Patent text expansion and enrichment method based on semantic knowledge graph and matching algorithm

Abstract

The invention discloses a patent text extension and enrichment method based on a semantic knowledge graph and a matching algorithm. The method comprises the following steps: constructing an initial semantic knowledge graph based on a patent text to be extended to extract key technical features; retrieving related technical support documents in a technical document database by adopting a preset matching algorithm; generating novel contents related to the key technical features through a text generation model; based on the novel content and the initial patent text, the expansion and enrichment process of the patent text is completed, and the technical relevance and accuracy of the expansion part are ensured. According to the method, the problem of keeping the correlation and accuracy of technical contents during patent text extension can be solved.

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a patent text expansion and enrichment method based on a semantic knowledge graph and a matching algorithm. Background Art

[0002] Patent text expansion and enrichment methods refer to expanding the content of the original document by adding more information or extracting additional data from existing technical documents on the basis of the original patent document, so as to make the patent application more comprehensive and detailed, which is conducive to the patent examiner's more accurate understanding of the patent technology. This method plays an important role in the field of patent writing, especially when it is necessary to better demonstrate the novelty or creativity of the technology. However, one challenge encountered in the implementation of this technology is how to ensure that the newly expanded content remains relevant and accurate to the original technical description. Specifically, as the amount of text increases, content or erroneous information that is irrelevant to the core invention may be introduced, which may interfere with the examiner's judgment on the innovativeness of the core technology, thereby affecting the quality of the patent and the final authorization decision. Therefore, it becomes an important task to solve the problem of ensuring the intra-technical relevance and accuracy of the patent text during the expansion process. Summary of the invention

[0003] The purpose of the present invention is to propose a patent text expansion and enrichment method based on semantic knowledge graph and matching algorithm to overcome the above-mentioned deficiencies in the prior art.

[0004] To achieve the above technical objectives, the technical solution of the present invention is implemented as follows:

[0005] Patent text expansion and enrichment methods based on semantic knowledge graph and matching algorithm, including:

[0006] Build an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features;

[0007] Retrieve relevant technical support documents from the technical document database using a preset matching algorithm;

[0008] Generate novel content related to the key technical features through a text generation model;

[0009] Based on the novel content and the initial patent text, the patent text expansion and enrichment process is completed, and the technical relevance and accuracy of the expanded part are ensured.

[0010] Preferably, the step of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features specifically includes:

[0011] Identify key technical and lexical units from the patent text to be expanded;

[0012] A semantic association matrix was constructed to quantify the strength of relationships between technical and lexical units;

[0013] Apply natural language processing techniques to identify semantic relationships between terms;

[0014] Using graph databases to store semantic relationships and key technologies;

[0015] Representative key technical features are extracted through the depth-first search algorithm of the graph database.

[0016] Preferably, the step of using a preset matching algorithm to retrieve relevant technical support documents in a technical document database specifically includes:

[0017] Convert the extracted key technical features into query strings;

[0018] Use fuzzy matching algorithms to improve query flexibility and matching quality;

[0019] Perform semantic vector comparison to determine the importance of related technical support documents;

[0020] Set a threshold to filter candidate technical documents with higher matching scores;

[0021] Search results are sorted by match to return the most relevant technical support documents.

[0022] Preferably, the step of generating novel content related to the key technical features through a text generation model specifically includes:

[0023] Pre-trained Transformer-based deep learning model for generating natural language sentences;

[0024] Construct conditional probability distribution to ensure the consistency of produced content and input key technical features;

[0025] Introduce constraints to avoid generating content that already exists in the literature and ensure the independence and originality of the output;

[0026] Follow the grammar checker to ensure the grammar of the model output is correct;

[0027] Use feedback loop mechanisms to iteratively optimize text to improve the logical coherence and technical depth of the content.

[0028] Preferably, the steps of completing the expansion and enrichment process of the patent text based on the novel content and the initial patent text and ensuring the technical relevance and accuracy of the expanded part specifically include:

[0029] Incorporate and restructure novel content to fit into the patent’s structured framework;

[0030] Application expert review reviews the accuracy of the use of technical terms and makes necessary corrections;

[0031] Achieve a balance between logical and technical coherence, ensuring that the extension is clear and easy to understand;

[0032] Verify that the added content does not conflict with existing legal documents or prior definitions of patent rights;

[0033] Regularly validate the content of new extensions to confirm compliance with the latest patent drafting standards and maintain accuracy.

[0034] Preferably, the step of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features includes:

[0035] Perform word segmentation on the original patent text and break it down into several basic units;

[0036] Identify the relationship between basic units based on the word segmentation results and construct a relationship graph;

[0037] Mining the key features and relevance of technical texts by analyzing relationship graphs;

[0038] Based on the above results, the keywords that best reflect the technical characteristics are selected as technical features.

[0039] Preferably, the method for constructing a relationship graph is specifically as follows: when the semantic relevance between the keywords is greater than 7, it is determined that there is a strong correlation between the keywords, thereby establishing corresponding node connections.

