Intelligent scientific and technological information retrieval consultation method and system based on big data

Abstract

The invention relates to the technical field of retrieval analysis, in particular to a scientific and technological information intelligent retrieval consultation method and system based on big data. The method comprises the following steps: obtaining a structured database, an unstructured literature and a patent text corresponding to science and technology information, and carrying out science and technology knowledge semantic association to construct a multi-mode science and technology information knowledge network graph; acquiring a corresponding scientific and technological information retrieval request input by a user to perform retrieval probability analysis and retrieval reasoning analysis so as to generate an optimal retrieval path of the scientific and technological information of the user; performing user retrieval personalized consultation matching on the multi-modal science and technology information knowledge network atlas based on the optimal retrieval path of the user science and technology information to obtain a personalized retrieval consultation result of the user science and technology information; and carrying out science and technology time sequence analysis and science and technology period evolution prediction on the personalized retrieval consultation result of the science and technology information of the user to generate a life cycle evolution path of the science and technology information retrieved by the user. According to the invention, precise retrieval and intelligent consultation of science and technology information can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of retrieval analysis, and in particular to a scientific and technological information intelligent retrieval and consultation method and system based on big data. BACKGROUND

[0002] With the rapid development of information technology, the amount of knowledge and information in the field of science and technology is growing explosively, especially under the promotion of the Internet, big data and artificial intelligence, the acquisition and management of scientific and technological information have become more complex. Researchers, engineers and enterprise managers often need to quickly extract valuable data from a large amount of scientific and technological literature, patents, academic reports, technical standards and other information to support technological innovation, research and development decision-making and market analysis.

[0003] In recent years, intelligent retrieval methods based on big data analysis and artificial intelligence technology have gradually attracted attention. Through the use of natural language processing (NLP), deep learning (DL), machine learning (ML) and other technologies, intelligent retrieval methods can deeply understand the semantics of query requests, identify valuable information hidden in a large amount of data, and significantly improve the efficiency and accuracy of retrieval. However, existing scientific and technological information retrieval systems usually use keyword-based or simple text matching algorithms for retrieval. This method often ignores the semantics and context relationships of information, resulting in the relevance and accuracy of the retrieval results being unable to meet the efficient acquisition of scientific and technological information retrieval services, making it difficult to efficiently process large-scale information across disciplines and disciplines, and thus unable to comprehensively and accurately extract effective scientific and technological information. SUMMARY

[0004] Therefore, it is necessary to provide a scientific and technological information intelligent retrieval and consultation method and system based on big data to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a scientific and technological information intelligent retrieval and consultation method based on big data comprises the following steps:

[0006] Step S1: Obtain the structured database, unstructured literature and patent text corresponding to the scientific and technological information, and perform scientific knowledge semantic association according to the structured database, unstructured literature and patent text corresponding to the scientific and technological information, to construct a multi-modal scientific and technological information knowledge network graph;

[0007] Step S2: Obtain the scientific and technological information retrieval request corresponding to the user input, and perform retrieval technique analysis on the scientific and technological information retrieval request corresponding to the user input to generate a scientific and technological information retrieval vector including the technical field, innovation point and application scenario; based on the multi-modal scientific and technological information knowledge network graph, perform retrieval reasoning analysis on the scientific and technological information retrieval vector to generate an optimal user scientific and technological information retrieval path;

[0008] Step S3: User retrieval personalized consultation matching of the multi-modal scientific and technological information knowledge network graph based on the user scientific and technological information optimal retrieval path, to obtain the user scientific and technological information personalized retrieval consultation result;

[0009] Step S4: Scientific and technological time sequence analysis is performed on the user scientific and technological information personalized retrieval consultation result to generate user retrieval scientific and technological information time sequence arrangement data; scientific and technological cycle evolution prediction is performed according to the user retrieval scientific and technological information time sequence arrangement data to generate a user retrieval scientific and technological information life cycle evolution path.

