Method and system for evaluating value of patent combination

Abstract

The invention relates to a mechanism evaluation method and system for patent combinations, and the method comprises the steps: obtaining four types of patent combinations in a patent database based on theme field classification, a keyword matching technology, a reference relation between patents, and a patent classification number, selecting N > = 2 patent documents meeting a preset screening condition from the four types of patent combinations as a target patent combination, secondly, vectorization processing is carried out on the N patent documents, the technical similarity between the N patent documents is obtained by calculating the comprehensive similarity between corresponding N vectorization processing results to serve as the node association degree, technical nodes recognized for the N patent documents serve as nodes, and a technical topological graph is constructed; and analyzing the technical topological graph by adopting a pre-trained value evaluation generation type language model to obtain technical complementarity among the N patent documents and market association information and legal association information of the N patent documents so as to scientifically and accurately generate value evaluation information of the target patent combination.

Description

Technical Field

[0001] The present invention relates to the field of patent evaluation technology, and in particular to a method and system for evaluating the value of a patent portfolio. Background Art

[0002] Intellectual property management, covering patent analysis, intellectual property protection, technology competition reports, etc., aims to help enterprises or universities effectively manage their intellectual property, protect innovative achievements, and improve market competitiveness.

[0003] Among them, the patent value assessment strategy based on the designated evaluation system is an analysis strategy based on patent data, which aims to help enterprises or universities protect and utilize the advantages of patent results. With stock patents as the carrier, from the aspects of laws and regulations, core technologies, market demand, etc., it provides companies or universities with a decision-making basis for the distribution of patent assets, the reasonable allocation of management resources, and the transfer and transformation of patent results, thereby promoting the precise operation of patents, clarifying the value of stock patents, effectively utilizing existing patent results, improving the sustainability of innovation, and maintaining market share.

[0004] However, existing methods lack support for natural language processing, resulting in inaccurate extraction and analysis of patent technology elements, making it difficult to deeply understand the core value and application potential of patents. The lack of system architecture and statistical analysis tools makes existing methods rely on empirical judgment in evaluating patent technology and citation scope, lacks scientific algorithm support, and limits the effectiveness of their optimization strategies. In addition, existing technologies lack support for text association technology and lack evaluation of the correlation and combination value between patents, which limits the comprehensiveness of their optimization strategies. Existing methods lack innovation and comprehensiveness in the evaluation strategy of patent portfolios, making it difficult to provide effective protection solutions. Summary of the invention

[0005] In a first aspect, the present application provides a method for evaluating the value of a patent portfolio, including:

[0006] In the patent database, based on the subject field classification, obtain the first category of patent portfolios belonging to the same subject field;

[0007] In the patent database, based on the keyword matching technology, a second category of patent combinations with consistent keyword matching is obtained;

[0008] In the patent database, based on the citation relationship between patents, a third-category patent portfolio belonging to the same technology chain is obtained;

[0009] In the patent database, based on the patent classification number, a fourth category patent combination belonging to the same classification number is obtained;

[0010] From the four types of patent portfolios obtained, select N patent documents that meet the preset screening conditions as the target patent portfolio; N ≥ 2;

[0011] Performing vectorization processing on the N patent documents respectively, and obtaining the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results;

[0012] The obtained technical similarity is used as the node association degree, and the technical nodes identified for the N patent documents are used as nodes to construct a technical topology map; wherein each technical node is associated with the document information of the corresponding patent document;

[0013] A pre-trained value assessment generative language model is used to analyze the technical topology map to obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value assessment information of the target patent portfolio.

[0014] In a second aspect, the present application embodiment provides a patent portfolio value assessment system, including:

[0015] An acquisition module is used to obtain a first-category patent combination belonging to the same subject field in a patent database based on subject field classification; obtain a second-category patent combination with consistent keyword matching in the patent database based on keyword matching technology; obtain a third-category patent combination belonging to the same technology chain based on the citation relationship between patents in the patent database; obtain a fourth-category patent combination belonging to the same classification number in the patent database based on patent classification numbers;

[0016] A selection module is used to select N patent documents that meet preset screening conditions from the four types of patent combinations obtained as the target patent combination; N≥2;

[0017] A processing module, used for performing vectorization processing on the N patent documents respectively, and obtaining the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results;

[0018] A construction module is used to construct a technology topology map by using the obtained technology similarity as the node association degree and the technology nodes identified for the N patent documents as nodes; wherein each technology node is associated with the document information of the corresponding patent document;

[0019] An evaluation module is used to use a pre-trained value evaluation generative language model to analyze the technical topology map, obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value evaluation information of the target patent portfolio.

[0020] Optionally, the acquisition module is used to obtain the first category of patent combinations belonging to the same subject field in the patent database based on the subject field classification, including:

[0021] For each patent document in the patent database, the LDA model is used to maximize the probability distribution of each document segmentation in each patent document to the subject segmentation, thereby obtaining the document subject of each patent document;

[0022] Clustering is performed on the document topics of each patent document in the patent database, each document cluster generated by the processing is used as the first type of patent combination, and the cluster center of each document cluster is used as the subject area of ​​each patent document in the first type of patent combination.

[0023] Optionally, the obtaining module is used to obtain the second category patent combination with consistent keyword matching in the patent database based on the keyword matching technology, including:

[0024] For each patent document in the patent database, using TF-IDF technology to extract document keywords of each patent document;

[0025] The document keywords of each patent document are matched with the document keywords of other patent documents respectively, and the patent documents with consistent matching are combined as the second type of patent combination.

