Patent behavior knowledge graph construction and analysis method based on innovation subject perspective

By building a knowledge graph of multiple patent behavior relationships, identifying core innovation entities and recommending potential partners, the limitations of single patent behavior network analysis in the existing technology are solved, and accurate layout identification and high-quality cooperative recommendation of subdivided technical fields are achieved. , meets the innovation needs of enterprises.

CN118917393AInactive Publication Date: 2024-11-08张桐瑞
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202410959942.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing patent behavior network analysis only considers a single patent behavior, ignores the relationship between multiple types of patent behaviors, resulting in the identification of core subjects and potential collaborators' exploration of enterprises that is difficult to meet the scientific and technological innovation needs of enterprises, and the layout and attention differences in the subdivided technology fields have not been Fully considering it will lead to identification bias.

Method used

Using the patent behavior knowledge graph construction analysis method based on the perspective of innovative subjects, by obtaining enterprise patent data, identifying core IPC classification numbers, conducting technical cluster analysis, extracting patent behavior data, building a knowledge graph of multiple relationships, and using relationship strength and control intensity to identify Core subjects and recommend potential collaborators based on path strength.

Benefits of technology

It realizes accurate layout identification of innovative subjects in subdivided technology fields and intelligent recommendation of potential partners, promotes collaborative high-quality development, meets the enterprise's scientific and technological innovation needs, improves the accuracy of core subject identification and effective cooperative relationships. sex.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917393A_ABST
    Figure CN118917393A_ABST
Patent Text Reader

Abstract

The invention relates to the field of analysis and research of intelligence information, in particular to a patent behavior knowledge graph construction and analysis method based on an innovation subject perspective. The technical problem is that the single patent behavior network analysis research in the prior art ignores the layout and attention difference of each innovation subject in the subdivision technical field, and is difficult to meet the actual scientific and technological innovation demand of the innovation subject; according to the technical scheme, the patent behavior knowledge graph construction and analysis method based on the innovation subject view angle comprises the steps that multiple times of data updating and consistency judgment are repeatedly conducted according to the time sequence to achieve entity alignment; according to the invention, on the basis of carrying out clustering analysis on patent technology layout, patent behavior relationships of innovation subjects in a technology cluster are analyzed by constructing a patent behavior knowledge graph comprising multiple relationships of patent cooperation application, patent right transfer and patent permission use; a patent behavior core subject automatic identification and potential partner intelligent recommendation method is constructed, and collaborative development of innovation subjects is promoted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of analysis and research of intelligence data, and in particular to a method for constructing and analyzing a patent behavior knowledge graph based on the perspective of an innovation subject. Background Art

[0002] Social networks can describe the meshed interactive relationships between subjects and are widely used in the analysis of patent behavior relationships such as patent cooperation, patent rights transfer, and patent licensing, in order to clarify the industrial technology layout and evolutionary laws and explore the impact mechanism of network characteristics on scientific and technological innovation. Patent cooperation analysis, patent rights transfer analysis, and patent licensing relationship analysis based on social networks clarify the structural characteristics and evolutionary laws of the technical field, and confirm the intelligence significance of patent behavior for innovative subjects to find potential collaborators and build patent cooperation relationships. In addition, social networks based on single patent behavior can also predict potential cooperation relationships to a certain extent through the construction of indicator systems and link analysis. However, the existing patent behavior network only considers a single patent behavior and ignores the relationship links between innovative subjects based on multi-type patent behavior relationships, resulting in the difficulty of core subject identification and potential collaborator mining based on link analysis to meet the technological innovation needs of enterprises. The single patent behavior network analysis research under the existing technology ignores the layout and attention differences of various innovative subjects in the subdivided technology fields. The existence of industrial noise makes the core subject identification and potential collaborator recommendation based on industrial data deviate from the actual situation of the subdivided fields that enterprises pay attention to, making it difficult to meet the actual technological innovation needs of innovative subjects. Summary of the invention

[0003] The single patent behavior network analysis research under the existing technology ignores the differences in the layout and focus of various innovation entities in the subdivided technology fields. The existence of industrial noise makes the core entity identification and potential collaborator recommendation based on industrial data deviate from the actual situation of the subdivided fields that enterprises pay attention to, making it difficult to meet the actual technological innovation needs of innovation entities.

[0004] The technical solution of the present invention is: a patent behavior knowledge graph construction and analysis method based on the innovation subject perspective:

[0005] S1: To obtain enterprise patent data, it is first necessary to identify the target enterprises and collect their patent data, which usually involves retrieving and downloading patent documents of relevant enterprises from patent databases, such as the China National Intellectual Property Administration database, the European Patent Office database, and the United States Patent and Trademark Office database; data collection covers a sufficient time range to fully reflect the technological innovation process of the enterprise, and also covers detailed information on the patent, including but not limited to patent number, application date, publication date, applicant, IPC classification number, etc.;

[0006] S2: Identification of core IPC classification codes: Analyze the collected patent data to identify the IPC classification codes most commonly used by the enterprise. These classification codes represent the core technology areas of the enterprise. By counting the frequency of occurrence of each classification code, it is possible to determine which classification codes best represent the enterprise's technology focus.

[0007] S3: Technology cluster analysis based on IPC classification numbers. After determining the core IPC classification numbers, use these classification numbers to perform cluster analysis on patents to identify different technology clusters. Cluster analysis can classify similar patents together to form technology groups. Each group represents a specific technology field or technology direction to reveal the technology layout and key development direction within the enterprise.

[0008] S4: Obtain patent data of technology clusters. Based on the results of technology cluster analysis, conduct secondary data collection for each technology cluster to obtain more detailed and in-depth patent data for subsequent patent behavior analysis and knowledge graph construction;

[0009] S5: Extraction of bibliographic items. Bibliographic items are the basic information in patent documents, including applicant, inventor, application date, publication date, etc. By extracting these key information from patent documents, the foundation is laid for subsequent identification of behavioral subjects and behavior analysis.

[0010] S6: Patent behavior extraction, including but not limited to patent cooperation application behavior, patent right transfer behavior and patent licensing behavior. These behavior data reflect the cooperation and interaction relationship between innovation entities;

[0011] S7: Behavior subject alignment: consistency check and alignment of the extracted behavior subjects to ensure the uniqueness and accuracy of each entity in the knowledge graph;

[0012] S8: Patent behavior knowledge graph construction: Based on the patent behavior data extracted and processed in the above steps, a patent behavior knowledge graph of the technology cluster is constructed; the knowledge graph displays the patent behavior relationship between innovation entities in the form of a graph, including cooperation, transfer and licensing, etc. This graphical representation method helps to intuitively understand the subject relationship and technology flow within the technology cluster;

[0013] S9: Intelligent identification of core entities, using indicators such as relationship strength and control strength to automatically identify core innovation entities within a technology cluster;

[0014] S10: Intelligent recommendation of potential collaborators, based on indicators such as the relationship level and path strength in the knowledge graph, to achieve intelligent recommendation of potential collaborators; by analyzing the cooperation history and potential cooperation opportunities between different entities, the system recommends potential collaborators and cooperation paths.

