Scientific and technological novelty search report automatic generation method and system based on database and large model

Abstract

The invention discloses a method and system for automatically generating a scientific and technological novelty search report based on a database and large models, and the method comprises the steps: inputting a plurality of content modules into a corresponding large model respectively, outputting a keyword set of the content modules by the corresponding large model, and enabling the plurality of content modules to belong to a to-be-searched text; inputting all the keyword sets into a preset word expansion model, wherein the word expansion model outputs a corresponding expanded keyword set; generating a novelty search report according to all the expansion keyword sets and a preset database; through module division, similarity evaluation and matching degree evaluation, inaccuracy of an analyzed innovativeness result caused by a single evaluation standard can be avoided, the innovativeness of each content module is analyzed from multiple dimensions, and comprehensiveness and accuracy of a report innovativeness evaluation result can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific and technological novelty search, and in particular to a method and system for automatically generating a scientific and technological novelty search report based on a database and a large model. Background Art

[0002] Traditional scientific and technological novelty search reporting methods mainly rely on manual operations and have many limitations. Some current automatic scientific and technological novelty search methods mostly use a single model for text innovation evaluation, without considering the writing characteristics and differences of different modules. In the text comparison process, the evaluation factors for innovation are single, which will cause the evaluation results to be relatively one-sided and affect the comprehensiveness and accuracy of the innovation evaluation results.

[0003] For example, the Chinese patent application with publication number CN118427303A discloses a method, device, electronic device and storage medium for generating a scientific and technological novelty search report, including: characterizing the query text and the text to be matched by a trained large language model, and determining the reference text related to the query text in the text to be matched; extracting the key information of the reference text by a trained large language model and a multi-step prompting method of a thinking chain based on the reference text; generating a novelty search report for the text to be queried by a trained large language model and a multi-step prompting method of a thinking chain based on the text to be queried and the key information.

[0004] The above-mentioned prior art has the problems raised by the background technology: the innovation evaluation is conducted from the perspective of the text as a whole, ignoring the writing characteristics and content features of different modules of the text; in the text comparison process, the evaluation factor for innovation is single, which will cause the evaluation result to be relatively one-sided, affecting the comprehensiveness and accuracy of the innovation evaluation result; in order to solve at least one of the above problems, the present invention proposes a method and system for automatically generating scientific and technological novelty search reports based on a database and a large model. Summary of the invention

[0005] In view of the shortcomings of the prior art, the main purpose of the present invention is to provide a method and system for automatically generating a scientific and technological novelty search report based on a database and a large model, which can effectively solve the problems in the background technology. The specific technical solutions of the present invention are as follows: The automatic generation method of scientific and technological novelty search report based on database and large model includes: Inputting each of the plurality of content modules into a corresponding large model, wherein the corresponding large model outputs a keyword set of the content module, wherein the plurality of content modules belong to a text to be searched; Input all keyword sets into a preset word expansion model, and the word expansion model outputs the corresponding expanded keyword set; According to all the expanded keyword sets and the preset database, the relevant technical texts of each content module are obtained; According to the relevant technical text of each content module and the preset keyword comparison model, the comparison result corresponding to each content module is obtained; Combined with the causal relationship between each content module, the comparison results in each content module are connected by using directed edges to obtain multiple result chains, wherein the directed edges point from the cause node to the result node; A novelty search report is generated based on the multiple result chains and in combination with a preset prompt word model.

[0006] Specifically, all keyword sets are input into a preset word expansion model, and the word expansion model outputs a corresponding expanded keyword set, wherein the word expansion model includes a synonym replacement plug-in, a synonym rewriting plug-in and a word summarization plug-in, including: Using a preset synonym replacement plug-in, all keywords in the keyword set are replaced with synonyms to obtain a synonym set; Through the preset synonym rewriting plug-in, all keywords in the keyword set are rewritten into synonyms to obtain a synonym set; Through the preset word summarization plug-in, all keywords in the keyword set are summarized into hypernyms to obtain a hypernym set; The keyword set, synonym set, near-synonym set and hypernym set are combined to obtain an expanded keyword set.

[0007] Specifically, the relevant technical text of each content module is obtained according to all the expanded keyword sets and the preset database, including: Search in the database according to the expanded keyword set to obtain the search text; Using the preset evaluation model, the search text of each content module is screened respectively to obtain the relevant technical text of each content module.

[0008] Specifically, the preset evaluation model is used to screen the search text of each content module to obtain the relevant technical text of each content module, wherein the evaluation model includes a similarity evaluation model and a matching evaluation model, including: For each content module, the preset similarity evaluation model is used to screen the search texts for similarity, and similar texts of each module are obtained; For each content module, a preset matching evaluation model is used to screen the similar texts for matching, so as to obtain matching texts for each module; The matching texts of the modules are combined to obtain relevant technical texts of the modules.

[0009] Specifically, for each content module, a preset similarity evaluation model is used to perform similarity screening on the search text to obtain similar texts of each module, wherein the similarity evaluation model includes a text vectorization model and a similarity calculation plug-in, including: According to the keyword set, the to-be-queried text and the search text are respectively vectorized by a preset text vectorization model to obtain a to-be-queried text vector and multiple search text vectors; By using a preset similarity calculation plug-in, the similarity between the text vector to be searched and each search text vector is calculated respectively to obtain multiple similarity values; The corresponding search texts whose similarity values ​​are greater than a preset similarity threshold are combined to obtain similar texts.