[0026] Optionally, the obtaining module is used to obtain the third category patent combination belonging to the same technology chain in the patent database based on the citation relationship between patents, including:

[0027] In the patent database, obtaining cited patents in the examination information of each patent document;

[0028] Within a preset number of times, an iterative cited patent query process is performed on the cited patents to obtain the cited patents of the next iteration, so that each patent document and its cited patents are regarded as a third category patent combination belonging to the same technology chain.

[0029] Optionally, the obtaining module is used to obtain the fourth category patent combination belonging to the same classification number in the patent database based on the patent classification number, including:

[0030] Referring to a preset hierarchical structure for patent classification numbers, a patent classification number index is constructed, wherein the preset hierarchical structure includes: a category for distinguishing technical fields, a group for distinguishing specific aspects or applications in the technical fields, and a subgroup for specific patent description and classification, wherein the category includes at least one subgroup, and the subgroup includes at least one subgroup;

[0031] Extracting the patent classification number of each patent document in the patent database, and based on the patent classification number index, using a deep search algorithm to construct a patent classification library based on the patent classification number for the patent database;

[0032] In the patent classification library, patent documents corresponding to the same category, group or subgroup belonging to the same hierarchical structure are regarded as the fourth category patent combination belonging to the same classification number.

[0033] Optionally, the processing module includes:

[0034] Using Doc2Vec technology, respectively perform document embedding processing on the N patent documents to obtain document vector representations of the N patent documents;

[0035] For each patent document, the Euclidean distance between its document vector representation and other N-1 document vector representations is calculated as the technical similarity between each patent document and the other patent documents.

[0036] Optionally, the building block includes:

[0037] The N patent documents are constructed into N nodes, and based on the four types of patent combinations to which the N documents belong, the corresponding patent combination information is associated with each node; the patent combination information includes: subject area, keyword, technology chain, classification number;

[0038] For every two patent documents, the technical similarity between the two patent documents is used as the edge weight of the edge connecting the corresponding two nodes, and N-1 edges and their respective edge weights are obtained; the edge weight represents the node association between the two nodes;

[0039] Based on the N nodes, N-1 edges and their respective edge weights, a technical topology graph is constructed.

[0040] Optionally, the evaluation module includes:

[0041] Obtaining a value assessment generative language model that is retrained based on the pre-trained generative model and at least according to the patent database, product database, and legal database;

[0042] Input the technology topology graph into the value assessment generative language model, and retrieve matching market-related information and legal-related information from the product database and the legal database based on the patent portfolio information associated with each of the N node-represented patent documents;

[0043] Based on the edge weights of N-1 edges in the technology topology graph, patent documents associated with nodes with edge weights falling within a first edge weight threshold [0, K) are used as core patent documents in the target patent portfolio;

[0044] And, patent documents of nodes associated with edge weights falling within a second edge weight threshold [K, 1) are regarded as peripheral patent documents in the target patent portfolio;

[0045] Obtaining value assessment information of the target patent portfolio output by the value assessment generative language model; the value assessment information includes: a technical topology map, core patent documents and associated patent portfolio information, market-related information and legal-related information, peripheral patent documents and associated patent portfolio information, market-related information and legal-related information.

[0046] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor; the memory is used to store a computer program; the processor is used to implement any of the methods in the first part according to the computer program stored in the memory.

[0047] Beneficial effects of the present invention:

[0048] In the patent portfolio value assessment method proposed in this application, natural language processing technology is introduced. First, in the patent database, the first category of patent portfolios belonging to the same subject field is obtained based on the subject field classification, and the second category of patent portfolios with consistent keyword matching is obtained based on the keyword matching technology, and the third category of patent portfolios belonging to the same technology chain is obtained based on the citation relationship between patents, and the fourth category of patent portfolios belonging to the same classification number are obtained based on the patent classification number, thereby obtaining four categories of patent portfolios with objective content association. Then, from the obtained four categories of patent portfolios, N≥2 patent documents that meet the preset screening conditions are selected as the target patent portfolio, and vectorization processing is performed on these N patent documents respectively. By calculating the comprehensive similarity between the corresponding N vectorization processing results, the technical similarity between these N patent documents is obtained. Then, the obtained technical similarity is used as the node association, and the technical nodes identified for the N patent documents are used as nodes to construct a technical topology map, in which each technical node is associated with the document information of the corresponding patent document. Finally, the pre-trained generative language model for value assessment is used to analyze the technical topology map to obtain the technical complementarity between N patent documents, the market relevance information and legal relevance information of N patent documents, so as to generate the value assessment information of the target patent portfolio. Through the above methods, the objective division of patent portfolios and their comprehensive and accurate value assessment can be achieved, providing strong support for the technological innovation, market competition and patent layout of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a patent portfolio value assessment method proposed for this embodiment;

[0050] Figure 2 A schematic diagram of the structure of a patent portfolio value assessment system proposed in this embodiment;

[0051] Figure 3 This is a schematic diagram of the hardware structure of an electronic device proposed in this embodiment. DETAILED DESCRIPTION

[0052] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0053] It should be noted that the data collected and obtained in the embodiments of the present application must comply with national laws and regulations, and the processing and storage of data in the embodiments of the present application comply with national laws and regulations.

[0054] refer to Figure 1 As shown, a method for evaluating the value of a patent portfolio is provided for an embodiment of the present application.

[0055] S101: In the patent database, based on the subject field classification, obtain the first category of patent portfolios belonging to the same subject field.

[0056] S102: Based on the keyword matching technology, obtain the second category patent combination with consistent keyword matching in the patent database.

[0057] S103: In the patent database, based on the citation relationship between patents, obtain the third category patent combination belonging to the same technology chain.

[0058] S104: In the patent database, based on the patent classification number, obtain the fourth category patent combination belonging to the same classification number.

[0059] S105: Select N patent documents that meet preset screening conditions from the four types of patent combinations obtained as the target patent combination; N≥2.