[0015] Starting from the perspective of enterprises, we first construct the IPC classification code co-occurrence matrix and co-occurrence network based on enterprise patent data, identify the core IPC classification code of enterprise patents through statistical characteristics, and classify and aggregate the core IPC classification code through factor analysis based on principal component analysis. When identifying enterprise technology clusters, the following steps are included:

[0016] S201: Collect patent data of target enterprises, including key information such as patent title, abstract, application date, IPC classification number, CPC classification number, etc. The data source can be international or domestic patent databases such as the State Intellectual Property Office, the European Patent Office, and the United States Patent and Trademark Office;

[0017] S202: Perform data cleaning to remove duplicate, erroneous or irrelevant records, standardize data formats, and ensure consistency and accuracy of key fields such as IPC and CPC classification numbers;

[0018] S203: IPC and CPC extraction, extracting IPC and CPC classification codes from each patent;

[0019] S204: Create a two-dimensional matrix, where rows and columns represent different IPC and CPC classification codes respectively;

[0020] S205: By traversing all patents, whenever two IPC / CPC classification codes appear in the same patent, the co-occurrence count is increased in the corresponding matrix position;

[0021] S206: Calculate the frequency of each IPC and CPC classification code, that is, how many patents the classification code appears in;

[0022] S207: Use statistical indicators such as centrality and entropy to evaluate the importance of each classification number;

[0023] S208: Set thresholds based on frequency, centrality or other statistical characteristics to filter out core IPC and CPC classification numbers, which represent the company's key focus and layout in the technology field.

[0024] After setting thresholds by frequency, centrality or other statistical characteristics and screening out the core IPC and CPC classification numbers, we use factors based on principal component analysis to analyze and interpret the technology clusters, including the following steps:

[0025] S301: Data standardization: standardize the co-occurrence matrix to eliminate the dimension effect;

[0026] S302: Applying principal component analysis (PCA) to reduce the dimension of the co-occurrence matrix of the core IPC and PC classification numbers to extract the main features;

[0027] S303: Determine the number of principal components, usually based on the size of eigenvalues, cumulative contribution rate and other criteria;

[0028] S304: Perform varimax factor rotation to make factor loadings clearer and easier to interpret;

[0029] S305: Assign the core IPC and CPC classification codes to different technology clusters based on factor loadings;

[0030] S306: Based on PCA results and factor analysis, the core IPC / CPC classification codes are aggregated into different technology clusters. Each technology cluster represents the core technology combination and research direction of the enterprise in this field.

[0031] S307: Analyze the characteristics, technological relevance and development trends of each technology cluster, and use visualization tools such as network diagrams and heat maps to display the structure and relationship of technology clusters;

[0032] S308: Provide strategic advice to enterprises on patent layout, technology research and development, and market expansion based on the results of technology cluster analysis;

[0033] S309: Verify the accuracy and rationality of technology cluster analysis results through expert consultation, literature comparison, etc.;

[0034] S310: Adjust analysis methods and parameters based on verification results, optimize technology cluster identification models, continuously collect new patent data, and conduct continuous analysis and updates to reflect the latest developments in the company's technology layout.

[0035] Through the additional information such as the address and postal code provided in the patent specification, we build entity consistency verification and alignment steps based on this information to deal with the problems caused by the change of corporate name and ensure the accuracy and consistency of patent behavior data. The entity consistency verification and alignment of the information includes the following steps:

[0036] S401: Collect patent data and obtain comprehensive bibliographic information including patent applicants, patent owners, patent transferors, assignees, patent licensees and licensees from the patent data platform;

[0037] S402: Remove duplicate, erroneous or incomplete data records;

[0038] S403: Entity recognition, using natural language processing technology (NLP) to identify key entities such as patent applicants, patent holders, addresses, and postal codes from patent documents;

[0039] S404: Relationship extraction, establishing relationships between entities, such as application-authorization relationships, transfer relationships, licensing relationships, etc.;

[0040] S405: Address and postal code matching, standardizing all identified addresses and postal codes, including formatting and correcting typos; using address and postal code information for preliminary matching to identify records that may belong to the same entity but do not directly correspond due to a change in corporate name;

[0041] S406: For situations where direct matching by address is not possible, a more complex entity alignment algorithm, Jaccard similarity algorithm and edit distance algorithm are used to further identify potential consistent entities;

[0042] S407: Establishing an enterprise name change mapping table based on enterprise name change records in historical data to assist entity alignment;

[0043] S408: Represent the identified entities and the relationships between them in the form of a graph structure, where nodes represent entities, such as enterprises and individuals, and edges represent relationships, such as patent applications, rights transfers, and licensed uses;

[0044] S409: Add attribute information to nodes and edges, such as patent application date, transfer price, license period, etc., to enrich the content of the knowledge graph;

[0045] S410: Check errors and inconsistencies in the knowledge graph through cross-validation, manual review, etc., and make corrections;

[0046] S411: Optimize graph query algorithms to improve query efficiency and accuracy of knowledge graphs;

[0047] S412: Based on the constructed knowledge graph, conduct patent behavior analysis, such as patent transfer activity, licensing trends, etc.;

[0048] S413: Regularly obtain new data from the patent data platform and update the knowledge graph to ensure the timeliness and accuracy of information;

[0049] S414: Establish a user feedback mechanism to continuously optimize and improve the knowledge graph construction process based on user feedback.

[0050] For patents with multiple transfers, multiple data updates and consistency checks are repeated in chronological order to achieve entity alignment. For one-to-many, many-to-one, many-to-many and other patent transfers or patent licensing behaviors, multiple entities are split before entity consistency checks and alignment, and multiple one-to-one relationships of patent right transfers or patent licensing are established in sequence. The consistency and alignment of the subject of the patent right transfer relationship includes the following steps:

[0051] S501: input patent information;

[0052] S502: Determine whether the transferor and the applicant are the same; if the determination result is yes, proceed to S503; if the determination result is no, proceed to S506;

[0053] S503: The transferor consistency judgment is passed and the process goes to S504;

[0054] S504: Determine whether the assignee and the right holder are the same; if the judgment structure is yes, proceed to S505; if the judgment structure is no, proceed to S508;

[0055] S505: The assignee passes the consistency test and the process ends;

[0056] S506: Determine whether the address or zip code is consistent. If yes, proceed to S507; if no, proceed to S510;

[0057] S507: Set the transferor to be equal to the applicant and proceed to S503;

[0058] S508: Determine whether the address or zip code is consistent; if the determination result is yes, proceed to S509; if the determination result is no, proceed to S510;

[0059] S509: Make the assignee equal to the right holder and proceed to S505;

[0060] S510: Issue an early warning and request manual verification.

[0061] Patent licensing relationship consistency and alignment include the following steps:

[0062] S601: input patent information;

[0063] S602: Determine whether there is a transfer relationship; if the determination result is yes, proceed to S603; if the determination result is no, proceed to S604;

[0064] S603: Determine whether the licensed use is prior to the transfer of rights; if the determination result is yes, proceed to S604; if the determination result is no, proceed to S606;

[0065] S604: Determine whether the licensor and the applicant are the same; if the determination result is yes, proceed to S605; if the determination result is no, proceed to S607;

[0066] S605: The licensor consistency judgment is passed and the process ends;

[0067] S606: Determine whether the licensor and the right holder are the same; if the determination result is yes, proceed to S605; if the determination result is no, proceed to S609;

[0068] S607: Determine whether the addresses are consistent; if the determination result is yes, proceed to S608; if the determination result is no, proceed to S611;

[0069] S608: Let the licensee be equal to the applicant, and proceed to S605;

[0070] S609: Determine whether the addresses are consistent. If yes, proceed to S610; if no, proceed to S611;

[0071] S610: Set the licensor to be equal to the right holder and proceed to S605;

[0072] S611: Issue an early warning and request manual verification.