[0040] Preferably, the step of refining the technical points specifically included in the key technical features based on the relationship map includes:

[0041] Define the critical path formed by the combination of key nodes and their adjacent edges;

[0042] Calculate the coverage and importance of features based on the critical path, where coverage is calculated as the number of nodes on the path and importance is defined as the sum of the strengths of the edges on the path;

[0043] When the result of multiplying the importance value by the coverage value exceeds 0, it is considered that this critical path can provide a deep description of the technology;

[0044] Extract key information from these deep descriptions to form more precise technical features.

[0045] Preferably, among the important contents screened out based on the key technical features, the conditional judgment steps in the process of selecting the technical points that are not completely the same as the existing patent contents, are more specific, and can improve the understanding as the final technical features are:

[0046] Check whether the features are novel and determine whether the technical features have been recorded or mentioned in detail in the public information;

[0047] Based on the recorded repetition coefficient calculation, where the repetition coefficient is calculated as the average similarity between the feature and other public documents. If the repetition coefficient is less than 4, it is considered novel and can be adopted as a feature;

[0048] When the novelty is satisfied, the comprehensiveness and feasibility of the feature description will be further examined, and incomplete technical elements that only contain general principle descriptions will be excluded;

[0049] The elements that remain after passing these two checks are given a higher weight, and the high-value content finally selected is marked as important technical features.

[0050] Preferably, the specific operation method to ensure that the extracted key content is innovative and does not rely entirely on existing files is:

[0051] Ensure that the added content is new by evaluating the topic relevance of the extracted content, for example, a topic relevance coefficient greater than or equal to 75 is considered strong relevance;

[0052] If the new content does not appear frequently enough in the existing technical literature, that is, the frequency of mention in a pre-set literature set is less than or equal to 3%, it is considered to be innovative;

[0053] After the extracted content meets the judgment criteria of novelty and comprehensiveness (judgment of relevance and frequency of mention), its accuracy and completeness are further ensured. If the above conditions are met and the technical description clarity evaluation is higher than 4 / 5, it is judged to be an appropriate technical addition;

[0054] Finally, based on this judgment process, the specific content to be added to the patent text to expand the existing patent document is determined to ensure the technical relevance and accuracy of the new text.

[0055] Beneficial effects of the present invention:

[0056] The present invention can solve the problem of maintaining the relevance and accuracy of technical content when expanding patent text by constructing a semantic knowledge graph and a matching algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of a patent text expansion enrichment method;

[0058] Figure 2 It is a flowchart of the steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features;

[0059] Figure 3 A detailed flowchart of the steps of using a preset matching algorithm to retrieve relevant technical support documents from a technical document database;

[0060] Figure 4 A detailed flowchart of the steps to generate novel content related to key technical features through a text generation model;

[0061] Figure 5 A detailed flowchart of the steps to complete the process of expanding and enriching the patent text based on the novel content and the initial patent text and ensure the technical relevance and accuracy of the expanded part;

[0062] Figure 6 is a flowchart of the steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features;

[0063] Figure 7 The method of constructing a relationship graph is specifically to determine that there is a strong correlation between the keywords when the semantic relevance between the keywords is greater than 7, thereby establishing a flowchart of corresponding node connections;

[0064] Figure 8 It is a flowchart that refines the steps of the technical points specifically included in the key technical features based on the relationship map;

[0065] Fig. 9 It is a flow chart of the conditional judgment steps in the process of selecting technical points that are not completely the same as the existing patent content, are more specific, and can improve understanding as the final technical features from the important contents screened out based on the key technical features;

[0066] Fig.10 It is a flowchart that ensures that the extracted key content is innovative and does not completely rely on the specific operating methods of existing documents. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the embodiments of the present disclosure more clearly understood, the embodiments of the present disclosure are further described in detail below in conjunction with the embodiments and drawings.

[0068] In order to better understand the specific operation and implementation logic of the patent text expansion and enrichment method, the four core steps contained in the method will be explained one by one in the following description, and examples will be used to illustrate how this method ensures that the expanded content maintains both technical relevance and content accuracy.

[0069] In this process, the first step is to construct an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features. The purpose of this process is to identify and capture key technical information that can constitute or represent unique innovations in the field through in-depth analysis of the original text. This usually includes but is not limited to: technical goals, technical means and solutions. For example, if we need to expand a patent text on battery technology, we need to extract key information related to battery type (lithium ion), structure (multi-layer), energy density, etc. from the existing patent documents; then use automated algorithms or tools (such as NLP algorithms or semantic analysis platforms) to construct this information into a highly structured semantic knowledge graph for further content mining and generation.

[0070] The second step is to use a preset matching algorithm to retrieve relevant technical support documents in the technical document database. This step aims to find technical information that corresponds to the identified core technical elements to support the subsequent generation of new content. For example, in the lithium-ion battery technology in the above example, at this stage of the patent text expansion process, other relevant external documents will be screened based on the previously established knowledge system, such as other research articles or papers in the same field, or even related patents. Algorithms such as cosine similarity measurement and keyword frequency comparison may be needed here to ensure that the documents found are indeed related to the technical point and can be used to expand the original text.