[0010] Specifically, for each content module, a preset matching evaluation model is used to perform matching screening on the similar texts to obtain matching texts of each module, wherein the matching evaluation model includes a graph construction model and a matching calculation plug-in, including: According to the keyword set, the preset graph construction model is used to construct graphs for the text to be searched and similar texts respectively, so as to obtain a graph of the text to be searched and multiple graphs of similar texts; The matching degree between the to-be-queried text graph and each similar text graph is calculated respectively by a preset matching degree calculation plug-in to obtain a plurality of matching degree values; The corresponding similar texts whose matching values ​​are greater than a preset matching threshold are combined to obtain a matching text.

[0011] Specifically, the comparison result corresponding to each content module is obtained according to the relevant technical text of each content module and the preset keyword comparison model, wherein the content module includes a background content module, a solution content module and an effect content module, including: Extract keywords from the relevant technical texts respectively to obtain a keyword set for each content module of the relevant technology, wherein the keyword set includes keywords and keyword connection relationships; Split the keywords and keyword connection relationships in the keyword set of each content module of the related technology, and put them into the background library, solution library and effect library respectively; Using the preset keyword reorganization model, the keywords in the background library, solution library and effect library are reorganized to obtain multiple reorganized texts; The text to be checked is compared with each reorganized text respectively through a preset similarity comparison model to obtain a comparison result of each content module.

[0012] Specifically, generating a novelty search report based on the multiple result chains and a preset prompt word model includes: By using preset causal constraints, the multiple result chains are screened to obtain a target result chain; According to the target result chain, using a preset prompt word model, logically connect the comparison results of each module in the target result chain to obtain a combined result; Extracting feature vectors from the combination result using a preset feature vector extraction model to obtain a result feature vector; Inputting the result feature vector into a preset text generation model to generate report content; The format of the report content is adjusted through a preset format adjustment model to obtain a novelty search report.

[0013] The system for automatically generating scientific and technological novelty search reports based on a database and a large model is used to implement the method for automatically generating scientific and technological novelty search reports based on a database and a large model, comprising: A keyword extraction module, which inputs each of the multiple content modules into a corresponding large model, wherein the corresponding large model outputs a keyword set of the content modules, wherein the multiple content modules belong to a text to be searched; The keyword expansion module inputs all keyword sets into a preset word expansion model, and the word expansion model outputs the corresponding expanded keyword set; The novelty search module generates novelty search reports based on all expanded keyword sets and preset databases.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention divides the text into modules and analyzes the innovation of multiple modules at the same time, thereby improving the efficiency of novelty search. According to the writing characteristics and content characteristics of each module, the innovation analysis of each module is performed separately to avoid inaccurate innovation results analyzed due to a single standard. The recall rate and precision rate of the report innovation evaluation can be improved through similarity evaluation and matching evaluation, and the innovation of each module is analyzed from multiple dimensions to avoid the one-sidedness of single factor evaluation and improve the comprehensiveness and accuracy of the innovation evaluation results. By automatically generating and formatting the report, it can be ensured that the contents of various parts of the novelty search report are connected in an orderly manner, and automatic generation replaces manual writing, which greatly shortens the report generation time. The generated content follows the scientific and technological writing standards, which can improve the report quality and authority. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The flowchart is a method for automatically generating a scientific and technological novelty search report based on a database and a large model in Example 1 of the present invention; Figure 2 A schematic diagram of the division of text modules in Embodiment 1 of the present invention; Figure 3This is a workflow diagram of the keyword comparison model in Example 1 of the present invention; Figure 4 This is a working schematic diagram of the result combination model in Example 1 of the present invention; Figure 5 This is a schematic diagram of the structure of the automatic generation system of scientific and technological novelty search reports based on database and large model in Example 2 of the present invention. DETAILED DESCRIPTION

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0019] Embodiment 1:

[0020] This embodiment provides a method for automatically generating a scientific and technological novelty search report based on a database and a large model. Figure 1 As shown, the method for automatically generating a scientific and technological novelty search report based on a database and a large model includes: S101, inputting a plurality of content modules into a corresponding large model respectively, wherein the corresponding large model outputs a keyword set of the content module, and the plurality of content modules belong to a text to be searched; S102, inputting all keyword sets into a preset word expansion model, and the word expansion model outputs a corresponding expanded keyword set; S103, obtaining relevant technical texts of each content module according to all expanded keyword sets and a preset database; S104, obtaining a comparison result corresponding to each content module according to the relevant technical text of each content module and a preset keyword comparison model; S105, combining the causal relationship between each content module, connecting the comparison results in each content module using directed edges to obtain multiple result chains, wherein the directed edges point from the cause node to the result node; S106: Generate a novelty search report based on the multiple result chains and a preset prompt word model.

[0021] Since the current scientific and technological reports are written in blocks, including modules such as background technology, technical solutions, and technical effects, the use of a single innovation evaluation standard will ignore the writing differences between different modules, affecting the accuracy of the innovation evaluation results. The present invention provides an automatic generation method for scientific and technological novelty search reports based on a database and a large model, which performs innovation evaluation according to the report content blocks, taking into account the writing habits of different modules and the degree of influence on innovation, such as Figure 2 As shown, the text is divided into three modules according to the titles, and the innovation of the first module, the second module and the third module are evaluated respectively.