[0073] The definition of innovation subject relationship strength is the number of patent behavior relationships between the innovation subject and other subjects. The core innovation subject identification is helpful to help the innovation subject determine the main competitors and partners in the subdivided field, assist the enterprise in the selection of patent partners, communication and negotiation cost allocation, including the following steps:

[0074] S701: Define nodes, which represent different innovation entities, i.e. enterprises, research institutions or individuals;

[0075] S702: Define edges, which represent patent behavior relationships between innovation entities, such as joint applications, citations, and transfers;

[0076] S703: Establish a relationship network: Based on the collected patent data, build a patent behavior relationship network between innovation entities;

[0077] S704: Use the graph database or the visualization tool Cytoscape to display the knowledge graph in a graphical form;

[0078] S705: Count the number of relationships. For each innovation subject, count the number of patent behavior relationships between it and other innovation subjects, that is, the number of edges connected to it;

[0079] S706: Normalize the number of relationships of each innovation subject, and the calculation formula is:

[0080]

[0081] Among them, k i represents the number of relationships connected to innovation subject i, N represents the number of innovation subjects in the knowledge graph,

[0082] Indicates the strength of the relationship;

[0083] S707: Through normalization, the results are explained. The innovation subject with stronger relationship strength has more external collaborators in the network and is more likely to act as an intermediary to promote cooperation between other innovation subjects.

[0084] S708: Sort all innovation subjects according to the calculated relationship strength; set a relationship strength threshold according to research needs or actual conditions to identify core innovation subjects; innovation subjects with relationship strength higher than the threshold are regarded as core innovation subjects;

[0085] S709: Based on the identification results of core innovation entities, analyze their competition and cooperation in the sub-sectors, regularly update patent data, reconstruct the knowledge graph and calculate the relationship strength to monitor the changes of core innovation entities and their impact on corporate strategies.

[0086] The control intensity of an innovation subject is defined as the proportion of the shortest paths that pass through the innovation subject among the shortest paths that connect all innovation subjects. The establishment of a relationship between two non-adjacent innovation subjects depends on the intermediary subjects on their connection paths. These intermediary subjects have a controlling and restrictive effect on the interaction between the two non-adjacent innovation subjects. The greater the control intensity of an innovation subject, the greater its voice and control in building cooperative relationships among different enterprises, and the more significant its role as an intermediary in the establishment of future cooperative relationships. The more cooperative enterprises should maintain a good relationship with it in order to build more benign cooperative relationships. The specific steps include:

[0087] S801: Node definition, defining each innovation subject as a node in the knowledge graph;

[0088] S802: Relationship definition: Based on the information such as joint application, citation, and transfer in the patent data, define the cooperative relationship between innovation entities and use it as an edge in the knowledge graph;

[0089] S803: Use graph database or graph processing software Neo4j and Gephi to build a knowledge graph containing the relationship between innovation entities and their patent behaviors;

[0090] S804: Calculate the control intensity of the innovation subject;

[0091] S805: Identify key intermediaries. Based on the calculated control intensity, identify the innovation entities that play important intermediary roles in different cooperative relationships;

[0092] S806: Based on the analysis of these key intermediaries, enterprises can formulate targeted cooperation strategies and give priority to establishing or maintaining good relationships with innovation entities with high control intensity to promote the establishment of more benign cooperative relationships;

[0093] S807: As patent data is continuously updated, the knowledge graph is rebuilt regularly and the control strength is calculated to dynamically monitor market changes and adjust cooperation strategies in a timely manner.

[0094] Calculating the control intensity of the innovation entity includes the following steps:

[0095] S901: Identify the shortest path. For any two non-adjacent innovation entities A and B in the knowledge graph, calculate all the shortest paths between them.

[0096] S902: Count the paths passing through the nodes. For each innovation subject i in the graph, count the number of paths passing through i among all the shortest paths connecting s and t.

[0097] S903: The total number of shortest paths connecting s and t is recorded as g st ;

[0098] S904: Calculate the control strength of innovation subject i on the relationship between s and t

[0099] This step implements normalization so that the value of the control intensity is between 0 and 1; the calculation formula is as follows:

[0100]

[0101] S905: Perform the above calculations on all possible pairs of non-adjacent innovation entities in the graph to obtain the control strength of each innovation entity in multiple relationships.

[0102] The relationship level is defined as the shortest path length required for the innovation subject to establish a connection with the potential collaborator. The distance between the potential collaborator and the innovation subject and the difficulty of relationship building are compared, including the following steps:

[0103] S1001: Identify patent behavior relationships, based on cleaned and standardized data, identify patent behavior relationships between innovation entities, such as assignment, licensing, joint application, etc.;

[0104] S1002: Construct a relationship graph. Using graph theory, take innovation subjects as nodes and patent behavior relationships as edges to construct a patent behavior relationship network. Each node in the network represents an innovation subject, and each edge represents the patent behavior relationship between two innovation subjects.

[0105] S1003: Calculate the shortest path, using the shortest path algorithm in graph theory, namely the Floyd-Warshall algorithm, to calculate the shortest path length between any two innovation entities in the network;

[0106] S1004: Define the relationship level, according to the formula PL ij =min(Length(p)),p∈P ij , taking the shortest path length as the relationship level, where P ijrepresents the set of all accessible paths between innovation subject i and innovation subject j, and Length(p) represents the length function of path p; if there is a direct patent behavior relationship between the two innovation subjects, their relationship level is 1, and they are first-level collaborators with each other; if there is no direct accessible path between the two innovation subjects, the shortest path length between them is defined as +∞, and the relationship level is +∞, indicating that they cannot be directly connected through the existing relationship, and they are potential collaborators with ∞ level each other;

[0107] S1005: List potential collaborators. Based on the patent behavior relationship network, list all innovation entities that have potential cooperation relationships with the innovation entity as a list of potential collaborators;

[0108] S1006: Calculate the relationship level. For each potential collaborator in the list, calculate the relationship level between it and the target innovation subject, that is, the length of the shortest path between them;

[0109] S1007: Sort by relationship level, sort potential partners from small to large according to the relationship level; the smaller the relationship level, the greater the possibility of cooperation, and the lower the communication and negotiation costs;

[0110] S1008: Generate a sorted list, generating a sorted list of potential collaborators, the list reflecting the priority of the possibility of cooperation;

[0111] S1009: Identify collaborators with the same relationship level, and in the sorted list, identify a subset of potential collaborators with the same relationship level;

[0112] S1010: Calculate path strength. For each subset of collaborators with the same relationship level, calculate the path strength between the target innovation subject and each potential collaborator in the subset. Path strength can be defined as a measure of all possible paths connecting two nodes, such as path diversity, i.e. the number of different paths, the average importance of nodes on the path, i.e. the weighted sum based on indicators such as node degree and centrality, or path stability indicators, i.e. the stability assessment of the cooperation history between nodes on the path.

[0113] S1011: Sort by path strength. In the subset of collaborators with the same relationship level, potential collaborators are sorted from large to small according to path strength. The greater the path strength, the higher the stability of the relationship between the innovation subjects, the stronger the willingness to cooperate, and the more likely it is to form a long-term and stable cooperative relationship.

[0114] S1012: Combine the ranking results of relationship level and path strength to comprehensively evaluate potential partners; give priority to potential partners with small relationship level and large path strength as priority cooperation objects;

[0115] S1013: Based on the comprehensive evaluation results, formulate specific cooperation strategies for innovation entities, including cooperation priorities, cooperation methods, cooperation content, etc.

