patent analysis results include the highly relevant patents, overlapping technical features, and different technical features.

[0096] Optionally, in this embodiment, the semantic parsing module 20 includes: The paragraph segmentation submodule is used to segment the text to be applied for patent into paragraphs, resulting in segmented text. The feature extraction submodule is used to extract technical features from the segmented text to obtain multiple technical features; The feature set generation submodule is used to generate the structured technical feature set based on the plurality of technical features. The technical features in the structured technical feature set include at least one of feature identifier, feature name, technical description, technical dimension and corresponding claim.

[0097] Optionally, in this embodiment, the query generation module 21 includes: The first determining submodule is used to determine the target query field in the patent database for storing technical features based on the technical features in the structured technical feature set. The query condition generation submodule is used to generate query conditions corresponding to the target query field based on the technical features. The query statement generation submodule is used to generate the database query statement based on the target query field and the query conditions.

[0098] Optionally, in this embodiment, the query condition generation submodule includes: The keyword extraction unit is used to extract technical keywords from the structured technical feature set; The query condition generation unit is used to generate inclusion relationship query conditions for the target query field based on the technical keywords; The inclusion relationship query conditions include parallel query conditions and / or one-choice query conditions between at least two technical keywords.

[0099] Optionally, in this embodiment, the relevant patent screening module 22 includes: The second determining submodule is used to determine the technical field matching degree based on at least one of the technical field tags, classification numbers and technical keywords between the patent application text and each of the related patents; The acquisition submodule is used to generate a first semantic vector corresponding to the text of the patent application and to acquire a second semantic vector corresponding to each of the related patents; The third determining submodule is used to calculate the vector similarity between the first semantic vector and each of the second semantic vectors to obtain the semantic similarity.

[0100] Optionally, in this embodiment, the relevant patent screening module 22 further includes: The fourth determination submodule is used to determine the comprehensive relevance of each of the related patents based on the technical field matching degree and the semantic similarity. The fifth determination submodule is used to determine the relevant patents whose comprehensive relevance meets the preset relevance conditions as the highly relevant patents; The selection submodule is used to supplement the highly relevant patents in descending order of comprehensive relevance if the number of relevant patents that meet the preset relevance conditions is less than the preset number.

[0101] Optionally, in this embodiment, the technical feature comparison module 23 includes: The sixth determining submodule is used to determine the overlapping technical features between the patent application text and each of the highly relevant patents based on the comparison results between the structured technical feature set and the structured technical feature set corresponding to each of the highly relevant patents. The paragraph group generation submodule is used to generate a paragraph group between the paragraph of the patent application and the paragraph of the highly relevant patent based on the text position of the overlapping technical features in the text of the patent application and the text position of the highly relevant patent. The sorting submodule is used to sort the overlapping technical features in the paragraph group according to the order in which the overlapping technical features appear in the paragraph of the patent application; The seventh determination submodule is used to determine the overlap degree data of the technical features based on the sorted overlapping technical features.

[0102] Optionally, in this embodiment, the result generation module 24 includes: The paragraph group filtering submodule is used to filter target paragraph groups that meet the overlap condition from the paragraph groups; An improvement suggestion generation submodule is used to input the target paragraph group, the sorted overlapping technical features, and the overlap data of the technical features into a pre-trained analysis model to obtain improvement suggestions corresponding to the target paragraph group. The analysis results submodule is used to generate the patent analysis results based on the target paragraph group, the technical feature overlap data, and the improvement suggestions.

[0103] Another embodiment of the present invention relates to an electronic device, such as... Figure 3 As shown, it includes: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions executable by the at least one processor 301, the instructions being executed by the at least one processor 301 to enable the at least one processor 301 to execute the artificial intelligence-based patent analysis method in the above embodiments.

[0104] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0105] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A patent analysis method based on artificial intelligence, characterized in that, include: Semantic parsing is performed on the text of the patent application to generate a set of structured technical features corresponding to the text of the patent application; Based on the field structure of the patent database, the structured technical feature set is converted into a database query statement, and the relevant patent set is retrieved from the patent database using the database query statement. Determine the technical field matching degree and semantic similarity between the patent application text and each related patent in the related patent set, and filter out highly relevant patents from the related patent set based on the technical field matching degree and the semantic similarity; The structured technical feature set is compared with the structured technical feature set corresponding to each of the highly relevant patents to obtain technical feature overlap data. Patent analysis results are generated based on the overlap data of the technical features. The patent analysis results include highly relevant patents, overlapping technical features, and different technical features.