[0022] In this embodiment, each text is divided into multiple sections. For example, the background technology section usually includes an explanation of the current industry situation and an analysis of existing problems. The solution content module describes specific technical means and work processes. Different modules have different focuses and writing methods. Therefore, the text to be searched is divided into multiple modules according to the text content and title, and the innovation of each content module is evaluated separately. By dividing the modules, the innovation of multiple modules can be analyzed simultaneously, which can improve the efficiency of novelty retrieval. According to the writing characteristics and content characteristics of each content module, the innovation analysis of each module is performed separately to avoid inaccurate innovation results due to a single standard. The background part contributes less to the innovation evaluation of the report. When the innovation evaluation results of each module are integrated, a smaller weight can be given to this part to achieve dynamic adjustment of the comprehensive innovation evaluation results to meet the needs of different text contents.

[0023] Specifically, the multiple content modules include a background content module, a solution content module and an effect content module. Through the first model, the problem keywords and the logical connection relationship between the problem keywords in the background content module are extracted to obtain a problem keyword set. The core problem-related expressions in the background content module text are deeply analyzed, the key problem keywords are sorted out, and the logical connection relationship between the keywords is analyzed, so as to accurately grasp the core issues surrounding the background technology; through the second model, the solution keywords and the logical connection relationship between the solution keywords in the solution content module are extracted to obtain a solution keyword set, and the key expressions that can reflect specific technical means, technical implementation paths, etc. are identified, the solution keywords are extracted, and the logical relationship between them is analyzed, so as to clearly present the core components and mutual relationships of the technical solution; through the third model, the effect keywords and the logical connection relationship between the effect keywords in the effect content module are extracted to obtain an effect keyword set, and the effect keywords and their logical connection are obtained to clearly show the actual effect brought by the technology.

[0024] Specifically, the preset word expansion model is used, based on the existing problem, solution, and effect keyword sets, and the internal semantic association expansion, synonym replacement, related concept extension and other mechanisms are used to expand the original keywords and enrich the coverage of keywords to improve the comprehensiveness of subsequent retrieval; the expanded keyword set is used as the retrieval basis, and matching queries are performed in a large number of texts in the database to find texts containing these keywords or related to their semantic content, so as to obtain retrieval texts related to the text to be searched; a large number of training texts of the module are used for training to obtain a trained evaluation model for each content module. The evaluation model is trained based on the large model, and the extracted feature word set is compared with the retrieved text, and the similarity and matching are used for evaluation. The similarity evaluation can screen out texts with certain relevance, and the matching evaluation can further screen out highly relevant texts from the relevant texts. Highly relevant texts are of great value for evaluating the innovation of the report, which can improve the recall and precision of the report innovation evaluation. The innovation of each content module is analyzed from multiple dimensions, which can avoid the one-sidedness of single factor evaluation and improve the comprehensiveness and accuracy of the innovation evaluation results.

[0025] According to the relevant technical text of each content module and the preset keyword comparison model, the comparison result corresponding to each content module is obtained. In this embodiment, each content module includes multiple comparison results. There is a causal relationship between the comparison results of different modules. The causal relationship between modules is determined according to the theme and core content of each content module. It can be determined by consulting relevant literature, expert experience and understanding of business processes, and constructing the connection relationship between modules. For example, technical problems in the background will guide the formulation of technical solutions; by constructing the connection relationship, a clear logical context can be formed between the modules of the new search report to avoid loose and chaotic content. The comparison result of each content module is regarded as a node. According to the constructed connection relationship, a directed edge is set from the cause node to the result node, and the results in each content module are connected to obtain multiple result chains.

[0026] Specifically, a novelty search report is generated by combining multiple result chains and prompt word models. After the innovation evaluation results of each content module are calculated separately, the innovation evaluation results of each module are orderly spliced ​​based on the general logical structure and standard format of the scientific and technological novelty search report. The correlation and coherence between the innovation evaluation results of each module are analyzed through a preset text combination model, and combined into a relatively complete and coherent result. Since there may be some incoherence or proper noun errors between the results directly combined, the key feature vectors in the splicing report are extracted using a feature vector extraction model. Based on the key feature vectors, the preset text generation model is used to generate the report content, and the format problems in the report content are adjusted. By automatically generating and formatting the report, it can be ensured that the contents of the various parts of the novelty search report are connected in an orderly manner, and automatic generation replaces manual writing, which greatly shortens the report generation time. In addition, the generated content follows the scientific and technological writing standards, which can improve the quality and authority of the report.

[0027] The present invention can improve the efficiency of novelty search by dividing modules and analyzing the innovation of multiple modules at the same time. According to the writing characteristics and content characteristics of each module, the innovation analysis of each module is carried out separately to avoid inaccurate innovation results analyzed due to a single standard; the recall rate and precision rate of report innovation evaluation can be improved through similarity evaluation and matching evaluation, and the innovation of each module is analyzed from multiple dimensions to avoid the one-sidedness of single factor evaluation and improve the comprehensiveness and accuracy of innovation evaluation results; by automatically generating and formatting the report, it can ensure that the contents of various parts of the novelty search report are connected in an orderly manner, and automatic generation replaces manual writing, which greatly shortens the report generation time, and the generated content follows the scientific and technological writing standards, which can improve the quality and authority of the report.