[0116] Beneficial effects of the present invention:

[0117] 1. Compared with the existing single patent behavior network analysis research, which ignores the layout and attention differences of various innovation entities in the subdivided technology field, the existence of industrial noise makes the core subject identification and potential collaborator recommendation based on industrial data deviate from the actual situation of the subdivided field that enterprises pay attention to, and it is difficult to meet the actual technological innovation needs of innovation entities. Based on the cluster analysis of patent technology layout, this method analyzes the patent behavior relationship of innovation entities in the technology cluster by constructing a patent behavior knowledge map including multiple relationships such as patent cooperation application, patent right transfer, and patent licensing, and constructs a method for automatic identification of core subjects of patent behavior and intelligent recommendation of potential collaborators to promote the coordinated and high-quality development of innovation entities;

[0118] 2. Cluster analysis of patent layout of science and technology innovation enterprises in subdivided fields is conducted by constructing the co-occurrence matrix of IPC classification numbers of enterprise patents and factor analysis; after proposing the entity alignment method of patent behavior subjects, a patent behavior knowledge map including multiple relationships of patent cooperation application, patent right transfer, and patent licensing is constructed based on the patent data of technology clusters; the core innovation subjects are identified through relationship strength and control strength, and potential collaborators are recommended based on relationship level and path strength; intelligence functions such as evaluation of intellectual property rights / innovation activity of innovation subjects, market competitiveness measurement and recommendation of potential collaborators are realized

[0119] 3. Through the construction of patent behavior knowledge graph, a patent behavior knowledge graph of technology cluster is constructed based on the patent behavior data extracted and processed in the above steps; the knowledge graph displays the patent behavior relationship between innovation entities in the form of a graph, including cooperation, transfer and licensing; this graphical representation method helps to intuitively understand the subject relationship and technology flow within the technology cluster. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 What is shown is a schematic diagram of a research framework for constructing a patent behavior knowledge graph based on the perspective of an innovation subject, a patent behavior knowledge graph construction and analysis method of the present invention;

[0121] Figure 2 What is shown is a schematic diagram of the subject consistency and alignment flow chart of the patent rights transfer relationship of the patent behavior knowledge graph construction analysis method based on the innovation subject perspective of the present invention;

[0122] Figure 3Shown is a schematic diagram of the patent licensing relationship subject consistency and alignment flow chart of the patent behavior knowledge graph construction analysis method based on the innovation subject perspective of the present invention. DETAILED DESCRIPTION

[0123] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0124] See also Figure 1-3 The present invention provides an embodiment: a patent behavior knowledge graph construction and analysis method based on the innovation subject perspective:

[0125] S1: To obtain enterprise patent data, it is first necessary to identify the target enterprises and collect their patent data, which usually involves retrieving and downloading patent documents of relevant enterprises from patent databases, such as the China National Intellectual Property Administration database, the European Patent Office database, and the United States Patent and Trademark Office database; data collection covers a sufficient time range to fully reflect the technological innovation process of the enterprise, and also covers detailed information on the patent, including but not limited to patent number, application date, publication date, applicant, IPC classification number, etc.;

[0126] S2: Identification of core IPC classification codes: Analyze the collected patent data to identify the IPC classification codes most commonly used by the enterprise. These classification codes represent the core technology areas of the enterprise. By counting the frequency of occurrence of each classification code, it is possible to determine which classification codes best represent the enterprise's technology focus.

[0127] S3: Technology cluster analysis based on IPC classification numbers. After determining the core IPC classification numbers, use these classification numbers to perform cluster analysis on patents to identify different technology clusters. Cluster analysis can classify similar patents together to form technology groups. Each group represents a specific technology field or technology direction to reveal the technology layout and key development direction within the enterprise.

[0128] S4: Obtain patent data of technology clusters. Based on the results of technology cluster analysis, conduct secondary data collection for each technology cluster to obtain more detailed and in-depth patent data for subsequent patent behavior analysis and knowledge graph construction;

[0129] S5: Extraction of bibliographic items. Bibliographic items are the basic information in patent documents, including applicant, inventor, application date, publication date, etc. By extracting these key information from patent documents, the foundation is laid for subsequent identification of behavioral subjects and behavior analysis.

[0130] S6: Patent behavior extraction, including but not limited to patent cooperation application behavior, patent right transfer behavior and patent licensing behavior. These behavior data reflect the cooperation and interaction relationship between innovation entities;

[0131] S7: Behavior subject alignment: consistency check and alignment of the extracted behavior subjects to ensure the uniqueness and accuracy of each entity in the knowledge graph;

[0132] S8: Patent behavior knowledge graph construction: Based on the patent behavior data extracted and processed in the above steps, a patent behavior knowledge graph of the technology cluster is constructed; the knowledge graph displays the patent behavior relationship between innovation entities in the form of a graph, including cooperation, transfer and licensing, etc. This graphical representation method helps to intuitively understand the subject relationship and technology flow within the technology cluster;

[0133] S9: Intelligent identification of core entities, using indicators such as relationship strength and control strength to automatically identify core innovation entities within a technology cluster;

[0134] S10: Intelligent recommendation of potential collaborators, based on indicators such as the relationship level and path strength in the knowledge graph, to achieve intelligent recommendation of potential collaborators; by analyzing the cooperation history and potential cooperation opportunities between different entities, the system recommends potential collaborators and cooperation paths.

[0135] Starting from the perspective of enterprises, we first construct the IPC classification code co-occurrence matrix and co-occurrence network based on enterprise patent data, identify the core IPC classification code of enterprise patents through statistical characteristics, and classify and aggregate the core IPC classification code through factor analysis based on principal component analysis. When identifying enterprise technology clusters, the following steps are included:

[0136] S201: Collect patent data of target enterprises, including key information such as patent title, abstract, application date, IPC classification number, CPC classification number, etc. The data source can be international or domestic patent databases such as the State Intellectual Property Office, the European Patent Office, and the United States Patent and Trademark Office;

[0137] S202: Perform data cleaning to remove duplicate, erroneous or irrelevant records, standardize data formats, and ensure consistency and accuracy of key fields such as IPC and CPC classification numbers;

[0138] S203: IPC and CPC extraction, extracting IPC and CPC classification codes from each patent;

[0139] S204: Create a two-dimensional matrix, where rows and columns represent different IPC and CPC classification codes respectively;

[0140] S205: By traversing all patents, whenever two IPC / CPC classification codes appear in the same patent, the co-occurrence count is increased in the corresponding matrix position;

[0141] S206: Calculate the frequency of each IPC and CPC classification code, that is, how many patents the classification code appears in;

[0142] S207: Use statistical indicators such as centrality and entropy to evaluate the importance of each classification number;

[0143] S208: Set thresholds based on frequency, centrality or other statistical characteristics to filter out core IPC and CPC classification numbers, which represent the company's key focus and layout in the technology field.

[0144] After setting thresholds by frequency, centrality or other statistical characteristics and screening out the core IPC and CPC classification numbers, we use factors based on principal component analysis to analyze and interpret the technology clusters, including the following steps:

[0145] S301: Data standardization: standardize the co-occurrence matrix to eliminate the dimension effect;

[0146] S302: Applying principal component analysis (PCA) to reduce the dimension of the co-occurrence matrix of the core IPC and PC classification numbers to extract the main features;

[0147] S303: Determine the number of principal components, usually based on the size of eigenvalues, cumulative contribution rate and other criteria;

[0148] S304: Perform varimax factor rotation to make factor loadings clearer and easier to interpret;

[0149] S305: Assign the core IPC and CPC classification codes to different technology clusters based on factor loadings;

[0150] S306: Based on PCA results and factor analysis, the core IPC / CPC classification codes are aggregated into different technology clusters. Each technology cluster represents the core technology combination and research direction of the enterprise in this field.