[0060] S106: Perform vectorization processing on N patent documents respectively, and obtain the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results.

[0061] S107: Using the obtained technical similarity as the node association degree and the technical nodes identified for the N patent documents as nodes, a technical topology map is constructed; wherein each technical node is associated with the document information of the corresponding patent document.

[0062] S108: Using a pre-trained generative language model for value assessment, analyze the technology topology map to obtain the technical complementarity between N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value assessment information of the target patent portfolio.

[0063] In an optional implementation, for S101, in order to accurately and objectively obtain the first category of patent combinations belonging to the same subject area. In the embodiment of the present application, for each patent document in the patent database, the LDA model is used to maximize the probability distribution of the attribution of each document segmentation in each patent document to the subject segmentation, so as to obtain the document subject of each patent document; clustering processing is performed on the document subject of each patent document in the patent database, and each document cluster generated by the processing is used as the first category of patent combinations, and the cluster center of each document cluster is used as the subject area of ​​each patent document in the first category of patent combinations.

[0064] Specifically, in the LDA (Latent Dirichlet Allocation) model, the word segmentation in each patent document can be analogized to a "topic". These topics are implicit and are discovered through the LDA model. In the context of LDA, a document is composed of multiple words, and these words are assigned to different topics according to their degree of association with different topics. Each patent document includes multiple document types, such as abstracts, abstract drawings, claims, specifications, and specification drawings. When a new word segmentation (document) enters the library, the LDA model will assign it to the most appropriate topic based on the content (words) of the word segmentation. This process is probabilistic, that is, a book may belong to multiple topics at the same time, but the probability of belonging to each topic is different. Based on the fact that word segments are often assigned to the same topic together, a clear definition of each topic is formed. These definitions are based on the statistical characteristics of the word segmentation content, rather than on any pre-set classification criteria. The LDA model will then continuously adjust the distribution of word segments to optimize the purity and representativeness of each topic. Similar to the parameter estimation process in the LDA model, the topic distribution and word distribution are updated through iterative algorithms (such as Gibbs sampling or variational inference) to maximize the likelihood or posterior probability of the observed data.

[0065] In an alternative embodiment, in S102, to accurately and objectively obtain the second type of patent portfolio with consistent keyword matching. In the embodiments of the present application, for each patent document in the patent database, the TF-IDF technology is adopted to extract the document keywords of each patent document; the document keywords of each patent document are respectively matched with the document keywords of other patent documents, and the patent documents with consistent matching are combined as the second type of patent portfolio.

[0066] Specifically, for the extraction of the document keywords of each patent document, taking one patent document as an example. First, the patent document needs to be preprocessed, including: word segmentation, splitting the patent document text into words or phrases. In the case of compliance with laws and regulations, for Chinese patents, Chinese word segmentation tools such as jieba and HanLP can be used; stop word removal, removing those common words that are not helpful for text understanding (such as "of", "is", "in", etc.); word stemming (mainly for English), reducing the word to its basic form (such as reducing "running" to "run"); lemmatization (mainly for English), similar to word stemming, but usually more aggressively reducing the word to its root form. Then, the TF-IDF technology is adopted to evaluate the importance of a word for a document in a document set or a corpus. For easy understanding, a brief description of the TF-IDF technology principle is as follows. If a word or phrase has a high frequency TF in a patent document and rarely appears in other patent documents, then this word or phrase is considered to have good category discrimination ability and is suitable for classification. It mainly involves the following steps: calculating TF (term frequency), dividing the number of times a word appears in a document by the total number of words in the document; calculating IDF (inverse document frequency), dividing the total number of documents by the number of documents containing the word, and then taking the logarithm; calculating TF-IDF, multiplying TF by IDF. Finally, after calculating the TF-IDF values of each word in the patent document, select the top M (M>0) words with the highest values as keywords.

[0067] In addition, an external knowledge base (such as a professional dictionary, a thesaurus) can be introduced to optimize the keyword extraction effect.

[0068] In an alternative embodiment, in S103, to accurately and objectively obtain the third type of patent portfolio belonging to the same technology chain. In the embodiments of the present application, in the patent database, obtain the cited patents in the examination information of each patent document; under the preset number of times limit, perform iterative cited patent query processing on the cited patents to obtain the cited patents for the next iteration, so as to take each patent document and its cited patents as the third type of patent portfolio belonging to the same technology chain.

[0069] It should be noted that the patent database in the embodiment of the present application may be one or more databases containing the required patent data, such as USPTO (United States Patent and Trademark Office), EPO (European Patent Office), CNIPA (China National Intellectual Property Administration), etc.

[0070] Specifically, in the patent database, according to the research objectives and technical fields, appropriate search keywords and conditions are set to retrieve relevant patent documents from the database. Then, the examination information of the patent document is parsed, that is, for each retrieved patent document, its examination information part is parsed, especially the list of cited patents therein, which usually includes other patents cited by the patent in the writing process, reflecting the technical relevance and inheritance. Further, for the cited patent list of each patent, an iterative query is performed. That is, for each cited patent, its cited patents are searched again, and this process is repeated until the preset number of iterations is reached or no new cited patents are found. In order to control the complexity of the query and avoid infinite loops, a reasonable limit on the number of iterations is set. This limit depends on the specific research needs and the size of the data set. Then, each original patent document and all its cited patents obtained through iterative query are regarded as a set, that is, a patent combination belonging to the same technology chain. The aforementioned data is sorted out to create a clear document or database record for each patent combination, including information such as the original patent, cited patent, and number of iterations.

[0071] It should be noted that in the above iterative process, it is necessary to remove duplicate patent records to ensure that each patent is only counted once in the final combination. And when processing patent data, comply with relevant laws, regulations and privacy policies to ensure the legal use of data.