[0028] Furthermore, all keyword sets are input into a preset word expansion model, and the word expansion model outputs a corresponding expanded keyword set, wherein the word expansion model includes a synonym replacement plug-in, a synonym rewriting plug-in and a word summarization plug-in, including: S201, using a preset synonym replacement plug-in, replacing all keywords in a keyword set with synonyms to obtain a synonym set; S202, rewriting all keywords in the keyword set into synonyms through a preset synonym rewriting plug-in to obtain a synonym set; S203, using a preset word summarization plug-in, summarizing all the keywords in the keyword set to obtain a hypernym set; S204: Combine the keyword set, synonym set, near-synonym set and hypernym set to obtain an expanded keyword set.

[0029] In this embodiment, the word expansion model includes a synonym replacement plug-in, a synonym rewriting plug-in and a word summary plug-in, which respectively perform synonym replacement, synonym rewriting and word hypernym summary on the keywords of each content module. For example, if the keyword is "promote", the plug-in will find synonyms such as "improve" and "enhance" in the dictionary to generate a synonym set; the hypernym summary of the word can analyze the position of each keyword in the lexical semantic network, determine the conceptual category to which it belongs, and then find a more general hypernym. For example, for the keywords "smart phone" and "tablet computer", their hypernym is "mobile terminal device". Through synonym replacement, synonym rewriting and word hypernym summary, the search scope can be expanded, which helps to find a wider range of relevant documents and avoid missing. The original problem keyword set, solution keyword set, effect keyword set and synonym set, synonym set and hypernym set generated by the previous steps are merged. The integrated keyword set makes full use of the various vocabulary variants and concept extensions generated by the previous steps. These vocabulary information can be fully utilized in the subsequent database search, and the search is performed from multiple angles and multiple levels to maximize the possibility of retrieving relevant documents.

[0030] Furthermore, the relevant technical text of each content module is obtained according to all the expanded keyword sets and the preset database, including: S301, searching in a database according to an expanded keyword set to obtain a search text; S302: Using a preset evaluation model, the search text of each content module is screened to obtain relevant technical text of each content module.

[0031] Furthermore, the preset evaluation model is used to screen the search text of each content module to obtain the relevant technical text of each content module, wherein the evaluation model includes a similarity evaluation model and a matching evaluation model, including: S401, for each content module, using a preset similarity evaluation model to perform similarity screening on the search texts to obtain similar texts of each module; S402: for each content module, using a preset matching evaluation model, respectively, the similar texts are screened for matching, and matching texts of each module are obtained; S403: Combine the matching texts of the modules to obtain relevant technical texts of the modules.

[0032] In this embodiment, each content module is evaluated for innovation. The innovation evaluation is divided into similarity evaluation and matching evaluation. The evaluation model includes a similarity evaluation model and a matching evaluation model. For similarity evaluation, the similarity between the text to be checked and the database text is calculated through a pre-trained similarity evaluation model. For example, the similarity value between the text to be checked and a certain database text is 0.7, indicating that the two have a high correlation in semantics and technical topics; the similarity value with another database text is 0.3, then the database text does not contribute much to the evaluation of the text to be checked. By calculating the similarity between the text to be checked and the database text, the degree of closeness between the text to be checked and the existing knowledge can be accurately displayed, breaking the ambiguity of traditional qualitative judgments, and providing intuitive and reliable quantitative indicators for the definition of innovation; it can quickly locate texts similar to the text to be checked in the massive database texts, and improve the comparison efficiency.

[0033] After similarity evaluation, for database texts with high similarity, further calculate the matching degree between the text to be checked and the text with high similarity, evaluate the overlap between the text to be checked and the existing achievements in the database text, and then judge the innovation of the text to be checked. Through the pre-trained matching evaluation model, calculate the matching degree between the text to be checked and the database text with high similarity. For example, the matching value between the text to be checked and a database text is 0.8, which means that the two are highly matched in terms of technical architecture and functional implementation, indicating that the technical solution of the text to be checked is very similar to the existing technology to a certain extent, and then the technical solution of the text to be checked is not highly innovative. By calculating the matching degree, the matching degree between the technology of the text to be checked and the existing technology can be deeply analyzed from the aspects of technical architecture and technical effect, so as to accurately judge the technical innovation of the text to be checked. Filter out the matching texts that are highly relevant to the text to be checked from the large amount of retrieved texts, obtain the relevant technical texts of each module, reduce the subsequent calculation amount, and improve the efficiency of innovation analysis.

[0034] Furthermore, for each content module, a preset similarity evaluation model is used to perform similarity screening on the search text to obtain similar texts of each module, wherein the similarity evaluation model includes a text vectorization model and a similarity calculation plug-in, including: S501, according to the keyword set, vectorize the to-be-queried text and the search text respectively through a preset text vectorization model to obtain a to-be-queried text vector and multiple search text vectors; S502, using a preset similarity calculation plug-in, respectively calculating the similarity between the to-be-queried text vector and each search text vector to obtain a plurality of similarity values; S503: Combine corresponding search texts whose similarity values ​​are greater than a preset similarity threshold to obtain similar texts.