[0010] Further, step S1 includes the following steps:

[0011] Step S11: Obtain the structured database corresponding to the scientific and technological information;

[0012] Step S12: Obtain the unstructured literature corresponding to the scientific and technological information;

[0013] Step S13: Obtain the patent text corresponding to the scientific and technological information;

[0014] Step S14: Scientific entity recognition and entity attribute analysis are performed on the structured database, unstructured literature and patent text corresponding to the scientific and technological information to obtain each scientific and technological information entity and the scientific and technological attributes between each entity, wherein each scientific and technological information entity includes technical terms, research institutions, scientific researchers and technical indicators; entity relationship network extraction is performed on each scientific and technological information entity based on the scientific and technological attributes between each entity to generate an entity relationship network between each scientific and technological entity;

[0015] Step S15: The potential association relationship between each scientific and technological entity is obtained through each scientific and technological entity prediction, and the entity relationship network between each scientific and technological entity is connected based on the potential association relationship between each scientific and technological entity to generate a scientific and technological information initial knowledge network graph; real-time data stream corresponding to the scientific and technological information is obtained, and the scientific and technological information initial knowledge network graph is updated based on the real-time data stream corresponding to the scientific and technological information to construct a multi-modal scientific and technological information knowledge network graph.

[0016] Further, step S2 includes the following steps:

[0017] Step S21: Obtain the scientific and technological information retrieval request corresponding to the user input;

[0018] Step S22: Construct a retrieval request semantic understanding module based on a BERT pre-training language model, and perform natural language deep analysis on the scientific and technological information retrieval request corresponding to the user input based on the retrieval request semantic understanding module to obtain a scientific and technological natural language retrieval request corresponding to the user input;

[0019] Step S23: Tokenization processing and part-of-speech tagging are performed on the corresponding scientific natural language search request of the user input to generate scientific information basic language features corresponding to the user input search request;

[0020] Step S24: Based on the multi-modal scientific information knowledge network graph, the scientific information basic language features corresponding to the user input search request are retrieved and semantically expanded to generate scientific information retrieval vectors corresponding to the technical field, innovation points, and application scenarios;

[0021] Step S25: Based on the multi-modal scientific information knowledge network graph, the scientific information retrieval vectors are analyzed by retrieval reasoning to generate the optimal retrieval path of the user's scientific information.

[0022] Further, step S24 includes the following steps:

[0023] Scientific and technical language entity recognition is performed on the scientific information basic language features corresponding to the user input search request to identify scientific and technical language entities corresponding to the user input search request, research institutions, and scientific researchers;

[0024] Based on the multi-modal scientific information knowledge network graph, the scientific and technical language entities corresponding to the user input search request are semantically expanded and mapped to the concept nodes in the knowledge graph, wherein the concept nodes include the technical field label, innovation point feature vector, and application scenario description corresponding to the user input search request. The technical field label, innovation point feature vector, and application scenario description corresponding to the user input search request are combined into a retrieval vector to generate scientific information retrieval vectors corresponding to the technical field, innovation points, and application scenarios.

[0025]

[0026] Further, step S25 includes the following steps:

[0027] Step S251: The corresponding different modal scientific information in the multi-modal scientific information knowledge network graph is deconstructed and reconstructed to take the concept nodes corresponding to the different modal scientific information as topological nodes, and the connection edges between the topological nodes are constructed according to the semantic association, logical deduction relationship, and structural similarity between the scientific information to generate a multi-modal scientific semantic topological weaving body;

[0028] Step S252: The scientific information retrieval vector is taken as an attractor entity, and retrieval attraction field mapping calculations are performed between each topological node in the multi-modal scientific semantic topological weaving body and the attractor entity to calculate the attraction degree between the retrieval vector and the corresponding graph different scientific information nodes, and to construct a retrieval vector attraction field mapping matrix;