[0072] Through the above steps, the third-category patent portfolio belonging to the same technology chain can be obtained more accurately and objectively, providing support for further target patent portfolio analysis and decision-making.

[0073] In an optional implementation, in S104, in order to accurately and objectively obtain the fourth category patent combination belonging to the same classification number. In the embodiment of the present application, a patent classification number index is constructed with reference to a preset hierarchical structure for patent classification numbers, and the preset hierarchical structure includes: categories for distinguishing technical fields, groups for distinguishing specific aspects or applications in technical fields, and groups for specific patent descriptions and classifications, and the categories include at least one group, and the groups include at least one group; the patent classification number of each patent document in the patent database is extracted, and based on the patent classification number index, a deep search algorithm is used to construct a patent classification library based on the patent classification number for the patent database; in the patent classification library, patent documents corresponding to the same category, group or group belonging to the same hierarchical structure are regarded as the fourth category patent combination belonging to the same classification number.

[0074] Specifically, in order to understand the hierarchical structure of classification numbers, we can first deeply understand and clarify the preset hierarchical structure of patent classification numbers. This usually includes several levels, such as major categories (used to distinguish technical fields), middle categories (used to distinguish specific aspects or applications in technical fields), minor categories (used for specific patent descriptions and classifications), etc. Secondly, based on this hierarchical structure, a patent classification number index is constructed. This index should be able to support fast search and classification, ensuring that each classification number can be accurately mapped to its corresponding level and category.

[0075] Then, all relevant patent documents are obtained from the patent database. The corresponding patent classification number is extracted from each patent document, which is usually located in a specific part of the patent document, such as "IPC classification number" or "USPC classification number". The extracted patent classification number is then cleaned to remove duplicate, erroneous or irrelevant data. A deep search algorithm (which can also be further combined with breadth-first search) is used to traverse the patent database and classify each patent document into its corresponding hierarchical structure based on the patent classification number index. Based on the search results, a patent classification library based on patent classification numbers is constructed. This classification library should be able to clearly reflect the position of each patent document in its classification number hierarchy. Then, in the patent classification library, the corresponding patent documents are grouped together according to the same category, grouping or group of the same hierarchical structure. These combinations are the fourth category patent combinations belonging to the same classification number.

[0076] Furthermore, the constructed patent portfolio can be verified to ensure that they do belong to the same classification number and conform to the expected hierarchical structure. In this way, the constructed fourth-category patent portfolio can help with the subsequent technical analysis of the target portfolio patents, assist in identifying key technology fields, analyze technology development trends, and evaluate the technology competition landscape.

[0077] In an optional implementation, in S105, in order to obtain a target patent combination that meets the business requirements, in the embodiment of the present application, N≥2 patent documents that meet the preset screening conditions are selected from the four types of patent combinations obtained as the target patent combination.

[0078] Among them, the above-mentioned preset screening conditions are screening conditions set according to business needs, such as release time, patent owner, country / region, etc. For a number of (greater than or equal to N) patent documents that meet the set screening conditions, for each patent document, the four types of patent combinations containing the patent documents are obtained based on S101 to S104, and N patent documents are randomly selected from the intersection or union of the four types of patent combinations. Here, the aforementioned selection conditions may also include preset sorting and selection rules, and use machine learning sorting algorithms (such as RankNet, LambdaMART) to sort the candidate patents in the intersection or union, and select the top N.

[0079] In an optional implementation, in S106, in order to accurately and objectively obtain the technical similarity between N patent documents. In the embodiment of the present application, Doc2Vec technology is used to perform document embedding processing on the N patent documents to obtain the document vector representation of each of the N patent documents; for each patent document, the Euclidean distance between its document vector representation and other N-1 document vector representations is calculated as the technical similarity between each patent document and other patent documents.

[0080] Specifically, for the process of obtaining the document vector representation of each of the N patent documents mentioned above, first, it is necessary to read the text content of the N patent documents from the patent database, clean the patent documents, including removing HTML tags, special characters, stop words, etc., and perform word segmentation (for Chinese patents, word segmentation and removal of redundant words may be required), build the words in all documents into a vocabulary, and assign a unique index to each unique word. Then, it is necessary to train the Doc2Vec model to predict the randomly sampled words in the document through the document vector, including: setting model parameters: vector dimension (d), which is usually tens to hundreds of dimensions, window size (window), which is the number of context words, number of iterations (epochs), which is the number of training cycles, learning rate (lr), which is the step size for controlling weight updates, and other possible parameters such as negative sampling size (negative), etc.; setting the model loss function as follows:

[0081]

[0082] Where T is the total number of training samples, wt is the target word in the tth training sample, d is the corresponding document vector, and p(wt|d) is the probability of predicting word wt given the document vector d.

[0083] Then, based on the above settings, the model is trained, using the preprocessed patent documents as input, and the Doc2Vec model is trained according to the selected model architecture and set parameters. During the training process, the model tries to minimize the loss function by adjusting the weights of the document vector and the vocabulary vector.

[0084] After training, the model is able to generate a fixed-dimensional vector representation for each patent document. These vectors capture the semantic relationship between documents, making similar documents closer in the vector space. The vector representation of each patent document is stored here for subsequent analysis and comparison.

[0085] In an optional implementation, in S107, a technical topology map is constructed accurately, objectively and clearly. In the embodiment of the present application, N patent documents are constructed into N nodes, and the corresponding patent combination information is associated with each node based on the four types of patent combinations to which the N documents belong; the patent combination information includes: subject area, keyword, technology chain, classification number; for every two patent documents, the technical similarity between the two patent documents is used as the edge weight of the edge connecting the corresponding two nodes, and N-1 edges and their respective edge weights are obtained; the edge weight represents the node association between the two nodes; based on the N nodes, N-1 edges and their respective edge weights, a technical topology map is constructed.