[0035] In this embodiment, the similarity between the text to be searched and the search text is first evaluated. The similarity evaluation model includes a text vectorization model and a similarity calculation plug-in. The text is vectorized according to the keyword set to facilitate the quantitative calculation of the similarity. The text is converted into a numerical vector form that can be efficiently processed by the computer through a pre-trained text vectorization model. When the text to be searched and the search text are input, the model maps them into vectors of fixed length respectively. These vectors can represent the semantic features of the text in a high-dimensional space. The vector dimension can be set according to actual needs. Through text vectorization, texts of different forms and lengths can be uniformly converted into fixed-length vectors to facilitate the calculation of similarity.

[0036] Specifically, by calculating the similarity between vectors, the similarity between corresponding texts is reflected. The cosine similarity between vectors is calculated using a preset similarity calculation plug-in, with a value range of -1 to 1. The larger the cosine similarity value is, the more similar the corresponding texts are. Through similarity calculation, the fuzzy similarity concepts between texts can be converted into specific numerical values, so that the similarity between the text to be searched and the searched text can be quantitatively analyzed. It can quickly perform pairwise calculations on a large number of database text vectors and text vectors to be searched, greatly improving the comparison efficiency. By setting a similarity threshold, texts that are highly relevant to the text to be searched can be screened out from a large number of search texts. The similarity threshold can be set according to actual calculation needs. Texts with similarity values ​​greater than the similarity threshold are selected as similar texts. These texts have a high degree of overlap with the innovative information in semantics. They are used as key comparison texts for subsequent in-depth analysis, which helps to focus on key comparison materials and improve the efficiency of new searches.

[0037] Furthermore, for each content module, a preset matching evaluation model is used to perform matching screening on the similar texts to obtain matching texts of each module, wherein the matching evaluation model includes a graph construction model and a matching calculation plug-in, including: S601, based on the keyword set, using a preset graph construction model, respectively construct graphs for the to-be-queried text and similar texts, to obtain a graph of the to-be-queried text and multiple similar text graphs; S602, using a preset matching degree calculation plug-in, respectively calculating the matching degree between the to-be-queried text graph and each similar text graph to obtain a plurality of matching degree values; S603: Combine the corresponding similar texts whose matching values ​​are greater than a preset matching threshold to obtain matching texts.

[0038] In this embodiment, after obtaining similar texts, the matching degree of the texts is further calculated. The matching degree evaluation model includes a graph construction model and a matching degree calculation plug-in. Through the preset graph construction model, graphs are constructed for the text to be checked and the similar texts respectively. The graph construction model uses natural language processing technology to identify professional terms and key concepts as nodes, and then determines the logical relationship between nodes based on text semantic analysis, such as cause and effect, composition, and application relationships as edges, to construct a graph that reflects the text's technical architecture and functional implementation; through graph construction, the text's disorganized technical information is converted into a clear and rigorously structured graph form, which is convenient for in-depth analysis of the text's innovation.

[0039] Specifically, after constructing the graphs of the to-be-queried text and similar texts, the matching degree between the graphs is quantitatively calculated, and the degree of fit between the two graphs is quantified through the pre-trained matching degree calculation plug-in. The model calculates the matching degree based on the combination of the graph edit distance algorithm and the subgraph isomorphism algorithm, so as to take into account the overall structural similarity of the graphs and the matching of local key technologies. For example, for the to-be-queried text graph of the smart home security system and the similar text graph of the constructed database, the model first calculates the preliminary edit distance value based on the graph edit distance algorithm, and then combines the subgraph isomorphism algorithm to judge the matching of key subgraphs, and comprehensively obtains a matching degree value of 0.6. This operation is repeated for similar text graphs in other databases, and the matching degree values ​​with the to-be-queried text graph are calculated in turn. Through the quantitative calculation of the matching degree, the matching degree between the innovative information and the existing knowledge can be quantified from the underlying details such as the technical architecture and the function implementation process, so as to more accurately reflect the uniqueness of the innovation.

[0040] After calculating the matching value, set the matching threshold and select the database texts with matching values ​​greater than the matching threshold. For example, select 3 highly matching texts from a large number of database texts and use these texts as matching texts. The matching texts cover the core technical solutions, key function implementations and other information that are similar to the innovation points of the text to be checked, which is crucial for accurately judging innovation.

[0041] Furthermore, the comparison result corresponding to each content module is obtained according to the relevant technical text of each content module and the preset keyword comparison model, wherein the content module includes a background content module, a solution content module and an effect content module, including: S701, extract keywords from the relevant technical texts respectively to obtain a keyword set for each content module of the relevant technology, wherein the keyword set includes keywords and keyword connection relationships; S702, splitting the keywords and keyword connection relationships in the keyword set of each content module of the related technology and putting them into a background library, a solution library and an effect library respectively; S703, using a preset keyword recombination model, reorganizing the keywords in the background library, the solution library, and the effect library to obtain a plurality of reorganized texts; S704: Using a preset similarity comparison model, the text to be checked is compared with each reorganized text to obtain a comparison result for each content module.