[0071] The third part is to generate novel content related to key technical features through text generation models (which can be language models based on deep learning, such as Transformer architecture). With the above foundation, it is time to generate substantive content. By using advanced text generation mechanisms (such as BERT, T5, etc.), based on the technical support materials that have been obtained and the inherent characteristics of the patent, new content that meets the requirements of the topic and the context is generated to further enrich and improve the framework and structural layout of the entire discussion. Assuming that the above example is to expand on methods to improve battery performance, the system may generate details related to improving energy conversion efficiency, optimizing cycle durability, etc.; at the same time, it will also add a brief description of the scientific basis behind the concepts / views proposed in the newly added part (such as experimental design methods, chemical mechanism analysis).

[0072] In the fourth step, the additional materials and the original patent text are combined to form an updated, richer and more detailed version of the document. At this time, all the previous results are combined and the logical relationship between them is evaluated again to ensure that the content of the patent supplement is highly consistent with the original description and achieves the required accuracy; if necessary, the first to third steps can be run again to further improve the expression of details.

[0073] In general, in the entire process of expanding and enriching patent texts, the use of modern information technology and natural language processing (NLP) tools not only realizes the automatic selection of effective factors from huge amounts of information to support the needs of the extension of existing works, but also ensures that the generated content is both novel and has a rigorous scientific basis through the assistance of intelligent algorithm models. Therefore, this strategy effectively answers the goal requirement of solving the difficult problem of maintaining the relevance and accuracy of technical content when expanding patent texts, and brings better experience effects to subsequent readers or users on the basis of ensuring the integrity and innovative value of the original text.

[0074] Next, the steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features are described, which specifically involve the following key operation steps and their specific manifestations in practice:

[0075] Identify key technologies and vocabulary units from the patent text to be expanded; this step refers to extracting important technical information and terms from a large amount of patent documents or text materials. For example, when reading a patent document on wireless charging technology, the system may find professional terms such as inductive coupling and energy transmission module, as well as key words directly related to specific technical implementation methods.

[0076] A semantic association matrix is ​​constructed to quantify the strength of the relationship between technologies and vocabulary units; this means that an algorithm is used to determine the degree of quantitative connection between the previously extracted technologies and terms. For example, a matrix can be constructed to represent the strength of the mutual association between the core technology concept of inductive coupling and other words such as coupler type or power management strategy.

[0077] Natural language processing technology is used to identify the semantic relationship between terms; at this stage, methods such as syntactic analysis or dependency relationship analysis are used to further refine the technical elements and logical chains in the sentence, and to express their relationships in a structured manner. Taking the wireless charging patent as an example, the relationship between the energy transfer device and the high-frequency transformer in the sentence may be analyzed to confirm whether they constitute the constituent units of the technology.

[0078] Use graph databases to store semantic relationships and key technologies; that is, choose appropriate graph data storage tools to record and index all previously organized technical points and their semantic structures, in order to facilitate subsequent information retrieval and analysis. Assume that all the information mentioned above will be stored on a platform such as the Neo4j graph database, so that users can query technical elements and their relationship chains on demand.

[0079] The representative key technical features are extracted through the depth-first search algorithm of the graph database. This refers to the in-depth exploration in the final stage, using specific methods such as depth-first search DFS to find the most prominent key technical features. Back to the field of wireless charging, after searching, it may be found that the implementation of magnetic induction charging technology is one of the core contents of the entire document, and it is listed as a core technical feature. Through the above steps, we can ensure that we obtain the truly important invention points and technical advantages.

[0080] The above is the basic process of building a semantic knowledge graph and an example of its practical application. This method not only helps to quickly understand the key points of complex patent texts, but also supports data analysis and development trend insights in the field of technological innovation.

[0081] Next, the steps of using the preset matching algorithm to retrieve relevant technical support documents in the technical document database are described, which mainly include the following steps:

[0082] 1. First, the key technical features extracted from the patent text are converted into one or more query strings for subsequent query processing.

[0083] - Example: For example, from a patent application about high-speed signal transmission lines, characteristic words such as low signal attenuation and large bandwidth are extracted and converted into the following query string: signal attenuation low OR largebandwidth.

[0084] 2. Then use fuzzy matching algorithm to further expand keyword queries, improve query flexibility and improve matching quality in different language environments.

[0085] -For example, for the query string previously formed, the system can automatically match documents with similar but not identical expressions, such as: low signal loss or high data throughput. This can better capture the information of related documents.

[0086] 3. Based on fuzzy query, more accurate semantic understanding is further used to score the importance of the found documents to ensure that the technical support documents that best match the target technical requirements are returned.

[0087] -Example: By using word vectorization technology on each article in the search results and calculating the cosine similarity between the feature set and the original text, the similarity value of each file is given. The larger the value, the higher the correlation.