[0029] ​Step S253: According to the retrieval vector attraction field mapping matrix, the node with the maximum attraction strength is taken as the starting point, and the retrieval conduction path of the scientific and technological information retrieval vector in the multi-modal scientific and technological information knowledge network graph is simulated based on the starting point, so as to follow the corresponding scientific and technological information logical relationship and field knowledge context in the multi-modal scientific and technological information knowledge network graph. After passing through each node, the conduction path is continuously extended according to the weight of the node and the connection relationship with the adjacent node, and a user scientific and technological information retrieval conduction path tree is generated, wherein the branches and nodes of the tree reflect the corresponding potential retrieval propagation path and importance of the scientific and technological information retrieval vector in the multi-modal scientific and technological information knowledge network graph.

[0030] Step S254: The user scientific and technological information retrieval conduction path tree is subjected to retrieval reasoning constraint optimization to generate a user scientific and technological information optimal retrieval path.

[0031] Further, the step S254 comprises the following steps:

[0032] The corresponding professional logical rules and discipline knowledge system of the scientific and technological field are obtained, and the corresponding retrieval request constraint conditions are obtained from the user input corresponding scientific and technological information retrieval request;

[0033] The user scientific and technological information retrieval conduction path tree is pruned and optimized based on the corresponding professional logical rules and discipline knowledge system of the scientific and technological field and the retrieval request constraint conditions, so as to remove the path branches that do not conform to the logical rules, deviate greatly from the retrieval target or correspond to information redundancy. At the same time, the weight adjustment and optimization sorting of the retained path branches are carried out, so as to find the path branch combination that best meets the logic and retrieval request in the pruned user scientific and technological information retrieval conduction path tree by introducing the shortest path algorithm in graph theory, thereby generating a logic constraint pruning and optimization retrieval path tree;

[0034] Each path in the logic constraint pruning and optimization retrieval path tree is subjected to retrieval reasoning constraint optimization based on the scientific and technological information retrieval vector, so as to assign a dynamic weight to each path in the logic constraint pruning and optimization retrieval path tree. The weight is determined by the timeliness, authority and matching degree of the scientific and technological information through weighted calculation, and the retrieval path with the highest comprehensive score is selected from the logic constraint pruning and optimization retrieval path tree, so as to clearly indicate the optimal retrieval path for obtaining the most relevant and high-quality scientific and technological information in the multi-modal scientific and technological information knowledge network graph starting from the scientific and technological information retrieval vector, thereby generating a user scientific and technological information optimal retrieval path.

[0035] Further, the step S3 comprises the following steps:

[0036] Step S31: Based on the user scientific and technological information optimal retrieval path, the multi-modal scientific and technological information knowledge network graph is subjected to user intention retrieval processing, and a user scientific and technological optimal path intention retrieval result is obtained.

[0037] Step S32: Obtain the user's historical search behavior and corresponding historical browsing records, and construct a user's science and technology information interest model based on the user's historical search behavior and corresponding historical browsing records;

[0038] Step S33: Obtain the user's interested corresponding science and technology field and research direction through the user's science and technology information interest model, and perform personalized consulting sorting matching on each science and technology information search result in the user's science and technology optimal path intention search result based on the user's interested corresponding science and technology field and research direction, to obtain a user's science and technology information personalized search consulting result.

[0039] Further, step S4 includes the following steps:

[0040] Step S41: Perform science and technology time sequence analysis on each search recommended science and technology information in the user's science and technology information personalized search consulting result, to generate user search science and technology information time sequence arrangement data;

[0041] Step S42: Perform technical time sequence development mining analysis on the user search science and technology information time sequence arrangement data, to obtain user search science and technology information technology development law data;

[0042] Step S43: Based on the user search science and technology information technology development law data, perform technical development stagnation point identification on the user search science and technology information time sequence arrangement data, to obtain a user search science and technology information technology development time sequence stagnation point;

[0043] Step S44: Based on the user search science and technology information technology development time sequence stagnation point, perform science and technology cycle evolution prediction on the user search science and technology information time sequence arrangement data, to generate a user search science and technology information life cycle evolution path.