[0042] In this embodiment, if Figure 3 For the relevant technical texts obtained after screening, the problem keyword set, solution keyword set and effect keyword set in each relevant technical text are extracted respectively. The keyword set includes keywords / sentences and the logical connection relationship (and, or, cause and effect) between keywords / sentences. For example, for the solution content module, the key expressions involving technical implementation methods, process flow, etc. are excavated as solution keywords and corresponding logical connection relationships; the keywords and connection relationships in the extracted keyword set of relevant technical texts are split to obtain independent keywords, and they are placed in the corresponding background library, solution library and effect library. By splitting the keywords and their connection relationships, the keywords involving technical problems are placed in the background library, the relevant technical solutions are stored in the solution library, and the keywords reflecting the technical effects are placed in the effect library. In this way, the information can be sorted according to different technical dimensions, which is convenient for subsequent reorganization and comparison operations.

[0043] Specifically, the solutions or effects between different technical documents have certain correlations. Randomly reorganizing the keywords of different technical documents will result in new technical documents. Through the preset keyword reorganization model, the keywords stored in the background library, solution library and effect library are recombined and matched to generate multiple reorganized texts. It will consider the reasonable semantic coherence and technical correlation between the keywords in different libraries, try to construct various text expressions, and simulate different technical scenarios and situations; by generating multiple reorganized texts, it can create more diverse text forms to simulate various technical situations, increase the number of samples compared with the text to be checked, and thus more comprehensively detect the similarity between the text to be checked and the existing related technologies in different technical expressions and scenarios, and avoid missing important similarities due to a single comparison sample.

[0044] According to the reorganized text, a preset similarity comparison model is used to perform similarity analysis on the text to be checked and each reorganized text in terms of semantics, vocabulary usage, logical structure, etc., and the similarity value between them is calculated. A threshold is set. When the similarity value is greater than the threshold, the existing technology has a huge impact on the technology of the text to be checked. The text to be checked and each reorganized text are compared one by one, and the similarities between the text to be checked and the existing related technologies are analyzed from multiple angles and in multiple technical scenarios. The comparison results of each reorganized text in each content module are obtained respectively, so as to fully understand the innovation of the technology to be checked in the entire technical field, which is conducive to the accurate generation of scientific and technological novelty search reports.

[0045] Furthermore, the generating of a novelty search report based on the multiple result chains and a preset prompt word model includes: S801. Screening the multiple result chains through preset causal constraints to obtain a target result chain; S802: According to the target result chain, using a preset prompt word model, logically connect the comparison results of each module in the target result chain to obtain a combined result; S803, extracting a feature vector from the combination result using a preset feature vector extraction model to obtain a result feature vector; S804, inputting the result feature vector into a preset text generation model to generate report content; S805: Adjust the format of the report content using a preset format adjustment model to obtain a novelty search report.

[0046] In this embodiment, if Figure 4 Each content module includes multiple comparison results. There is a causal relationship between the comparison results of different modules. The causal relationship between modules is determined according to the theme and core content of each content module. It can be determined by consulting relevant literature, expert experience and understanding of business processes, and constructing the connection relationship between modules. For example, technical problems in the background will guide the formulation of technical solutions; by constructing the connection relationship, a clear logical context can be formed between the various modules of the novelty report to avoid loose and confusing content.

[0047] Specifically, the comparison result of each content module is regarded as a node, and a directed edge pointing from the cause node to the result node is set according to the constructed connection relationship. The results in each content module are connected to obtain multiple result chains, among which the logical relationships and results between some result chains cannot fully reflect the innovativeness of the text to be searched. The multiple result chains are screened to select the most logical and representative result chains. According to the preset causal constraints, for example, the causal relationships in the result chains are stipulated to have corresponding credibility; each result chain is evaluated through the preset causal constraint rules to select the most representative target result chain that meets all the constraints. By selecting the target result chain that meets the causal constraints, the logical rigor and accuracy of the content of the novelty search report can be ensured to avoid the inclusion of unreasonable or invalid information.

[0048] Specifically, according to the target result chain and the logical relationship between the result chains, the preset prompt model (prompt model) is used to generate logical keywords, and the logical keywords are used to combine the various comparison results in the target result chain. The logical keywords are inserted into the appropriate positions so that the various results are combined into a complete sentence in a logical order. The discrete module comparison results can be converted into coherent and smooth text content, so that readers can clearly understand the logical deduction process of the report content.

[0049] In this embodiment, a novelty search report is written based on the comparison results between the analyzed text to be searched and each combined text. The comparison results of each content module include the comparison results with each combined text. This embodiment selects the most logical and representative comparison results from multiple comparison results based on the causal relationship between modules, organizes the scattered comparison results according to certain rules, and forms a coherent and logical overall result.

[0050] Further based on the overall results, the report content is automatically generated through a preset text generation model. The text generation model is based on natural language generation (NLG) technology. It makes use of the language knowledge, expression paradigms and logical structures accumulated on massive scientific and technological texts and novelty report examples during the pre-training stage. Through the internal multi-head attention mechanism, it deeply understands the innovation relationship between each module, imitates the professional writing style, and converts the structured innovation points into fluent, professional, and industry-compliant natural language text paragraphs to generate the report content. Through the automatic generation of report content, the structured innovation evaluation can be quickly converted into a complete report. The generated text follows the scientific and technological writing standards, thereby improving the quality and readability of the report.