[0088] 4. Based on the obtained relevance evaluation scores, set a certain threshold to eliminate non-matching technical documents, and only retain those materials that are closely related to the topic and have a large amount of information.

[0089] - In actual applications, each discovered file may be assigned a score based on the cosine distance and arranged from high to low to be displayed to the query initiator. When browsing, only the top options (for example, scores greater than or equal to 70%) are presented. This makes it easier to quickly find the most ideal answer while saving time, energy and resources.

[0090] 5. Finally, after sorting by the degree of matching, a series of documents are presented to the user in descending order of importance, making it easier to read, understand, and learn how to use these reference materials to improve project work.

[0091] -For example, in the high-speed transmission line query scenario in the above example, the first few results that are finally screened out after a series of matching and evaluation may be research reports on performance improvement solutions for the same product series, design guidelines for similar products, or common questions and solutions that customers have when using such products.

[0092] This intelligent patent extended document indexing tool, which is built based on key technical information identification and deep comparison technology, can greatly enhance the integration and utilization of text resources in the research and development process and improve overall work efficiency.

[0093] Next, the specific steps of generating novel content related to the key technical features through the text generation model are described. This step mainly includes several important components.

[0094] First, we need to pre-train a deep learning model based on the Transformer architecture to generate natural language sentences. This step is to build a powerful machine learning model that is capable of understanding and generating human language. Take a practical example: suppose we are dealing with a design patent for a new energy-saving circuit. In order to better assist in the description and expansion of this innovative work, we use the Transformer framework model with a self-attention mechanism for training, so that it is familiar with the terminology and technical specifications of the relevant field.

[0095] Subsequently, a conditional probability distribution is constructed to ensure the consistency of the produced content with the input key technical features, that is, to make the content generated by the model closely related to the given key technical features. For example, when the key feature of the technology is low energy consumption in circuit design, the training process will emphasize and ensure that the output text content focuses on this feature and can clearly and accurately describe the relevant technical details and advantages, etc., so as to achieve a high degree of fit with the key technical points.

[0096] Next, constraints need to be introduced to avoid generating existing document content and ensure output independence and originality. This involves using some algorithms or databases to check for duplicate content, confirming that each sentence output generated by the model does not directly copy data from existing documents, and ensuring that each generated information is a new technical discovery, concept explanation, or improvement plan, etc. Continuing with the scenario of the above case, we set up a retrieval mechanism during the training process to find similarities between the generated content and past documents, ensuring that each concept about low-energy design is a novel and unprecedented creative expression.

[0097] At the same time, the requirements of the grammar checker must be followed to ensure the correct structure of the text. This means that in addition to being unique, the generated content must also be fluent and logically reasonable. In practice, we can add additional rules to ensure that the output language conforms to the rules. In this example, we will use tools such as spelling and word order detection to check the accuracy of all sentences and whether they read in accordance with the standards, ensuring that the discourse describing technical features is fluent and easy to understand.

[0098] Finally, apply feedback loop mechanisms to continuously improve the quality of the manuscript - this includes content coherence and depth. This step can be achieved by re-investing user evaluations or automatic evaluation results into the training of the model as a feedback mechanism to achieve self-iteration, so as to gradually generate better sentence fragments, chapters, and entire documents; with the goal of ultimately providing a more in-depth and detailed discussion of the characteristics and advantages of new technology inventions, helping researchers to effectively expand and improve patent descriptions. Taking the low-energy energy-saving circuit project as an example, the continuous iteration process can not only make the content of each new version gradually deepened at the technical level (such as adding different types of electronic component combinations and the performance comparisons they bring), but also make the article as a whole smoother and easier to understand (including chapter transitions, the attractiveness of the introduction, etc.).

[0099] The above steps are accomplished through the coordinated use of specific algorithms, tools and models - aiming to generate technical patent documents with high creativity and practical value while ensuring the quality of the manuscript, which will help to better promote innovative results and lay the foundation for their protection.

[0100] Next, the steps of completing the expansion and enrichment process of the patent text based on the novel content and the initial patent text and ensuring the technical relevance and accuracy of the expanded part are described, which are the core elements of patent expansion and optimization:

[0101] Incorporate and restructure novel content to fit into the structured framework of the patent.

[0102] First, you need to understand the basic concepts of the original invention and identify the key technologies and improvements in the new content; then reorganize this information so that it can be logically integrated into the existing patent text. For example, in a patent related to the field of electronic communications, suppose the inventor has discovered a new algorithm that can reduce the packet loss rate. When expanding the text, you need to ensure that the specific operation of the algorithm is combined with the existing hardware implementation details, and clearly describe the integration and advantages of this technology.

[0103] Application expert review reviews the accuracy of the use of technical terms and makes necessary corrections.

[0104] In this step, professional technicians in the industry check the accuracy and consistency of the professional vocabulary and technical expressions used. Taking the above-mentioned communication algorithm as an example, the review team may include engineers with software engineering backgrounds or researchers in the field; they are responsible for verifying whether specific terms such as algorithm names, bit error rates or delay indicators are accurate and appropriate during use; at the same time, correct possible translation inconsistencies or terminology errors to ensure that the document is clear and accurate.