[0086] Specifically, first, N patent documents are taken as input, and each patent document is converted into a node. In this way, there are N nodes in the constructed technology topology diagram, and each node represents a patent document. Secondly, for each node (i.e., each patent document), according to the four types of patent combinations to which it belongs (here it is based on the classification method predefined in the embodiment of the present application, such as the first type of patent combination, the second type of patent combination, the third type of patent combination, the fourth type of patent combination, etc.), the corresponding patent combination information is associated. This information includes but is not limited to: subject area, describing the main technical field or subject to which the patent belongs; keywords, which are words extracted from patent documents that can summarize the core content of the document; technology chain, which indicates the position of the patent in the process of technological development, and may involve the identification of predecessor technology, successor technology or related technology chains; classification number, the official classification number to which the patent belongs, used to identify its technical field and classification.

[0087] Secondly, for every two patent documents (i.e., every two nodes), the technical similarity between them is calculated. This can be achieved through a variety of methods, such as similarity algorithms based on text content (such as cosine similarity), similarity assessment based on patent classification numbers, or a composite similarity model combining multiple features. The calculated technical similarity is used as the edge weight of the edge connecting the corresponding two nodes. The higher the edge weight, the stronger the technical similarity or correlation between the two nodes (i.e., the two patent documents).

[0088] Then, based on N nodes, N-1 edges (note that N-1 is a simplified statement here, and the actual number of edges may be much larger than N-1, because the connections between nodes in most graphs are complex mesh structures) and their respective edge weights, a technical topology graph is constructed using graph theory methods.

[0089] Optionally, in order to more intuitively display the technology topology map, you can use graphical tools for visualization. At the same time, you can analyze the map, such as identifying key nodes (technical core or important patents), analyzing the technology evolution path, and discovering technology intersections.

[0090] It should be noted that with the addition of new patents and the elimination of old patents, the technology topology map should be able to be dynamically updated to reflect the latest status of technological development.

[0091] Through the above steps, an accurate, objective and clear technology topology map can be constructed to provide strong support for the technical analysis, decision support and strategic planning of the target patent portfolio.

[0092] In an optional implementation, in S108, the value assessment information of the target patent portfolio is generated accurately, objectively, clearly and comprehensively. In the embodiment of the present application, a value assessment generative language model based on a pre-trained generative model is obtained, which is retrained based on at least a patent database, a product database and a legal database; a technical topology map is input into the value assessment generative language model, and the matching market-related information and legal-related information are retrieved from the product database and the legal database based on the patent portfolio information associated with each of the N nodes representing the patent documents; based on the edge weights of the N-1 edges in the technical topology map, the patent documents associated with the nodes with edge weights falling into the first edge weight threshold [0, K) are used as the core patent documents in the target patent portfolio; and the patent documents associated with the nodes with edge weights falling into the second edge weight threshold [K, 1) are used as the peripheral patent documents in the target patent portfolio; the value assessment information of the target patent portfolio output by the value assessment generative language model is obtained; the value assessment information includes: a technical topology map, core patent documents and associated patent portfolio information, market-related information and legal-related information, and peripheral patent documents and associated patent portfolio information, market-related information and legal-related information.

[0093] It can be divided into the following four parts:

[0094] Part 1: Data preparation and preprocessing.

[0095] First, collect relevant data from patent databases, product databases, and legal databases. The patent database should contain information such as patent documents, classification numbers, keywords, etc. The product database should contain product descriptions, market positioning, sales data, etc. The legal database should contain the legal status of the patent, litigation records, licensing agreements, etc. Then clean the collected data to remove duplicate, incomplete, or erroneous data. Extract features such as subject areas, keywords, and technology chains from patent documents; extract market-related information (such as market share, competitors, user feedback, etc.) from the product database; and extract legal-related information (such as patent validity period, litigation history, patent licensing fees, etc.) from the legal database.

[0096] The second part is model building and training.

[0097] First, select a pre-trained model suitable for generative tasks, such as the GPT series, BERT series, etc. These models perform well in text generation and understanding. Then retrain the pre-trained model using the prepared data. During the training process, the model needs to learn how to combine patent documents, product information, and legal information to generate valuable evaluation information. In addition, optimize the model: improve the accuracy and efficiency of the model by adjusting model parameters, optimizing algorithms, and training strategies.

[0098] Part 3: Technology topology map input and value assessment.

[0099] First, the constructed technology topology map is used as input, where N nodes represent patent documents, and N-1 edges and their weights represent the technical similarity and relevance between patent documents. Then, the product database and legal database are used to retrieve matching market-related information and legal-related information for each node (i.e., each patent document) in the technology topology map. Patent documents are then divided into core patent documents and peripheral patent documents based on edge weight thresholds (such as the first edge weight threshold [0, K] and the second edge weight threshold [K, 1)). Core patent documents occupy a more important position in the technology topology map and have a greater impact on the overall value assessment.

[0100] Part 4: Generation of value assessment information.

[0101] First, the technical topology map, core patent documents and their associated patent portfolio information, market-related information and legal-related information, as well as peripheral patent documents and their related information, are input into the value assessment generative language model. The model then generates a value assessment report based on the input information, which describes in detail the technical characteristics, market potential, legal risks and overall value assessment of the target patent portfolio. The generated value assessment information helps to focus on the core patent documents in the target patent portfolio, understand the strengths and weaknesses of the target patent portfolio, and potential market and legal risks, so that the value assessment results can be used as a decision support tool to help companies formulate patent strategies, product planning and market strategies.