[0044] Further, step S44 includes the following steps:

[0045] Perform potential technical bottleneck mining analysis on the user search science and technology information technology development time sequence stagnation point, to obtain a potential technical bottleneck point corresponding to the user search science and technology information time lag;

[0046] Perform technical bottleneck impact assessment according to the potential technical bottleneck point corresponding to the user search science and technology information time lag, to obtain a science and technology potential technical bottleneck impact degree corresponding to the user search science and technology information time lag;

[0047] Based on the user search science and technology information technology development time sequence stagnation point, perform search technology competition detection on the corresponding time lag time node on the user search science and technology information time sequence sorting data, to obtain science and technology competition feedback data corresponding to the user search science and technology information time lag;

[0048] The corresponding user search for scientific and technological information time lag corresponding to the scientific and technological competition feedback data obtains the corresponding user search for scientific and technological competition threat coefficient, and based on the user search for scientific and technological information time lag corresponding to the scientific and technological potential technical bottleneck influence degree and the user search for scientific and technological competition threat coefficient, the scientific and technological life cycle attenuation quantization is carried out on the corresponding time lag time node of the user search for scientific and technological information time sequence sorting data, and the scientific and technological life cycle attenuation amplitude corresponding to the user search for scientific and technological information time lag is obtained.

[0049] Based on the scientific and technological life cycle attenuation amplitude corresponding to the user search for scientific and technological information time lag and in combination with the user search for scientific and technological information technology development law data, the scientific and technological cycle evolution prediction is carried out on the user search for scientific and technological information time sequence arrangement data, so as to generate the user search for scientific and technological information life cycle evolution path.

[0050] Further, the present application also provides a scientific and technological information intelligent retrieval consulting system based on big data, which is used for executing the scientific and technological information intelligent retrieval consulting method based on big data as described above, and the scientific and technological information intelligent retrieval consulting system based on big data comprises:

[0051] A scientific knowledge network construction module is used for acquiring the structured database, unstructured literature and patent text corresponding to scientific information, and performing scientific knowledge semantic association according to the structured database, unstructured literature and patent text corresponding to scientific information, so as to construct a multi-modal scientific information knowledge network graph.

[0052] A user request retrieval reasoning module is used for acquiring the scientific information retrieval request corresponding to user input, and performing retrieval technique analysis on the scientific information retrieval request corresponding to user input, so as to generate the scientific information retrieval vector corresponding to the technical field, innovation point and application scene; the scientific information retrieval vector is subjected to retrieval reasoning analysis based on the multi-modal scientific information knowledge network graph, so as to generate the user scientific information optimal retrieval path.

[0053] A user retrieval consulting module is used for performing user retrieval individual consulting matching on the multi-modal scientific information knowledge network graph based on the user scientific information optimal retrieval path, so as to obtain the user scientific information individualized retrieval consulting result.

[0054] A consulting scientific cycle evolution module is used for performing scientific time sequence analysis on the user scientific information individualized retrieval consulting result, so as to generate the user search for scientific and technological information time sequence arrangement data; and the scientific cycle evolution prediction is carried out according to the user search for scientific and technological information time sequence arrangement data, so as to generate the user search for scientific and technological information life cycle evolution path.

[0055] The present application has the following beneficial effects:

[0056] 1、The intelligent retrieval and consultation method of scientific and technological information based on big data, compared with the prior art, the beneficial effects of the present application are that by obtaining various types of scientific and technological information sources, such as structured databases (such as technical documents, scientific data tables), unstructured documents (such as scientific papers, academic reports) and patent texts, a multi-modal knowledge graph covering various types of scientific and technological information can be constructed, structured data provides efficient retrieval capability and data integration advantage, unstructured documents are rich in detailed background information, innovation point analysis and technology trend, patent texts contain proprietary knowledge of technological innovation and protection content of intellectual property, through in-depth semantic analysis and correlation of these information, a multi-dimensional and multi-level knowledge graph can be constructed, each information node represents a scientific and technological topic, and each edge represents the correlation between the nodes, the graph can reveal the potential connection in the scientific and technological field, help to identify the intersection of the technical field, the innovation point and the scientific research direction which is not widely concerned, and better analyze the semantic and context relationship of scientific and technological information, this process also lays a solid data foundation for subsequent scientific and technological information retrieval and reasoning analysis. Secondly, according to the input request of the user, keyword analysis, context analysis and intention recognition are carried out, so as to extract the specific technical field, innovation point and application scene concerned by the user, this process is called retrieval technology analysis, which helps the system to clearly understand the core direction of user demand, and converts it into a scientific and technological information retrieval vector, based on the retrieval vector, deep reasoning and correlation analysis are carried out in the multi-modal scientific and technological information knowledge network graph, so as to generate the optimal retrieval path, which can guide the user to more accurately find the related field of scientific research, technological development and trend, and make personalized adjustment to the user's demand. For example, if the user is interested in the technical point containing new application scene or breakthrough innovation, the related latest research results and technical scheme will be preferentially pushed, and the content most suitable for the user's demand can be quickly screened out from the mass information, avoiding information overload and invalid retrieval, improving retrieval efficiency and accuracy. Then, according to the generated user optimal retrieval path, further deep matching is carried out on the multi-modal scientific and technological information knowledge graph, and the scientific and technological information most suitable for the user's demand is obtained, through the integration of multi-modal data, customized retrieval consultation results can be provided for the user, covering technical solutions, industry development trend, scientific research literature recommendation and best practice of technical application, the generation of personalized retrieval results not only considers the accurate matching of the current demand of the user, but also combines the past retrieval history, the scientific research field concerned and the personal interest point of the user, so that the retrieval results are more suitable for the actual retrieval service of the user, and the user can also be guided to explore new research direction in the retrieval process, even to propose potential technical application which has not been found by the user, promote scientific and technological innovation and cross-field and cross-disciplinary scientific and technological information knowledge fusion.Finally, by conducting time-series analysis on users' scientific and technological information searches, and arranging user search behaviors and interests chronologically, the lifecycle of their research trends and technological interests can be revealed. This analysis not only helps understand changes in users' interest in scientific and technological information but also predicts future technological evolution paths based on this data. For example, by analyzing users' past search data, it is possible to predict technological development trends in a particular field, predict future research hotspots and technological breakthroughs, and help users understand upcoming technological advancements in advance. This prediction of technological cycle evolution not only provides users with forward-looking knowledge insights but also helps them make more informed decisions when choosing research topics or investing in technology. Furthermore, as the scientific and technological field develops and changes, user needs will also evolve. The recommendation strategy can be adjusted in real time based on this to ensure that the provided scientific and technological information remains at the forefront of the times. This time-series analysis is not only a customized service for users' personalized needs but also an effective tool for promoting the dynamic evolution and accurate prediction of scientific and technological information, thereby enabling the more comprehensive and accurate extraction of effective scientific and technological information.

[0057] 2. The intelligent retrieval and consultation system for scientific and technological information based on big data proposed in this invention is composed of a scientific and technological knowledge network construction module, a user request retrieval and reasoning module, a user retrieval and consultation module, and a consultation technology cycle evolution module. It can realize any intelligent retrieval and consultation method for scientific and technological information based on big data as described in this invention. It is used to combine the operations between computer programs running on each module to realize the intelligent retrieval and consultation method for scientific and technological information based on big data. The internal structure of the system cooperates with each other, which can greatly reduce repetitive work and manpower input, and can quickly and effectively provide a more accurate and efficient intelligent retrieval and consultation process for scientific and technological information based on big data, thereby simplifying the operation process of the intelligent retrieval and consultation system for scientific and technological information based on big data. Attached Figure Description

[0058] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0059] Figure 1 This is a flowchart illustrating the steps of the intelligent retrieval and consultation method for scientific and technological information based on big data according to the present invention.