[0105] Achieve a balance between logical and technical coherence to ensure that the extension is clear and easy to understand.

[0106] This stage emphasizes the natural connection between the various parts of the document and the purpose of ensuring that non-professionals can also understand the entire invention content and its operating mechanism. In this case, the patent writer needs to use concise language to explain how the aforementioned new technology for reducing packet loss rate complements the existing architecture, and provide examples or diagrams to illustrate the working principle and advantages.

[0107] Verify that the added content does not conflict with existing legal documents or prior definitions of patent rights.

[0108] This step requires comparing the expanded text with the requirements of local or international intellectual property laws. Taking the US patent regulations as an example, it is necessary to ensure that the specific parameters and operating environment settings of the new model of communication equipment in the newly added chapters do not violate the current Federal Code, that is, to ensure that it meets the legal novelty conditions and patentable standard regulations.

[0109] Regularly validate the content of new extensions to confirm compliance with the latest patent drafting standards and maintain accuracy.

[0110] In the final stage, the current best writing practices and official guidelines should be reviewed and compared periodically to maintain the standardization of the text. For example, after updating the working mode description of a communication system, it is necessary to check whether the description format and requirements are consistent with the latest version of the Patent Guide or the International Patent Classification Guide, and ensure that the latest developments and trends in the technical fields involved are reflected. This includes not only the above examples, but also factors such as data security and privacy protection.

[0111] Next, the steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features are described.

[0112] The main purpose of this step is to extract the most representative and key information from the original patent text. First, after preprocessing the original patent text, professional natural language processing technology such as Chinese or English word segmenters are used to split the original text into basic units, such as single words and phrases. These basic units form the basis for building a deeper semantic understanding.

[0113] For example, suppose we have a patent text involving wireless charging technology: A method for achieving high-efficiency power transmission using magnetic field resonance... In this step, the sentence will first be segmented into more fine-grained words and phrases such as a, utilizing, magnetic field, resonance, achieving, high efficiency, and power, transmission, and method.

[0114] The second step is to use algorithms such as dependency grammar parsing to determine the logical connections and grammatical structures between words based on the basic units obtained, so as to form a relationship graph that intuitively shows the relationship between words. Continuing with the above example, in the process of relationship identification, it may be found that the use of - magnetic field resonance - to achieve such a connection link means that magnetic field resonance is described as part of a method. In this way, the relationship between the two words "magnetic field" and "resonance" is clearly identified. In the relationship graph, they will appear as a node pair, and the lines between the nodes will have directions to indicate dependencies.

[0115] The third step requires further in-depth analysis of the generated graph to identify the most important technical features and concepts and explore how these elements interact with each other. This step may use graph theory algorithms and specific statistical methods to evaluate the weights of different nodes and connections to quantify the influence of key technical features and their status in the text content.

[0116] Still based on the example mentioned above, further exploration may find that magnetic field and resonance are often mentioned together in the entire technical solution, and there is a strong link relationship with other words, so it can be concluded that these words are core technical features. In addition, attention should be paid to the specific parameters and operating conditions that describe magnetic field resonance, because they play an important role in defining and limiting the scope of the technology.

[0117] The final stage is to screen the most important technical terms after completing all the data analysis and understanding the key elements and their interactions, forming a list of highly relevant words - the keyword library. Keywords can be single words or specific combinations of words as the final technical characteristics. Going back to the situation mentioned above, if the importance of magnetic field resonance for efficient power transmission has been determined through the previous steps, this phrase is likely to be one of the most core technical keywords.

[0118] To sum up the whole process, we can see that it transitions from simple text analysis to complex relationship network construction and deep semantic reasoning, thereby achieving a concise and effective extraction of the key aspects of a technology or an innovation. This process helps to understand and interpret the content and technical value of the patent to be evaluated.

[0119] Next, the method of constructing a relationship graph is described as follows: when the semantic relevance between keywords is greater than 7, it is determined that there is a strong correlation between the keywords, thereby establishing corresponding node connections. This is actually one of the methods that combines natural language processing and graph database technology. This method aims to create a structured knowledge network model based on the similarity or relevance between keywords to help understand and explore the relationship between documents.

[0120] Based on this definition, we can summarize the process into the following steps:

[0121] -The first step is to identify keywords in the text. Use text analysis techniques such as TF-IDF and TextRank algorithms to determine the key words in the text content. For example, given a text discussing the development of artificial intelligence, the system may extract keywords such as machine learning, natural language processing, neural networks, etc.

[0122] -The second step is to calculate the semantic relevance between any two keywords. Semantic relevance can be calculated by using cosine similarity to calculate the similarity between vectors or by using an existing vocabulary (such as WordNet) to measure the semantic distance between words. If the semantic relevance score calculated by machine learning and natural language processing on a certain corpus exceeds 7 points, it is considered that there is a close connection between the two.