2. The method according to claim 1, characterized in that, The semantic parsing of the patent application text to generate a set of structured technical features corresponding to the patent application text includes: The text to be applied for patent is divided into paragraphs to obtain segmented text; Technical features are extracted from the segmented text to obtain multiple technical features; Based on the aforementioned multiple technical features, a structured technical feature set is generated. The technical features in the structured technical feature set include at least one of feature identifier, feature name, technical description, technical dimension, and corresponding claim.

3. The method according to claim 1, characterized in that, The field structure based on the patent database converts the structured technical feature set into a database query statement, including: Based on the technical features in the structured technical feature set, determine the target query field in the patent database used to store the technical features; Generate query conditions corresponding to the target query field based on the aforementioned technical features; The database query statement is generated based on the target query field and the query conditions.

4. The method according to claim 3, characterized in that, The step of generating query conditions corresponding to the target query field based on the technical features includes: Extract technical keywords from the structured technical feature set; Generate inclusion relationship query conditions for the target query field based on the technical keywords; The inclusion relationship query conditions include parallel query conditions and / or one-choice query conditions between at least two technical keywords.

5. The method according to claim 1, characterized in that, The determination of the technical field matching degree and semantic similarity between the patent application text and each relevant patent in the relevant patent set includes: The technical field matching degree is determined based on at least one of the technical field tags, classification numbers and technical keywords between the patent application text and each of the relevant patents. Generate a first semantic vector corresponding to the text of the patent application to be filed, and obtain a second semantic vector corresponding to each of the related patents; The semantic similarity is obtained by calculating the vector similarity between the first semantic vector and each of the second semantic vectors.

6. The method according to claim 1, characterized in that, The process of selecting highly relevant patents from the relevant patent set based on the technical field matching degree and the semantic similarity includes: Based on the technical field matching degree and the semantic similarity, the comprehensive relevance degree corresponding to each of the relevant patents is determined; Patents whose overall relevance meets the preset relevance conditions are identified as highly relevant patents; If the number of relevant patents that meet the preset relevance conditions is less than the preset number, then the highly relevant patents will be selected in descending order of the overall relevance.

7. The method according to claim 1, characterized in that, The step of comparing the structured technical feature set with the structured technical feature sets corresponding to each of the highly relevant patents to obtain technical feature overlap data includes: Based on the comparison results between the structured technical feature set and the structured technical feature set corresponding to each of the highly relevant patents, the overlapping technical features between the patent application text and each of the highly relevant patents are determined. Based on the text positions of the overlapping technical features in the text of the patent application and the text positions in the highly relevant patent, a paragraph group is generated between the paragraph of the patent application and the paragraph of the highly relevant patent. The overlapping technical features in the paragraph group are sorted according to the order in which they appear in the paragraph of the patent application; Based on the sorted overlapping technical features, the overlap degree data of the technical features is determined.

8. The method according to claim 7, characterized in that, The process of generating patent analysis results based on the overlap data of the technical features includes: Select target paragraph groups that meet the overlap criteria from the paragraph groups; The target paragraph group, the sorted overlapping technical features, and the overlap data of the technical features are input into a pre-trained analysis model to obtain improvement suggestions corresponding to the target paragraph group. The patent analysis results are generated based on the target paragraph group, the technical feature overlap data, and the improvement suggestions.

9. A patent analysis system based on artificial intelligence, characterized in that, include: The semantic parsing module is used to perform semantic parsing on the patent application text and generate a set of structured technical features corresponding to the patent application text. The query generation module is used to convert the structured technical feature set into a database query statement based on the field structure of the patent database, and use the database query statement to retrieve the relevant patent set in the patent database. The relevant patent screening module is used to determine the technical field matching degree and semantic similarity between the patent application text and each relevant patent in the relevant patent set, and to screen out highly relevant patents from the relevant patent set based on the technical field matching degree and the semantic similarity. The technical feature comparison module is used to compare the structured technical feature set with the structured technical feature set corresponding to each of the highly relevant patents to obtain technical feature overlap data. The result generation module is used to generate patent analysis results based on the technical feature overlap data. The patent analysis results include highly relevant patents, overlapping technical features, and different technical features.

10. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the method according to any one of claims 1 to 8.

Images