[0051] Specifically, with the result feature vector as the core, the preset text generation model is used to automatically write and generate the report content. The text generation model is based on natural language generation (NLG) technology. Through the internal multi-head attention mechanism, it captures the innovative associations in the feature vector in all directions, selects precise vocabulary and reasonable sentence patterns from the rich corpus, and converts the structured feature vector into a coherent, professional, and readable natural language paragraph. The result feature vector is input into the text generation model, and necessary metadata is attached according to the text attributes, such as the text name "Research and Development of New High Energy Density New Energy Vehicle Batteries", the field "Intersection of New Energy and Materials Science", and the search requirement keywords "Application of High Nickel Ternary Materials", so that the generated content is closely related to the theme, meets professional requirements, and enhances the professionalism and standardization of the content.

[0052] Specifically, there are some repeated contents or grammatical errors in the directly generated report content, and the report content needs to be corrected and adjusted. Through the preset format adjustment model, the unsmooth and semantically repeated locations in the report content are automatically identified, and the format is adjusted or rewritten. The adjusted novelty search report has a reasonable layout, standardized format, and smooth sentences, thereby increasing the standardization and readability of the novelty search report.

[0053] Furthermore, the report content is adjusted by a preset format adjustment model to obtain a novelty search report, including: S901, according to the report content, perform grammatical structure analysis through a preset grammar analysis model to obtain a text grammar tree; S902, according to the text syntax tree, detecting the incorrect positions of the report content format by using a preset format matching model to obtain text difference points; S903: adjusting the language and format of the text difference points through a preset language adjustment model to obtain a novelty search report.

[0054] In this embodiment, the directly generated report content has grammatical logic incoherence or repeated content, and the report content is logically corrected and adjusted to improve the readability and standardization of the novelty search report. First, the report content is parsed grammatically, and the parts of speech (such as nouns, verbs, adjectives, etc.) in the text content are identified through a preset grammatical analysis model, and the sentence components (subject, predicate, object, attributive, adverbial, and complement) and the dependencies between the components are determined to construct a text grammar tree that reflects the grammatical logic of the text. For example, for the sentence "The application of new high-nickel ternary materials in batteries has significantly improved the energy density, thanks to its unique crystal structure." The model identifies "new high-nickel ternary materials" as a noun phrase as the subject, "application in batteries" as a prepositional phrase as a postpositioned attributive to modify the subject, "significantly improved" as a verb phrase as the predicate, "energy density" as the object, and "this is due to its unique crystal structure" as a causal adverbial clause, and then constructs a text grammar tree containing these nodes and dependencies. By constructing a text syntax tree, the internal logical organization of the text can be reflected. When identifying format errors, the position of the sentence and phrase level can be accurately located instead of blindly searching, thereby improving adjustment efficiency and accuracy.

[0055] Specifically, according to the text syntax tree, the incorrect position of the text content is detected. Through the preset format matching model, based on the pre-set standard format template, the template covers detailed specifications such as font, font size, paragraph format (indentation, line spacing, paragraph spacing), numbering rules, title level style, etc. The format matching model traverses each part of the report text according to each node of the text syntax tree, and compares the actual text presentation style with the template requirements one by one. Using the rule-based matching algorithm, once it is found that the actual format does not meet the standard, such as repeated content in the body paragraph, wrong title font size, etc., the position is immediately marked as a text difference point, and compared with the template through the syntax tree structured guidance, it can quickly and accurately identify problems such as non-standard content in the text.

[0056] Specifically, after identifying the text differences, the erroneous content is corrected, and the language and format of the content of the differences are adjusted through the preset language adjustment model. The language adjustment model integrates rule-based text processing and machine learning-assisted optimization capabilities. For the text content involved in the format differences, on the one hand, the preset rule sets are used, such as terminology unification rules (ensuring consistent wording for the same concept, such as the unification of "artificial intelligence algorithm" and "AI algorithm"), typesetting rules (resetting paragraph indents and line spacing according to standards); on the other hand, machine learning components are introduced to judge the correctness of the spelling of professional terms based on the classifier and use the language model to fine-tune the fluency of sentences; for example, the difference text fragment is input into the language model, and the model is used to predict a better expression, such as optimizing the awkward sentence "By implementing a special treatment process on the material, the purpose of improving battery performance is achieved" to "By adopting a special treatment process on the material, the battery performance is improved", so as to improve the readability of the text; by adjusting the language and format of the text content, the report can be made more professional and standardized, and the readability and quality of the report can be improved.

[0057] Embodiment 2:

[0058] In this embodiment, if Figure 5 , providing a system for automatically generating scientific and technological novelty search reports based on a database and a large model, which is used to implement the method for automatically generating scientific and technological novelty search reports based on a database and a large model, including: A keyword extraction module, which inputs each of the multiple content modules into a corresponding large model, wherein the corresponding large model outputs a keyword set of the content modules, wherein the multiple content modules belong to a text to be searched; The keyword expansion module inputs all keyword sets into a preset word expansion model, and the word expansion model outputs the corresponding expanded keyword set; The novelty search module generates novelty search reports based on all expanded keyword sets and preset databases.