[0123] -The last step is to model and represent keywords with strong connections in the relationship graph. For the above example, machine learning and natural language processing are connected as nodes to form part of the knowledge network. As the number of keywords and their relationships increases, a knowledge map that fully describes the technology ecosystem and evolution path will be formed, and further tools such as graph query can be used for advanced data analysis and trend prediction.

[0124] This method is particularly suitable for analyzing complex documents or large data sets. It provides users with more in-depth and broad information services by mining hidden connections between documents, and promotes interdisciplinary research and development collaboration.

[0125] Next, the steps of refining the specific technical points contained in the key technical features based on the relationship map are described. The first is to define the key path formed by the combination of key nodes and their adjacent edges. Key nodes refer to important technologies and innovations in the map, and adjacent edges refer to the relationships connecting these nodes. The key path is a channel formed by a series of these connections from the starting point to a certain end point; for example, in the patent analysis map of the automotive power system, the electric motor node is a component of the core technology innovation, and the connection between the closely related technologies such as the battery management system and the deceleration device can be regarded as one of the key paths.

[0126] Then, two quantitative values ​​are calculated based on the critical path - coverage and importance. The coverage can be simply understood as the number of technical elements passing through this critical path. For example, the coverage in the previous example may be calculated as three (i.e., including electric motors, battery management, and deceleration devices). The evaluation of the importance of the critical path is based on its connection relationship or the accumulation of weights between adjacent nodes. For example, if the connection between the electric motor and the battery management system mentioned above is frequent and tight (meaning that these components are often used together), it can be considered that this connection is stronger, and the corresponding critical path as a whole has more technical advantages. If the electric motor, battery management, and deceleration mechanism are strongly connected, then the overall importance is strong;

[0127] Then, when the product of importance and coverage is greater than zero, it is considered that a key path that can further describe the core technology of the patent has been found. The realization of this step means that the key paths that have passed the previous two steps of verification are sufficiently representative and influential. For example, if we find that a key path involves a wide range of technical fields and has significant correlation and citation numbers in the graph network, we can judge that it accounts for a large proportion of the innovation of the patent text involved.

[0128] Finally, we extract specific in-depth information from the key paths found, and use this to improve the feature description of the key technical points, so that we can have a more accurate grasp and understanding of the content and characteristics of the original text. For example, the aforementioned key technical path related to the electric motor may reveal the core of a new hybrid energy solution. By sorting out the specific technical elements in such a path (such as what type of reducer is used, the type of battery, and how to integrate it efficiently), a more detailed technical description can be obtained, which further helps people understand the unique value of the patented technology.

[0129] Next, the conditional judgment steps in the process of selecting technical points that are not completely the same as the existing patent content, are more specific, and can improve understanding as the final technical features from the important contents screened out based on the key technical features are described as follows:

[0130] The first step is to check the novelty of a particular feature. Determining novelty means evaluating whether the technology has already appeared in previously published documents or patents. If a particular feature has not been described or mentioned in detail in the existing materials, then this feature is considered novel and proceeds to the next step of analysis.

[0131] For example, suppose we have a new sensor for detecting air pollution, and its design uses a new material. We need to check whether similar technologies have been publicly reported; if so, we need to compare the similar technical literature involved with the features in this invention.

[0132] Next, the second step is to use the so-called repetition coefficient to measure the originality of the above features. This coefficient is a quantitative value of the similarity between the technical features and the previous documents in the public knowledge base; if the average repetition coefficient of the description of a technology (that is, the relevance score with existing knowledge) is less than or equal to 4, the technology is considered to be innovative enough to be adopted as a feature.

[0133] Using the sensor example above, if we calculate that the average repetition factor of the information about the adopted new material in the literature database is 2, this means that it is relatively unique and rarely mentioned, and thus can continue to be considered for inclusion in the final shortlist.

[0134] Once novelty is determined, the third stage will be carried out - an in-depth evaluation of the remaining features, focusing on the completeness of the description (whether it can form a set of feasible methods rather than just vague general principles) and its implementation potential. This step aims to remove those parts that only contain general overviews but lack specific guidance for actual application scenarios;

[0135] Continuing with our sensor example, we might notice that some of the literature mentions a particular type of sensor design concept, but none of them mentions how to specifically construct the device and a detailed list of materials required, so the general idea is not adopted, while the new material itself is retained in the list of characteristics with its more detailed processing instructions.

[0136] The final treatment is to give a higher level of attention to all items that have been screened at two levels - identify them as key attributes and strengthen them. The content that is truly practical and innovative after a fine screening process can be clearly distinguished and marked as the most core key technical elements.

[0137] The results obtained through the above screening method will be core features that have high added value and are easy to implement, which is an effective strategy for patent text expansion.