[0102] In addition, in order to better understand the above-mentioned value assessment generative language model (hereinafter briefly described as generative language model), the following additional explanation is given for the model's reasoning process. The reasoning process of generative language models (such as GPT, BERT, etc.) is learned through a large amount of training data. They use complex neural network structures to capture the dependencies and semantic information between texts.

[0103] In other words, inside the model, each input text (such as patent documents, product descriptions, legal terms, etc.) is first converted into a high-dimensional vector (embedding vector). This process can be regarded as an implicit "function", which can be expressed as: Embedding(x) = ex, where x is the input text and ex is the corresponding embedding vector.

[0104] In generative models, the representation of the current text is usually affected by its context (such as the previous text, the next text, or the entire technology topology graph). This context encoding can be seen as a complex function that combines multiple embedding vectors to generate a representation containing context information: ContextEncoding(e1,e2,...,eN)=c, where e1,e2,...,eN are the embedding vectors of each node in the technology topology graph, and c is the encoding vector containing context information.

[0105] The calculation of edge weights in a technical topology graph is usually based on the similarity between nodes. This similarity can be obtained by comparing their embedding vectors or context encoding vectors. A simple similarity calculation can be cosine similarity: where v1 and v2 can be either embedding vectors or context encoding vectors.

[0106] The final valuation is a complex process that integrates multiple factors (such as technical importance, market potential, legal risks). Although this process is difficult to express with a single formula, we can think of it as an estimate based on a weighted sum or conditional probability of all the above information:

[0107]

[0108] Or ValueAssessment≈P(HighValue|Topology,MarketInfo,LegalInfo), where Topology is the technical topology, MarketInfo and LegalInfo are market-related information and legal-related information respectively, wi is the weight of node i (which may be related to its position in the technical topology or edge weight), and fi is a function used to calculate the value contribution of the node based on node information, market information and legal information.

[0109] It should be noted that fi and wi in the above expression are not calculated by simple formulas, but are automatically learned by the generative language model during training. In addition, due to the complexity of the generative model, the actual reasoning process may involve multiple levels of encoding, decoding, and attention mechanisms, which will not be elaborated here.

[0110] In summary, through the above methods, it is possible to achieve an objective division of patent portfolios and a comprehensive and accurate assessment of their value, providing strong support for enterprises' technological innovation, market competition and patent layout.

[0111] This embodiment proposes a schematic diagram of the structure of a patent portfolio value assessment system, such as Figure 2 As shown, the value assessment system 20 of the patent combination of this embodiment is used to implement the operation corresponding to the electronic device in any of the above method embodiments. The value assessment system 20 of the patent combination of this embodiment includes:

[0112] The acquisition module 201 is used to obtain a first-category patent combination belonging to the same subject field in the patent database based on the subject field classification; obtain a second-category patent combination with consistent keyword matching based on the keyword matching technology in the patent database; obtain a third-category patent combination belonging to the same technology chain based on the citation relationship between patents in the patent database; obtain a fourth-category patent combination belonging to the same classification number based on the patent classification number in the patent database;

[0113] The selection module 202 is used to select N patent documents that meet the preset screening conditions from the obtained four types of patent combinations as the target patent combination; N≥2;

[0114] The processing module 203 is used to perform vectorization processing on the N patent documents respectively, and obtain the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results;

[0115] A construction module 204 is used to construct a technology topology map by using the obtained technology similarity as the node association degree and the technology nodes identified for the N patent documents as nodes; wherein each technology node is associated with the document information of the corresponding patent document;

[0116] The evaluation module 205 is used to use a pre-trained value evaluation generative language model to analyze the technical topology map, obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value evaluation information of the target patent portfolio.

[0117] Optionally, the obtaining module 201 is used to obtain the first category of patent combinations belonging to the same subject field in the patent database based on the subject field classification, including:

[0118] For each patent document in the patent database, the LDA model is used to maximize the probability distribution of each document segmentation in each patent document to the subject segmentation, thereby obtaining the document subject of each patent document;

[0119] Clustering is performed on the document topics of each patent document in the patent database, each document cluster generated by the processing is used as the first type of patent combination, and the cluster center of each document cluster is used as the subject area of ​​each patent document in the first type of patent combination.

[0120] Optionally, the obtaining module 201 is used to obtain the second category patent combination with consistent keyword matching in the patent database based on the keyword matching technology, including:

[0121] For each patent document in the patent database, using TF-IDF technology to extract document keywords of each patent document;

[0122] The document keywords of each patent document are matched with the document keywords of other patent documents respectively, and the patent documents with consistent matching are combined as the second type of patent combination.

[0123] Optionally, the obtaining module 201 is used to obtain the third category patent combination belonging to the same technology chain in the patent database based on the citation relationship between patents, including:

[0124] In the patent database, obtaining cited patents in the examination information of each patent document;

[0125] Within a preset number of times, an iterative cited patent query process is performed on the cited patents to obtain the cited patents of the next iteration, so that each patent document and its cited patents are regarded as a third category patent combination belonging to the same technology chain.

[0126] Optionally, the obtaining module is used to obtain the fourth category patent combination belonging to the same classification number in the patent database based on the patent classification number, including:

[0127] Referring to a preset hierarchical structure for patent classification numbers, a patent classification number index is constructed, wherein the preset hierarchical structure includes: a category for distinguishing technical fields, a group for distinguishing specific aspects or applications in the technical fields, and a subgroup for specific patent description and classification, wherein the category includes at least one subgroup, and the subgroup includes at least one subgroup;

[0128] Extracting the patent classification number of each patent document in the patent database, and based on the patent classification number index, using a deep search algorithm to construct a patent classification library based on the patent classification number for the patent database;

[0129] In the patent classification library, patent documents corresponding to the same category, group or subgroup belonging to the same hierarchical structure are regarded as the fourth category patent combination belonging to the same classification number.