[0151] S307: Analyze the characteristics, technological relevance and development trends of each technology cluster, and use visualization tools such as network diagrams and heat maps to display the structure and relationship of technology clusters;

[0152] S308: Provide strategic advice to enterprises on patent layout, technology research and development, and market expansion based on the results of technology cluster analysis;

[0153] S309: Verify the accuracy and rationality of technology cluster analysis results through expert consultation, literature comparison, etc.;

[0154] S310: Adjust analysis methods and parameters based on verification results, optimize technology cluster identification models, continuously collect new patent data, and conduct continuous analysis and updates to reflect the latest developments in the company's technology layout.

[0155] Through the additional information such as the address and postal code provided in the patent specification, we build entity consistency verification and alignment steps based on this information to deal with the problems caused by the change of corporate name and ensure the accuracy and consistency of patent behavior data. The entity consistency verification and alignment of the information includes the following steps:

[0156] S401: Collect patent data and obtain comprehensive bibliographic information including patent applicants, patent owners, patent transferors, assignees, patent licensees and licensees from the patent data platform;

[0157] S402: Remove duplicate, erroneous or incomplete data records;

[0158] S403: Entity recognition, using natural language processing technology (NLP) to identify key entities such as patent applicants, patent holders, addresses, and postal codes from patent documents;

[0159] S404: Relationship extraction, establishing relationships between entities, such as application-authorization relationships, transfer relationships, licensing relationships, etc.;

[0160] S405: Address and postal code matching, standardizing all identified addresses and postal codes, including formatting and correcting typos; using address and postal code information for preliminary matching to identify records that may belong to the same entity but do not directly correspond due to a change in corporate name;

[0161] S406: For situations where direct matching by address is not possible, a more complex entity alignment algorithm, Jaccard similarity algorithm and edit distance algorithm are used to further identify potential consistent entities;

[0162] S407: Establishing an enterprise name change mapping table based on enterprise name change records in historical data to assist entity alignment;

[0163] S408: Represent the identified entities and the relationships between them in the form of a graph structure, where nodes represent entities, such as enterprises and individuals, and edges represent relationships, such as patent applications, rights transfers, and licensed uses;

[0164] S409: Add attribute information to nodes and edges, such as patent application date, transfer price, license period, etc., to enrich the content of the knowledge graph;

[0165] S410: Check errors and inconsistencies in the knowledge graph through cross-validation, manual review, etc., and make corrections;

[0166] S411: Optimize graph query algorithms to improve query efficiency and accuracy of knowledge graphs;

[0167] S412: Based on the constructed knowledge graph, conduct patent behavior analysis, such as patent transfer activity, licensing trends, etc.;

[0168] S413: Regularly obtain new data from the patent data platform and update the knowledge graph to ensure the timeliness and accuracy of information;

[0169] S414: Establish a user feedback mechanism to continuously optimize and improve the knowledge graph construction process based on user feedback.

[0170] For patents with multiple transfers, multiple data updates and consistency checks are repeated in chronological order to achieve entity alignment. For one-to-many, many-to-one, many-to-many and other patent transfers or patent licensing behaviors, multiple entities are split before entity consistency checks and alignment, and multiple one-to-one relationships of patent right transfers or patent licensing are established in sequence. The consistency and alignment of the subject of the patent right transfer relationship includes the following steps:

[0171] S501: input patent information;

[0172] S502: Determine whether the transferor and the applicant are the same; if the determination result is yes, proceed to S503; if the determination result is no, proceed to S506;

[0173] S503: The transferor consistency judgment is passed and the process goes to S504;

[0174] S504: Determine whether the assignee and the right holder are the same; if the judgment structure is yes, proceed to S505; if the judgment structure is no, proceed to S508;

[0175] S505: The assignee passes the consistency test and the process ends;

[0176] S506: Determine whether the address or zip code is consistent. If yes, proceed to S507; if no, proceed to S510;

[0177] S507: Set the transferor to be equal to the applicant and proceed to S503;

[0178] S508: Determine whether the address or zip code is consistent; if the determination result is yes, proceed to S509; if the determination result is no, proceed to S510;

[0179] S509: Make the assignee equal to the right holder and proceed to S505;

[0180] S510: Issue an early warning and request manual verification.

[0181] Patent licensing relationship consistency and alignment include the following steps:

[0182] S601: input patent information;

[0183] S602: Determine whether there is a transfer relationship; if the determination result is yes, proceed to S603; if the determination result is no, proceed to S604;

[0184] S603: Determine whether the licensed use is prior to the transfer of rights; if the determination result is yes, proceed to S604; if the determination result is no, proceed to S606;

[0185] S604: Determine whether the licensor and the applicant are the same; if the determination result is yes, proceed to S605; if the determination result is no, proceed to S607;

[0186] S605: The licensor consistency judgment is passed and the process ends;

[0187] S606: Determine whether the licensor and the right holder are the same; if the determination result is yes, proceed to S605; if the determination result is no, proceed to S609;

[0188] S607: Determine whether the addresses are consistent; if the determination result is yes, proceed to S608; if the determination result is no, proceed to S611;

[0189] S608: Let the licensee be equal to the applicant, and proceed to S605;

[0190] S609: Determine whether the addresses are consistent. If yes, proceed to S610; if no, proceed to S611;

[0191] S610: Set the licensor to be equal to the right holder and proceed to S605;

[0192] S611: Issue an early warning and request manual verification.

[0193] The definition of innovation subject relationship strength is the number of patent behavior relationships between the innovation subject and other subjects. The core innovation subject identification is helpful to help the innovation subject determine the main competitors and partners in the subdivided field, assist the enterprise in the selection of patent partners, communication and negotiation cost allocation, including the following steps:

[0194] S701: Define nodes, which represent different innovation entities, i.e. enterprises, research institutions or individuals;

[0195] S702: Define edges, which represent patent behavior relationships between innovation entities, such as joint applications, citations, and transfers;

[0196] S703: Establish a relationship network: Based on the collected patent data, build a patent behavior relationship network between innovation entities;

[0197] S704: Use the graph database or the visualization tool Cytoscape to display the knowledge graph in a graphical form;

[0198] S705: Count the number of relationships. For each innovation subject, count the number of patent behavior relationships between it and other innovation subjects, that is, the number of edges connected to it;

[0199] S706: Normalize the number of relationships of each innovation subject, and the calculation formula is:

[0200]

[0201] Among them, k i represents the number of relationships connected to innovation subject i, N represents the number of innovation subjects in the knowledge graph,

[0202] Indicates the strength of the relationship;

[0203] S707: Through normalization, the results are explained. The innovation subject with stronger relationship strength has more external collaborators in the network and is more likely to act as an intermediary to promote cooperation between other innovation subjects.

[0204] S708: Sort all innovation subjects according to the calculated relationship strength; set a relationship strength threshold according to research needs or actual conditions to identify core innovation subjects; innovation subjects with relationship strength higher than the threshold are regarded as core innovation subjects;

[0205] S709: Based on the identification results of core innovation entities, analyze their competition and cooperation in the sub-sectors, regularly update patent data, reconstruct the knowledge graph and calculate the relationship strength to monitor the changes of core innovation entities and their impact on corporate strategies.