[0138] Next, we describe the specific methods to ensure that the extracted key content is innovative and does not rely entirely on existing documents:

[0139] Ensure that the added content is new by evaluating the topic relevance of the extracted content, for example, a topic relevance coefficient greater than or equal to 75 is considered strong relevance;

[0140] If the new content does not appear frequently enough in the existing technical literature, that is, the frequency of mention in a pre-set literature set is less than or equal to 3%, it is considered to be innovative;

[0141] After the extracted content meets the judgment criteria of novelty and comprehensiveness (judgment of relevance and frequency of mention), its accuracy and completeness are further ensured. If the above conditions are met and the technical description clarity evaluation is higher than 4 / 5, it is judged to be an appropriate technical addition;

[0142] Finally, based on this judgment process, the specific content to be added to the patent text to expand the existing patent document is determined to ensure the technical relevance and accuracy of the new text.

[0143] This method uses multiple steps to determine and select suitable innovative content to be added to the patent text to expand its content. The first step is to ensure that the selected key content is relevant and has a high level of innovation, and the second step is to verify the accuracy and completeness of the content, thereby ensuring that the new content can make an effective contribution to the existing patent document.

[0144] First step - relevance check:

[0145] By calculating the relevance between the proposed content and the core subject of the patent document, for example, using a related algorithm such as LDA (latent Dirichlet allocation), a relevance score or similarity between the two is obtained. If this score is equal to or greater than 75 points, it can be considered that the new content is closely related to the subject to a large extent and has a strong relevance.

[0146] Example description:

[0147] Assume that a new method is used to increase the speed of smartphone processors. Through correlation analysis, it is found that the new method has a high degree of overlap with keywords in existing literature in the processor field, such as CPU optimization, thermal management strategy, and clock signal. Since these terms are often mentioned frequently in technical discussions in related fields and have high scores, it can be determined that their correlation meets the requirement of more than 75 points.

[0148] Step 2 - Innovation Confirmation:

[0149] Check the frequency of the new content in a series of existing related technical documents. If the frequency of citation of the new content in a specific collection of documents is less than 3%, it indicates that it is relatively rare and novel.

[0150] Example description:

[0151] If the method mentioned above is only mentioned in a few documents or does not exist at all in the world's largest patent and scientific paper database, it means that this new technology has not received widespread attention and utilization worldwide. Therefore, it may be one of the new and innovative research results.

[0152] Step 3 - Check accuracy and completeness:

[0153] If the standards of relevance and mention rate have been met, the quality of the technical description part will be reviewed, including the accuracy of language used, the ability to elaborate on technical and functional features, etc. If this technical information can obtain an evaluation result of more than 4 / 5 of a technical review expert, the technical description is clear and reliable and has good technical description quality, and it can be determined that it meets the standards and is suitable for inclusion in the patent text extension part.

[0154] Example description:

[0155] Continuing with the CPU speed-up method example above, the expert review panel will carefully evaluate the key technical details in the description, such as the implementation method and its effect, as well as the possible impact on other related components, and give an evaluation. If the description is detailed, accurate, easy to understand, and has no obvious logical loopholes, it can get a high clarity score.

[0156] To sum up, after layers of checks to ensure that the newly added key technological innovations are accurate, they can be added into the patent document as valuable content, thereby achieving the purpose of enriching the content, while also improving the overall technical level and innovation of the patent document.

[0157] The patent text expansion and enrichment method of the present invention includes the following main steps, which are intended to improve the technical and logical consistency of patent documents during the content expansion process, thereby ensuring the relevance and accuracy of the newly added content:

[0158] 1. Constructing the initial semantic knowledge graph: The core of this step is to conduct an in-depth analysis of the patent text to be expanded, and build an initial semantic knowledge graph based on it, extracting and mapping the key technical features and their relationship networks from the text. This can form a structured knowledge framework as the basis for subsequent technical search and content creation;

[0159] 2. Retrieve relevant technical support documents: Based on the key technical information and structural features extracted from the knowledge graph obtained in the previous step, use efficient, pre-defined matching strategies and methods between large-scale technical literature libraries and technical specification databases to search and locate supporting materials. In this way, new information and background data that best fits the technical points described in the original patent document can be obtained;

[0160] 3. New content generation and integration: Then, using trained and optimized natural language processing (NLP) tools with strong text creation capabilities (such as text generation systems driven by artificial intelligence algorithms such as deep learning), the relevant information collected in the previous step is processed, transformed and integrated into new descriptive text fragments, so that these contents are closely connected with the original patents in technology and have novel and unique value and contribution;

[0161] 4. Complete the expansion and verification work: The last step is to merge the newly added content generated above into the original application documents to realize the synthesis of the final version of the document. At the same time, a series of review and inspection measures are adopted to ensure the integration and coordination of the various components (existing and new) in the entire work and comply with the laws, regulations and technical review standards related to intellectual property rights.

[0162] In this way, by combining a variety of automated and manual interaction methods, the goal of reasonably expanding the scope and details of patent descriptions can be achieved, effectively overcoming various problems and difficulties that may have arisen in previous processes, ensuring the integrity and accuracy of the patent text while also enhancing its practicality.

[0163] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present disclosure. It should be understood that the above description is only the specific implementation method of the embodiments of the present disclosure and is not intended to limit the protection scope of the embodiments of the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the protection scope of the embodiments of the present disclosure.