[0059] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for automatically generating a scientific and technological novelty search report based on a database and a large model, characterized in that: include: Inputting each of the plurality of content modules into a corresponding large model, wherein the corresponding large model outputs a keyword set of the content module, wherein the plurality of content modules belong to a text to be searched; Input all keyword sets into a preset word expansion model, and the word expansion model outputs the corresponding expanded keyword set; According to all the expanded keyword sets and the preset database, the relevant technical texts of each content module are obtained; According to the relevant technical text of each content module and the preset keyword comparison model, the comparison result corresponding to each content module is obtained; Combined with the causal relationship between each content module, the comparison results in each content module are connected by using directed edges to obtain multiple result chains, wherein the directed edges point from the cause node to the result node; A novelty search report is generated based on the multiple result chains and in combination with a preset prompt word model.

2. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 1 is characterized in that: All keyword sets are input into a preset word expansion model, and the word expansion model outputs a corresponding expanded keyword set, wherein the word expansion model includes a synonym replacement plug-in, a synonym rewriting plug-in and a word summarization plug-in, including: Using a preset synonym replacement plug-in, all keywords in the keyword set are replaced with synonyms to obtain a synonym set; Through the preset synonym rewriting plug-in, all keywords in the keyword set are rewritten into synonyms to obtain a synonym set; Through the preset word summarization plug-in, all keywords in the keyword set are summarized into hypernyms to obtain a hypernym set; The keyword set, synonym set, near-synonym set and hypernym set are combined to obtain an expanded keyword set.

3. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 1 is characterized in that: The relevant technical text of each content module is obtained according to all the expanded keyword sets and the preset database, including: Search in the database according to the expanded keyword set to obtain the search text; Using the preset evaluation model, the search text of each content module is screened respectively to obtain the relevant technical text of each content module.

4. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 3 is characterized in that: The preset evaluation model is used to screen the search text of each content module to obtain the relevant technical text of each content module, wherein the evaluation model includes a similarity evaluation model and a matching evaluation model, including: For each content module, the preset similarity evaluation model is used to screen the search texts for similarity, and similar texts of each module are obtained; For each content module, a preset matching evaluation model is used to screen the similar texts for matching, so as to obtain matching texts for each module; The matching texts of the modules are combined to obtain relevant technical texts of the modules.

5. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 4 is characterized in that: For each content module, a preset similarity evaluation model is used to perform similarity screening on the search text to obtain similar texts of each module, wherein the similarity evaluation model includes a text vectorization model and a similarity calculation plug-in, including: According to the keyword set, the to-be-queried text and the search text are respectively vectorized by a preset text vectorization model to obtain a to-be-queried text vector and multiple search text vectors; By using a preset similarity calculation plug-in, the similarity between the text vector to be searched and each search text vector is calculated respectively to obtain multiple similarity values; The corresponding search texts whose similarity values ​​are greater than a preset similarity threshold are combined to obtain similar texts.

6. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 4 is characterized in that: For each content module, a preset matching evaluation model is used to perform matching screening on the similar texts to obtain matching texts of each module, wherein the matching evaluation model includes a graph construction model and a matching calculation plug-in, including: According to the keyword set, the preset graph construction model is used to construct graphs for the text to be searched and similar texts respectively, so as to obtain a graph of the text to be searched and multiple graphs of similar texts; The matching degree between the to-be-queried text graph and each similar text graph is calculated respectively by a preset matching degree calculation plug-in to obtain a plurality of matching degree values; The corresponding similar texts whose matching values ​​are greater than a preset matching threshold are combined to obtain a matching text.

7. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 1 is characterized in that: The method obtains the comparison result corresponding to each content module according to the relevant technical text of each content module and the preset keyword comparison model, wherein the content module includes a background content module, a solution content module and an effect content module, including: Extract keywords from the relevant technical texts respectively to obtain a keyword set for each content module of the relevant technology, wherein the keyword set includes keywords and keyword connection relationships; Split the keywords and keyword connection relationships in the keyword set of each content module of the related technology, and put them into the background library, solution library and effect library respectively; Using the preset keyword reorganization model, the keywords in the background library, solution library and effect library are reorganized to obtain multiple reorganized texts; The text to be checked is compared with each reorganized text respectively through a preset similarity comparison model to obtain a comparison result of each content module.

8. The method for automatically generating a scientific and technological novelty search report based on a database and a large model according to claim 1 is characterized in that: The generating of a novelty search report according to the plurality of result chains and in combination with a preset prompt word model comprises: By using preset causal constraints, the multiple result chains are screened to obtain a target result chain; According to the target result chain, using a preset prompt word model, logically connect the comparison results of each module in the target result chain to obtain a combined result; Extracting feature vectors from the combination result using a preset feature vector extraction model to obtain a result feature vector; Inputting the result feature vector into a preset text generation model to generate report content; The format of the report content is adjusted through a preset format adjustment model to obtain a novelty search report.

9. The automatic generation system of scientific and technological novelty search report based on database and large model is characterized by: A method for automatically generating a scientific and technological novelty search report based on a database and a large model as claimed in any one of claims 1 to 8, comprising: A keyword extraction module, which inputs each of the multiple content modules into a corresponding large model, wherein the corresponding large model outputs a keyword set of the content modules, wherein the multiple content modules belong to a text to be searched; The keyword expansion module inputs all keyword sets into a preset word expansion model, and the word expansion model outputs the corresponding expanded keyword set; The novelty search module generates novelty search reports based on all expanded keyword sets and preset databases.

Images