[0206] The control intensity of an innovation subject is defined as the proportion of the shortest paths that pass through the innovation subject among the shortest paths that connect all innovation subjects. The establishment of a relationship between two non-adjacent innovation subjects depends on the intermediary subjects on their connection paths. These intermediary subjects have a controlling and restrictive effect on the interaction between the two non-adjacent innovation subjects. The greater the control intensity of an innovation subject, the greater its voice and control in building cooperative relationships among different enterprises, and the more significant its role as an intermediary in the establishment of future cooperative relationships. The more cooperative enterprises should maintain a good relationship with it in order to build more benign cooperative relationships. The specific steps include:

[0207] S801: Node definition, defining each innovation subject as a node in the knowledge graph;

[0208] S802: Relationship definition: Based on the information such as joint application, citation, and transfer in the patent data, define the cooperative relationship between innovation entities and use it as an edge in the knowledge graph;

[0209] S803: Use graph database or graph processing software Neo4j and Gephi to build a knowledge graph containing the relationship between innovation entities and their patent behaviors;

[0210] S804: Calculate the control intensity of the innovation subject;

[0211] S805: Identify key intermediaries. Based on the calculated control intensity, identify the innovation entities that play important intermediary roles in different cooperative relationships;

[0212] S806: Based on the analysis of these key intermediaries, enterprises can formulate targeted cooperation strategies and give priority to establishing or maintaining good relationships with innovation entities with high control intensity to promote the establishment of more benign cooperative relationships;

[0213] S807: As patent data is continuously updated, the knowledge graph is rebuilt regularly and the control strength is calculated to dynamically monitor market changes and adjust cooperation strategies in a timely manner.

[0214] Calculating the control intensity of the innovation entity includes the following steps:

[0215] S901: Identify the shortest path. For any two non-adjacent innovation entities A and B in the knowledge graph, calculate all the shortest paths between them.

[0216] S902: Count the paths passing through the nodes. For each innovation subject i in the graph, count the number of paths passing through i among all the shortest paths connecting s and t.

[0217] S903: The total number of shortest paths connecting s and t is recorded as g st ;

[0218] S904: Calculate the control strength of innovation subject i on the relationship between s and t

[0219] This step implements normalization so that the value of the control intensity is between 0 and 1; the calculation formula is as follows:

[0220]

[0221] S905: Perform the above calculations on all possible pairs of non-adjacent innovation entities in the graph to obtain the control strength of each innovation entity in multiple relationships.

[0222] The relationship level is defined as the shortest path length required for the innovation subject to establish a connection with the potential collaborator. The distance between the potential collaborator and the innovation subject and the difficulty of relationship building are compared, including the following steps:

[0223] S1001: Identify patent behavior relationships, based on cleaned and standardized data, identify patent behavior relationships between innovation entities, such as assignment, licensing, joint application, etc.;

[0224] S1002: Construct a relationship graph. Using graph theory, take innovation subjects as nodes and patent behavior relationships as edges to construct a patent behavior relationship network. Each node in the network represents an innovation subject, and each edge represents the patent behavior relationship between two innovation subjects.

[0225] S1003: Calculate the shortest path, using the shortest path algorithm in graph theory, namely the Floyd-Warshall algorithm, to calculate the shortest path length between any two innovation entities in the network;

[0226] S1004: Define the relationship level, according to the formula PL ij =min(Length(p)),p∈P ij , taking the shortest path length as the relationship level, where P ij represents the set of all accessible paths between innovation subject i and innovation subject j, and Length(p) represents the length function of path p; if there is a direct patent behavior relationship between the two innovation subjects, their relationship level is 1, and they are first-level collaborators with each other; if there is no direct accessible path between the two innovation subjects, the shortest path length between them is defined as +∞, and the relationship level is +∞, indicating that they cannot be directly connected through the existing relationship, and they are potential collaborators with ∞ level each other;

[0227] S1005: List potential collaborators. Based on the patent behavior relationship network, list all innovation entities that have potential cooperation relationships with the innovation entity as a list of potential collaborators;

[0228] S1006: Calculate the relationship level. For each potential collaborator in the list, calculate the relationship level between it and the target innovation subject, that is, the length of the shortest path between them;

[0229] S1007: Sort by relationship level, sort potential partners from small to large according to the relationship level; the smaller the relationship level, the greater the possibility of cooperation, and the lower the communication and negotiation costs;

[0230] S1008: Generate a sorted list, generating a sorted list of potential collaborators, the list reflecting the priority of the possibility of cooperation;

[0231] S1009: Identify collaborators with the same relationship level, and in the sorted list, identify a subset of potential collaborators with the same relationship level;

[0232] S1010: Calculate path strength. For each subset of collaborators with the same relationship level, calculate the path strength between the target innovation subject and each potential collaborator in the subset. Path strength can be defined as a measure of all possible paths connecting two nodes, such as path diversity, i.e. the number of different paths, the average importance of nodes on the path, i.e. the weighted sum based on indicators such as node degree and centrality, or path stability indicators, i.e. the stability assessment of the cooperation history between nodes on the path.

[0233] S1011: Sort by path strength. In the subset of collaborators with the same relationship level, potential collaborators are sorted from large to small according to path strength. The greater the path strength, the higher the stability of the relationship between the innovation subjects, the stronger the willingness to cooperate, and the more likely it is to form a long-term and stable cooperative relationship.

[0234] S1012: Combine the ranking results of relationship level and path strength to comprehensively evaluate potential partners; give priority to potential partners with small relationship level and large path strength as priority cooperation objects;

[0235] S1013: Based on the comprehensive evaluation results, formulate specific cooperation strategies for innovation entities, including cooperation priorities, cooperation methods, cooperation content, etc.

[0236] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.

Claims

1. A patent behavior knowledge graph construction and analysis method based on the innovation subject perspective, characterized by: The following steps are included: S1: To obtain enterprise patent data, it is first necessary to identify the target enterprises and collect their patent data, which usually involves retrieving and downloading patent documents of relevant enterprises from patent databases, such as the China National Intellectual Property Administration database, the European Patent Office database, and the United States Patent and Trademark Office database; data collection covers a sufficient time range to fully reflect the technological innovation process of the enterprise, and also covers detailed information on the patent, including but not limited to patent number, application date, publication date, applicant, IPC classification number, etc.; S2: Identification of core IPC classification codes: Analyze the collected patent data to identify the IPC classification codes most commonly used by the enterprise. These classification codes represent the core technology areas of the enterprise. By counting the frequency of occurrence of each classification code, it is possible to determine which classification codes best represent the enterprise's technology focus. S3: Technology cluster analysis based on IPC classification numbers. After determining the core IPC classification numbers, these classification numbers are used to perform cluster analysis on patents to identify different technology clusters. Cluster analysis can group similar patents together to form technology groups. Each group represents a specific technology field or technology direction to reveal the technology layout and key development direction within the enterprise. S4: Obtain patent data of technology clusters. Based on the results of technology cluster analysis, conduct secondary data collection for each technology cluster to obtain more detailed and in-depth patent data for subsequent patent behavior analysis and knowledge graph construction; S5: Extraction of bibliographic items. Bibliographic items are the basic information in patent documents, including applicant, inventor, application date, publication date, etc. By extracting these key information from patent documents, the foundation is laid for subsequent identification of behavioral subjects and behavior analysis. S6: Patent behavior extraction, including but not limited to patent cooperation application behavior, patent right transfer behavior and patent licensing behavior. These behavior data reflect the cooperation and interaction relationship between innovation entities; S7: Behavior subject alignment: consistency check and alignment of the extracted behavior subjects to ensure the uniqueness and accuracy of each entity in the knowledge graph; S8: Patent behavior knowledge graph construction: Based on the patent behavior data extracted and processed in the above steps, a patent behavior knowledge graph of the technology cluster is constructed; the knowledge graph displays the patent behavior relationship between innovation entities in the form of a graph, including cooperation, transfer and licensing, etc. This graphical representation method helps to intuitively understand the subject relationship and technology flow within the technology cluster; S9: Intelligent identification of core entities, using indicators such as relationship strength and control strength to automatically identify core innovation entities within a technology cluster; S10: Intelligent recommendation of potential collaborators, based on indicators such as the relationship level and path strength in the knowledge graph, to achieve intelligent recommendation of potential collaborators; by analyzing the cooperation history and potential cooperation opportunities between different entities, the system recommends potential collaborators and cooperation paths.