Claims

1. Patent text expansion and enrichment method based on semantic knowledge graph and matching algorithm, It is characterized in that The method comprises: Build an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features; Retrieve relevant technical support documents from the technical document database using a preset matching algorithm; Generate novel content related to the key technical features through a text generation model; Based on the novel content and the initial patent text, the patent text expansion and enrichment process is completed, and the technical relevance and accuracy of the expanded part are ensured.

2. The method according to claim 1, It is characterized in that The steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features specifically include: Identify key technical and lexical units from the patent text to be expanded; A semantic association matrix was constructed to quantify the strength of relationships between technical and lexical units; Apply natural language processing techniques to identify semantic relationships between terms; Using graph databases to store semantic relationships and key technologies; Representative key technical features are extracted through the depth-first search algorithm of the graph database.

3. The method according to claim 1, It is characterized in that The step of using a preset matching algorithm to retrieve relevant technical support documents in a technical document database specifically includes: Convert the extracted key technical features into query strings; Use fuzzy matching algorithms to improve query flexibility and matching quality; Perform semantic vector comparison to determine the importance of related technical support documents; Set a threshold to filter candidate technical documents with higher matching scores; Search results are sorted by match to return the most relevant technical support documents.

4. The method according to claim 1, It is characterized in that The step of generating novel content related to the key technical features through a text generation model specifically includes: Pre-trained Transformer-based deep learning model for generating natural language sentences; Construct conditional probability distribution to ensure the consistency of produced content and input key technical features; Introduce constraints to avoid generating content that already exists in the literature and ensure the independence and originality of the output; Follow the grammar checker to ensure the grammar of the model output is correct; Use feedback loop mechanisms to iteratively optimize text to improve the logical coherence and technical depth of the content.

5. The method according to claim 1, It is characterized in that The steps of completing the expansion and enrichment process of the patent text based on the novel content and the initial patent text and ensuring the technical relevance and accuracy of the expanded part specifically include: Incorporate and restructure novel content to fit into the patent’s structured framework; Application expert review reviews the accuracy of the use of technical terms and makes necessary corrections; Achieve a balance between logical and technical coherence, ensuring that the extension is clear and easy to understand; Verify that the added content does not conflict with existing legal documents or prior definitions of patent rights; Regularly validate the content of new extensions to confirm compliance with the latest patent drafting standards and maintain accuracy.

6. The method according to claim 1, It is characterized in that The steps of constructing an initial semantic knowledge graph based on the patent text to be expanded to extract key technical features include: Perform word segmentation on the original patent text and break it down into several basic units; Identify the relationship between basic units based on the word segmentation results and construct a relationship graph; Mining the key features and relevance of technical texts by analyzing relationship graphs; Based on the above results, the keywords that best reflect the technical characteristics are selected as technical features.

7. The method according to claim 6, It is characterized in that The method for constructing a relationship graph is specifically as follows: when the semantic relevance between the keywords is greater than 7, it is determined that there is a strong correlation between the keywords, thereby establishing corresponding node connections.

8. The method according to claim 7, It is characterized in that The steps of refining the technical points contained in the key technical features based on the relationship map include: Define the critical path formed by the combination of key nodes and their adjacent edges; Calculate the coverage and importance of features based on the critical path, where coverage is calculated as the number of nodes on the path and importance is defined as the sum of the strengths of the edges on the path; When the result of multiplying the importance value by the coverage value exceeds 0, it is considered that this critical path can provide a deep description of the technology; Extract key information from these deep descriptions to form more precise technical features.

9. The method according to claim 8, It is characterized in that Among the important contents screened out based on the key technical features, the conditional judgment steps in the process of selecting technical points that are not completely the same as the existing patent content, are more specific, and can improve understanding as the final technical features are as follows: Check whether the features are novel and determine whether the technical features have been recorded or mentioned in detail in the public information; Based on the recorded repetition coefficient calculation, where the repetition coefficient is calculated as the average similarity between the feature and other public documents. If the repetition coefficient is less than 4, it is considered novel and can be adopted as a feature; When the novelty is satisfied, the comprehensiveness and feasibility of the feature description will be further examined, and incomplete technical elements that only contain general principle descriptions will be excluded; The elements that remain after passing these two checks are given a higher weight, and the high-value content finally selected is marked as important technical features.

10. The method according to claim 9, It is characterized in that The specific operation method to ensure that the extracted key content is innovative and does not rely entirely on existing files is: Ensure that the added content is new by evaluating the extracted content for topical relevance; If the new content does not appear frequently enough in the existing technical literature, that is, the frequency of mention in a pre-set literature set is less than or equal to 3%, it is considered to be innovative; On the basis that the extracted content meets the judgment criteria of novelty and comprehensiveness, further ensure its accuracy and completeness. If the above conditions are met and the technical description clarity evaluation is higher than 4 / 5, it is judged as an appropriate technical addition; Finally, based on this judgment process, the specific content to be added to the patent text to expand the existing patent document is determined to ensure the technical relevance and accuracy of the new text.

Images