[0130] Optionally, the processing module 203 includes:

[0131] Using Doc2Vec technology, respectively perform document embedding processing on the N patent documents to obtain document vector representations of the N patent documents;

[0132] For each patent document, the Euclidean distance between its document vector representation and other N-1 document vector representations is calculated as the technical similarity between each patent document and the other patent documents.

[0133] Optionally, the construction module 204 includes:

[0134] The N patent documents are constructed into N nodes, and based on the four types of patent combinations to which the N documents belong, the corresponding patent combination information is associated with each node; the patent combination information includes: subject area, keyword, technology chain, classification number;

[0135] For every two patent documents, the technical similarity between the two patent documents is used as the edge weight of the edge connecting the corresponding two nodes, and N-1 edges and their respective edge weights are obtained; the edge weight represents the node association between the two nodes;

[0136] Based on the N nodes, N-1 edges and their respective edge weights, a technical topology graph is constructed.

[0137] Optionally, the evaluation module 205 includes:

[0138] Obtaining a value assessment generative language model that is retrained based on the pre-trained generative model and at least according to the patent database, product database, and legal database;

[0139] Input the technology topology graph into the value assessment generative language model, and retrieve matching market-related information and legal-related information from the product database and the legal database based on the patent portfolio information associated with each of the N node-represented patent documents;

[0140] Based on the edge weights of N-1 edges in the technology topology graph, patent documents associated with nodes with edge weights falling within a first edge weight threshold [0, K) are used as core patent documents in the target patent portfolio;

[0141] And, patent documents of nodes associated with edge weights falling within a second edge weight threshold [K, 1) are regarded as peripheral patent documents in the target patent portfolio;

[0142] Obtaining value assessment information of the target patent portfolio output by the value assessment generative language model; the value assessment information includes: a technical topology map, core patent documents and associated patent portfolio information, market-related information and legal-related information, peripheral patent documents and associated patent portfolio information, market-related information and legal-related information.

[0143] The patent portfolio value assessment system 20 provided in the embodiment of the present application can execute the above method embodiment. Its specific implementation principles and technical effects can be found in the above method embodiment, and this embodiment will not be repeated here.

[0144] This embodiment provides a hardware structure diagram of an electronic device. Figure 3 As shown, the electronic device 30 is used to implement the operations corresponding to the electronic device in any of the above method embodiments. The electronic device 30 of this embodiment may include: a memory 31, a processor 32 and a communication interface 34.

[0145] The memory 31 is used to store computer programs. The memory 31 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk.

[0146] The processor 32 is used to execute the computer program stored in the memory to implement the method in the above embodiment. For details, please refer to the relevant description in the above method embodiment. The processor 32 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0147] Optionally, the memory 31 may be independent or integrated with the processor 32 .

[0148] When the memory 31 is a device independent of the processor 32, the electronic device 30 may further include a bus 33. The bus 33 is used to connect the memory 31 and the processor 32. The bus 33 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0149] The communication interface 34 can be connected to the processor 31 via the bus 33. The processor 32 can control the communication interface 34 to obtain signals such as departure instructions.

[0150] The electronic device provided in this embodiment can be used to execute the above-mentioned patent portfolio value assessment method, and its implementation method and technical effect are similar, which will not be described in detail in this embodiment.

[0151] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the methods provided in the various embodiments described above.

[0152] Among them, the computer-readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, a computer-readable storage medium is coupled to a processor so that the processor can read information from the computer-readable storage medium and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the computer-readable storage medium can also exist in a communication device as discrete components.

[0153] Specifically, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0154] The present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor of a device can read the computer program from the computer-readable storage medium, and at least one processor executes the computer program so that the device implements the methods provided in the various embodiments described above.

[0155] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or modules, which can be electrical, mechanical or other forms.

[0156] Among them, each module can be physically separated, for example, installed in different locations of a device, or installed on different devices, or distributed on multiple network units, or distributed on multiple processors. Each module can also be integrated together, for example, installed in the same device, or integrated in a set of codes. Each module can exist in the form of hardware, or can also exist in the form of software, or can also be implemented in the form of software plus hardware. The present application can select some or all of the modules according to actual needs to achieve the purpose of the present embodiment.

[0157] When each module is implemented as an integrated module in the form of a software function module, it can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of each embodiment of the present application.

[0158] It should be understood that, although the various steps in the flowcharts in the above-described embodiments are sequentially displayed according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and they can be executed in other orders. Moreover, at least a portion of the steps in the figure may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily to be carried out sequentially, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0159] The above embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Any equivalent substitution or change made by a person skilled in the art based on the present invention is within the protection scope of the present invention.

Claims

1. A method for evaluating the value of a patent portfolio, the method comprising: In the patent database, based on the subject field classification, obtain the first category of patent portfolios belonging to the same subject field; In the patent database, based on the keyword matching technology, a second category of patent combinations with consistent keyword matching is obtained; In the patent database, based on the citation relationship between patents, a third-category patent portfolio belonging to the same technology chain is obtained; In the patent database, based on the patent classification number, a fourth category patent combination belonging to the same classification number is obtained; From the four types of patent portfolios obtained, select N patent documents that meet the preset screening conditions as the target patent portfolio; N ≥ 2; Performing vectorization processing on the N patent documents respectively, and obtaining the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results; The obtained technical similarity is used as the node association degree, and the technical nodes identified for the N patent documents are used as nodes to construct a technical topology map; wherein each technical node is associated with the document information of the corresponding patent document; A pre-trained value assessment generative language model is used to analyze the technical topology map to obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value assessment information of the target patent portfolio.