2. According to claim 1, a patent behavior knowledge graph construction and analysis method based on the innovation subject perspective is characterized by: Starting from the perspective of enterprises, we first construct the IPC classification code co-occurrence matrix and co-occurrence network based on enterprise patent data, identify the core IPC classification code of enterprise patents through statistical characteristics, and classify and aggregate the core IPC classification code through factor analysis based on principal component analysis. When identifying enterprise technology clusters, the following steps are included: S201: Collect patent data of target enterprises, including key information such as patent title, abstract, application date, IPC classification number, CPC classification number, etc. The data source can be international or domestic patent databases such as the State Intellectual Property Office, the European Patent Office, and the United States Patent and Trademark Office; S202: Perform data cleaning to remove duplicate, erroneous or irrelevant records, standardize data formats, and ensure consistency and accuracy of key fields such as IPC and CPC classification numbers; S203: IPC and CPC extraction, extracting IPC and CPC classification codes from each patent; S204: Create a two-dimensional matrix, where rows and columns represent different IPC and CPC classification codes respectively; S205: By traversing all patents, whenever two IPC / CPC classification codes appear in the same patent, the co-occurrence count is increased in the corresponding matrix position; S206: Calculate the frequency of each IPC and CPC classification code, that is, how many patents the classification code appears in; S207: Use statistical indicators such as centrality and entropy to evaluate the importance of each classification number; S208: Set thresholds based on frequency, centrality or other statistical characteristics to filter out core IPC and CPC classification numbers, which represent the company's key focus and layout in the technology field.

3. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: After setting thresholds by frequency, centrality or other statistical characteristics and screening out the core IPC and CPC classification numbers, we use factors based on principal component analysis to analyze and interpret the technology clusters, including the following steps: S301: Data standardization: standardize the co-occurrence matrix to eliminate the dimension effect; S302: Applying principal component analysis (PCA) to reduce the dimension of the co-occurrence matrix of the core IPC and PC classification numbers to extract the main features; S303: Determine the number of principal components, usually based on the size of eigenvalues, cumulative contribution rate and other criteria; S304: Perform varimax factor rotation to make factor loadings clearer and easier to interpret; S305: Assign the core IPC and CPC classification codes to different technology clusters based on factor loadings; S306: Based on PCA results and factor analysis, the core IPC / CPC classification codes are aggregated into different technology clusters. Each technology cluster represents the core technology combination and research direction of the enterprise in this field. S307: Analyze the characteristics, technological relevance and development trends of each technology cluster, and use visualization tools such as network diagrams and heat maps to display the structure and relationship of technology clusters; S308: Provide strategic advice to enterprises on patent layout, technology research and development, and market expansion based on the results of technology cluster analysis; S309: Verify the accuracy and rationality of technology cluster analysis results through expert consultation, literature comparison, etc.; S310: Adjust analysis methods and parameters based on verification results, optimize technology cluster identification models, continuously collect new patent data, and conduct continuous analysis and updates to reflect the latest developments in the company's technology layout.

4. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: Through the additional information such as the address and postal code provided in the patent specification, we build entity consistency verification and alignment steps based on this information to deal with the problems caused by the change of corporate name and ensure the accuracy and consistency of patent behavior data. The entity consistency verification and alignment of the information includes the following steps: S401: Collect patent data and obtain comprehensive bibliographic information including patent applicants, patent owners, patent transferors, assignees, patent licensees and licensees from the patent data platform; S402: Remove duplicate, erroneous or incomplete data records; S403: Entity recognition, using natural language processing technology (NLP) to identify key entities such as patent applicants, patent holders, addresses, and postal codes from patent documents; S404: Relationship extraction, establishing relationships between entities, such as application-authorization relationships, transfer relationships, licensing relationships, etc.; S405: Address and postal code matching, standardizing all identified addresses and postal codes, including formatting and correcting typos; using address and postal code information for preliminary matching to identify records that may belong to the same entity but do not directly correspond due to a change in corporate name; S406: For situations where direct matching by address is not possible, a more complex entity alignment algorithm, Jaccard similarity algorithm and edit distance algorithm are used to further identify potential consistent entities; S407: Establishing an enterprise name change mapping table based on enterprise name change records in historical data to assist entity alignment; S408: Represent the identified entities and the relationships between them in the form of a graph structure, where nodes represent entities, such as enterprises and individuals, and edges represent relationships, such as patent applications, rights transfers, and licensed uses; S409: Add attribute information to nodes and edges, such as patent application date, transfer price, license period, etc., to enrich the content of the knowledge graph; S410: Check errors and inconsistencies in the knowledge graph through cross-validation, manual review, etc., and make corrections; S411: Optimize graph query algorithms to improve query efficiency and accuracy of knowledge graphs; S412: Based on the constructed knowledge graph, conduct patent behavior analysis, such as patent transfer activity, licensing trends, etc.; S413: Regularly obtain new data from the patent data platform and update the knowledge graph to ensure the timeliness and accuracy of information; S414: Establish a user feedback mechanism to continuously optimize and improve the knowledge graph construction process based on user feedback.

5. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: For patents with multiple transfers, multiple data updates and consistency checks are repeated in chronological order to achieve entity alignment. For one-to-many, many-to-one, many-to-many and other patent transfers or patent licensing behaviors, multiple entities are split before entity consistency checks and alignment, and multiple one-to-one relationships of patent right transfers or patent licensing are established in sequence. The consistency and alignment of the subject of the patent right transfer relationship includes the following steps: S501: input patent information; S502: Determine whether the transferor and the applicant are the same; if the determination result is yes, proceed to S503; if the determination result is no, proceed to S506; S503: The transferor consistency judgment is passed and the process goes to S504; S504: Determine whether the assignee and the right holder are the same; if the judgment structure is yes, proceed to S505; if the judgment structure is no, proceed to S508; S505: The assignee passes the consistency test and the process ends; S506: Determine whether the address or zip code is consistent. If yes, proceed to S507; if no, proceed to S510; S507: Set the transferor to be equal to the applicant and proceed to S503; S508: Determine whether the address or zip code is consistent; if the determination result is yes, proceed to S509; if the determination result is no, proceed to S510; S509: Make the assignee equal to the right holder and proceed to S505; S510: Issue an early warning and request manual verification.

6. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 5 is characterized by: Patent licensing relationship consistency and alignment include the following steps: S601: input patent information; S602: Determine whether there is a transfer relationship; If the judgment result is yes, go to S603; if the judgment result is no, go to S604; S603: Determine whether the licensed use precedes the transfer of rights; If the judgment result is yes, go to S604; if the judgment result is no, go to S606; S604: Determine whether the licensor and the applicant are the same; If the judgment result is yes, go to S605; if the judgment result is no, go to S607; S605: The licensor consistency judgment is passed and the process ends; S606: Determine whether the licensor and the right holder are the same; If the judgment result is yes, go to S605; if the judgment result is no, go to S609; S607: Determine whether the addresses are consistent; If the judgment result is yes, go to S608; if the judgment result is no, go to S611; S608: Let the licensee be equal to the applicant, and proceed to S605; S609: Determine whether the addresses are consistent. If yes, proceed to S610; if no, proceed to S611; S610: Set the licensor to be equal to the right holder and proceed to S605; S611: Issue an early warning and request manual verification.

7. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: The definition of innovation subject relationship strength is the number of patent behavior relationships between the innovation subject and other subjects. The core innovation subject identification is helpful to help the innovation subject determine the main competitors and partners in the subdivided field, assist the enterprise in the selection of patent partners, communication and negotiation cost allocation, including the following steps: S701: Define nodes, which represent different innovation entities, i.e. enterprises, research institutions or individuals; S702: Define edges, which represent patent behavior relationships between innovation entities, such as joint applications, citations, and transfers; S703: Establish a relationship network: Based on the collected patent data, build a patent behavior relationship network between innovation entities; S704: Use the graph database or the visualization tool Cytoscape to display the knowledge graph in a graphical form; S705: Count the number of relationships. For each innovation subject, count the number of patent behavior relationships between it and other innovation subjects, that is, the number of edges connected to it; S706: Normalize the number of relationships of each innovation subject, and the calculation formula is: Among them, k i represents the number of relationships connected to innovation subject i, N represents the number of innovation subjects in the knowledge graph, and RD i Indicates the strength of the relationship; S707: Through normalization, the results are explained. The innovation subject with stronger relationship strength has more external collaborators in the network and is more likely to act as an intermediary to promote cooperation between other innovation subjects. S708: Sort all innovation subjects according to the calculated relationship strength; set a relationship strength threshold according to research needs or actual conditions to identify core innovation subjects; innovation subjects with relationship strength higher than the threshold are regarded as core innovation subjects; S709: Based on the identification results of core innovation entities, analyze their competition and cooperation in the sub-sectors, regularly update patent data, reconstruct the knowledge graph and calculate the relationship strength to monitor the changes of core innovation entities and their impact on corporate strategies.

8. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: The control intensity of an innovation subject is defined as the proportion of the shortest paths that pass through the innovation subject among the shortest paths that connect all innovation subjects. The establishment of a relationship between two non-adjacent innovation subjects depends on the intermediary subjects on their connection paths. These intermediary subjects have a controlling and restrictive effect on the interaction between the two non-adjacent innovation subjects. The greater the control intensity of an innovation subject, the greater its voice and control in building cooperative relationships among different enterprises, and the more significant its role as an intermediary in the establishment of future cooperative relationships. The more cooperative enterprises should maintain a good relationship with it in order to build more benign cooperative relationships. The specific steps include: S801: Node definition, defining each innovation subject as a node in the knowledge graph; S802: Relationship definition: Based on the information such as joint application, citation, and transfer in the patent data, define the cooperative relationship between innovation entities and use it as an edge in the knowledge graph; S803: Use graph database or graph processing software Neo4j and Gephi to build a knowledge graph containing the relationship between innovation entities and their patent behaviors; S804: Calculate the control intensity of the innovation subject; S805: Identify key intermediaries. Based on the calculated control intensity, identify the innovation entities that play important intermediary roles in different cooperative relationships; S806: Based on the analysis of these key intermediaries, enterprises can formulate targeted cooperation strategies and give priority to establishing or maintaining good relationships with innovation entities with high control intensity to promote the establishment of more benign cooperative relationships; S807: As patent data is continuously updated, the knowledge graph is rebuilt regularly and the control strength is calculated to dynamically monitor market changes and adjust cooperation strategies in a timely manner.

9. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: Calculating the control intensity of the innovation entity includes the following steps: S901: Identify the shortest path. For any two non-adjacent innovation entities A and B in the knowledge graph, calculate all the shortest paths between them. S902: Count the paths passing through the nodes. For each innovation subject i in the graph, count the number of paths passing through i among all the shortest paths connecting s and t. S903: The total number of shortest paths connecting s and t is recorded as g st ; S904: Calculate the control strength CB of the innovation subject i on the relationship between s and t i , this step realizes the normalization processing, so that the value of the control intensity is between 0 and 1; the calculation formula is as follows: S905: Perform the above calculations on all possible pairs of non-adjacent innovation entities in the graph to obtain the control strength of each innovation entity in multiple relationships.

10. The patent behavior knowledge graph construction and analysis method based on the innovation subject perspective according to claim 1 is characterized by: The relationship level is defined as the shortest path length required for the innovation subject to establish a connection with the potential collaborator. The distance between the potential collaborator and the innovation subject and the difficulty of relationship building are compared, including the following steps: S1001: Identify patent behavior relationships, based on cleaned and standardized data, identify patent behavior relationships between innovation entities, such as assignment, licensing, joint application, etc.; S1002: Construct a relationship graph. Using graph theory, take innovation subjects as nodes and patent behavior relationships as edges to construct a patent behavior relationship network. Each node in the network represents an innovation subject, and each edge represents the patent behavior relationship between two innovation subjects. S1003: Calculate the shortest path, using the shortest path algorithm in graph theory, namely the Floyd-Warshall algorithm, to calculate the shortest path length between any two innovation entities in the network; S1004: Define the relationship level, according to the formula PL ij =min(Length(p)), p∈P ij , taking the shortest path length as the relationship level, where P ij represents the set of all accessible paths between innovation subject i and innovation subject j, and Length(p) represents the length function of path p; if there is a direct patent behavior relationship between the two innovation subjects, their relationship level is 1, and they are first-level collaborators with each other; if there is no direct accessible path between the two innovation subjects, the shortest path length between them is defined as +∞, and the relationship level is +∞, indicating that they cannot be directly connected through the existing relationship, and they are potential collaborators with ∞ level each other; S1005: List potential collaborators. Based on the patent behavior relationship network, list all innovation entities that have potential cooperation relationships with the innovation entity as a list of potential collaborators; S1006: Calculate the relationship level. For each potential collaborator in the list, calculate the relationship level between it and the target innovation subject, that is, the length of the shortest path between them; S1007: Sort by relationship level, sort potential partners from small to large according to the relationship level; the smaller the relationship level, the greater the possibility of cooperation, and the lower the communication and negotiation costs; S1008: Generate a sorted list, generating a sorted list of potential collaborators, the list reflecting the priority of the possibility of cooperation; S1009: Identify collaborators with the same relationship level, and in the sorted list, identify a subset of potential collaborators with the same relationship level; S1010: Calculate the path strength. For each subset of collaborators with the same relationship level, calculate the path strength between the target innovation subject and each potential collaborator in the subset; Path strength can be defined as some measure of all possible paths connecting two nodes, such as path diversity, i.e., the number of different paths, the average importance of nodes on a path, i.e., the weighted sum based on indicators such as node degree and centrality, or path stability indicators, i.e., the stability assessment of the cooperation history between nodes on a path; S1011: Sort by path strength. In the subset of collaborators with the same relationship level, sort the potential collaborators from large to small according to the path strength. The greater the path intensity, the higher the stability of the relationship between innovation entities, the stronger the willingness to cooperate, and the more likely it is to form a long-term and stable cooperative relationship; S1012: Combine the ranking results of relationship level and path strength to conduct a comprehensive evaluation of potential partners; Prioritize potential partners with small relationship levels and high path strength as priority cooperation partners; S1013: Based on the comprehensive evaluation results, formulate specific cooperation strategies for innovation entities, including cooperation priorities, cooperation methods, cooperation content, etc.

Citation Information

Cited By

  • Patent navigation method based on knowledge acquisition

    CN120298166A

  • Knowledge graph and large language model-based interpretable patent recommendation method and device

    CN120541203A

  • Knowledge graph driven intelligent analysis system based on medical field

    CN120911582A

  • Rights processing support and settlement processing system and method

    JP7889937B1