2. The method according to claim 1, characterized in that The patent database, based on the subject field classification, obtains the first category of patent portfolios belonging to the same subject field, including: For each patent document in the patent database, the LDA model is used to maximize the probability distribution of each document segmentation in each patent document to the subject segmentation, thereby obtaining the document subject of each patent document; Clustering is performed on the document topics of each patent document in the patent database, each document cluster generated by the processing is used as the first type of patent combination, and the cluster center of each document cluster is used as the subject area of ​​each patent document in the first type of patent combination.

3. The method according to claim 1, characterized in that The second category of patent combinations with consistent keyword matching are obtained in the patent database based on the keyword matching technology, including: For each patent document in the patent database, using TF-IDF technology to extract document keywords of each patent document; The document keywords of each patent document are matched with the document keywords of other patent documents respectively, and the patent documents with consistent matching are combined as the second type of patent combination.

4. The method according to claim 1, characterized in that In the patent database, based on the citation relationship between patents, the third category of patent combinations belonging to the same technology chain are obtained, including: In the patent database, obtaining cited patents in the examination information of each patent document; Within a preset number of times, an iterative cited patent query process is performed on the cited patents to obtain the cited patents of the next iteration, so that each patent document and its cited patents are regarded as a third category patent combination belonging to the same technology chain.

5. The method according to claim 1, characterized in that In the patent database, based on the patent classification number, the fourth category patent combination belonging to the same classification number is obtained, including: Constructing a patent classification number index with reference to a preset hierarchical structure for patent classification numbers, wherein the preset hierarchical structure includes: a category for distinguishing technical fields, a group for distinguishing specific aspects or applications in technical fields, and a subgroup for specific patent description and classification, wherein the category includes at least one subgroup, and the subgroup includes at least one subgroup; Extracting the patent classification number of each patent document in the patent database, and based on the patent classification number index, using a deep search algorithm to construct a patent classification library based on the patent classification number for the patent database; In the patent classification library, patent documents corresponding to the same category, group or subgroup belonging to the same hierarchical structure are regarded as the fourth category patent combination belonging to the same classification number.

6. The method according to claim 1, characterized in that The performing vectorization processing on the N patent documents respectively and obtaining the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results includes: Using Doc2Vec technology, respectively perform document embedding processing on the N patent documents to obtain document vector representations of the N patent documents; For each patent document, the Euclidean distance between its document vector representation and other N-1 document vector representations is calculated as the technical similarity between each patent document and the other patent documents.

7. The method according to claim 1, characterized in that The obtained technical similarity is used as the node association degree, and the technical nodes identified for the N patent documents are used as nodes to construct a technical topology map; wherein each technical node is associated with the document information of the corresponding patent document, including: The N patent documents are constructed into N nodes, and based on the four types of patent combinations to which the N documents belong, the corresponding patent combination information is associated with each node; the patent combination information includes: subject area, keyword, technology chain, and classification number; For every two patent documents, the technical similarity between the two patent documents is used as the edge weight of the edge connecting the corresponding two nodes, and N-1 edges and their respective edge weights are obtained; the edge weight represents the node association between the two nodes; Based on the N nodes, N-1 edges and their respective edge weights, a technical topology graph is constructed.

8. The method according to any one of claims 1 to 7, characterized in that: The pre-trained value assessment generative language model is used to analyze the technology topology map to obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate the value assessment information of the target patent portfolio, including: Obtaining a value assessment generative language model that is retrained based on the pre-trained generative model and at least according to the patent database, product database, and legal database; Input the technology topology map into the value assessment generative language model, and retrieve matching market-related information and legal-related information from the product database and the legal database based on the patent portfolio information associated with each of the N node-represented patent documents; Based on the edge weights of N-1 edges in the technology topology graph, patent documents associated with nodes with edge weights falling within a first edge weight threshold [0, K) are used as core patent documents in the target patent portfolio; And, patent documents of nodes associated with edge weights falling within a second edge weight threshold [K, 1) are regarded as peripheral patent documents in the target patent portfolio; Obtaining value assessment information of the target patent portfolio output by the value assessment generative language model; the value assessment information includes: a technical topology map, core patent documents and associated patent portfolio information, market-related information and legal-related information, peripheral patent documents and associated patent portfolio information, market-related information and legal-related information.

9. A patent portfolio value assessment system, comprising: An acquisition module is used to obtain the first-category patent portfolio belonging to the same subject field in the patent database based on the subject field classification; In the patent database, based on the keyword matching technology, the second category of patent combinations with consistent keyword matching are obtained; in the patent database, based on the citation relationship between patents, the third category of patent combinations belonging to the same technology chain are obtained; in the patent database, based on the patent classification number, the fourth category of patent combinations belonging to the same classification number are obtained; A selection module is used to select N patent documents that meet preset screening conditions from the four types of patent combinations obtained as the target patent combination; N≥2; A processing module, used for performing vectorization processing on the N patent documents respectively, and obtaining the technical similarity between the N patent documents by calculating the comprehensive similarity between the corresponding N vectorization processing results; A construction module is used to construct a technology topology map by using the obtained technology similarity as the node association degree and the technology nodes identified for the N patent documents as nodes; wherein each technology node is associated with the document information of the corresponding patent document; An evaluation module is used to use a pre-trained value evaluation generative language model to analyze the technical topology map, obtain the technical complementarity between the N patent documents, the market relevance information and the legal relevance information of the N patent documents, so as to generate value evaluation information of the target patent portfolio.

10. An electronic device, characterized in that: The device comprises: a memory and a processor; The memory is used to store a computer program; the processor is used to implement the method according to any one of claims 1 to 8 according to the computer program stored in the memory.

Images