[0060] Figure 2 for Figure 1 A detailed flowchart of step S1;

[0061] Figure 3 for Figure 1 A detailed flowchart of step S2. Detailed Implementation

[0062] The technical method of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0063] To achieve the above-mentioned purpose, please refer to Figures 1 to 3 The present application provides a big data-based intelligent retrieval and consultation method for scientific and technological information, which comprises the following steps:

[0064] Step S1: Obtain the structured database, unstructured literature and patent text corresponding to the scientific and technological information, and perform scientific and technological knowledge semantic association according to the structured database, unstructured literature and patent text corresponding to the scientific and technological information, so as to construct a multi-modal scientific and technological information knowledge network graph;

[0065] Step S2: Obtain the scientific and technological information retrieval request corresponding to the user input, and perform retrieval technique analysis on the scientific and technological information retrieval request corresponding to the user input, so as to generate a scientific and technological information retrieval vector including the technical field, innovation point and application scene; perform retrieval reasoning analysis on the scientific and technological information retrieval vector based on the multi-modal scientific and technological information knowledge network graph, so as to generate an optimal user scientific and technological information retrieval path;

[0066] Step S3: Perform user retrieval personalized consultation matching on the multi-modal scientific and technological information knowledge network graph based on the optimal user scientific and technological information retrieval path, so as to obtain a user scientific and technological information personalized retrieval and consultation result;

[0067] Step S4: Perform scientific and technological time sequence analysis on the user scientific and technological information personalized retrieval and consultation result, so as to generate user retrieval scientific and technological information time sequence arrangement data; perform scientific and technological cycle evolution prediction according to the user retrieval scientific and technological information time sequence arrangement data, so as to generate a user retrieval scientific and technological information life cycle evolution path.

[0068] In the embodiments of the present application, please refer to Figure 1 The present application provides a big data-based intelligent retrieval and consultation method for scientific and technological information, which comprises the following steps:

[0069] Step S1: Obtain the structured database, unstructured literature and patent text corresponding to the scientific and technological information, and perform scientific and technological knowledge semantic association according to the structured database, unstructured literature and patent text corresponding to the scientific and technological information, so as to construct a multi-modal scientific and technological information knowledge network graph;

[0070] In the embodiment of the present application, the scientific and technological information is obtained from multiple channels through a data interface, a structured database in the field of new energy vehicle technology is downloaded from a government scientific and technological statistics platform, containing scientific research project data in the past five years, each data record contains project number, project unit, research direction, research and development investment and other fields; 100,000 pieces of unstructured literature related to new energy vehicles are extracted from well-known academic websites, covering full-text papers and research reports; 20,000 pieces of new energy vehicle patent text are obtained by connecting with the patent bureau database, including patent name, claim, specification and other contents, the obtained data is associated with scientific and technological knowledge, for structured data, the research direction field of “lithium battery research and development project” is matched with the related papers of “lithium battery materials” in unstructured literature, if the keyword coincidence degree is more than 70%, the association is established. The natural language processing technology is used for unstructured literature and patent text to extract technical terms such as “solid-state battery” and “hydrogen fuel cell”, the semantic similarity between terms is calculated through the word vector model, and the connection is established when the similarity is greater than 0.8. For example, the “solid-state battery” patent is associated with multiple papers studying its performance due to semantic similarity, a multi-modal scientific and technological information knowledge network graph is constructed with technical terms, research institutions, scientific researchers as nodes, semantic association, cooperative relationship and the like as edges, each node is given an attribute, such as the “Ningde Times” node is labeled as “lithium battery research and development enterprise”; each edge is set with a weight, the value is 0-1 according to the association strength, such as the weight of the close cooperative relationship edge is set to 0.9, and the weight of the ordinary reference relationship edge is set to 0.3, finally a complete knowledge network graph is formed.

[0071] Step S2: obtaining a scientific and technological information retrieval request corresponding to the user input, and performing retrieval technology analysis on the scientific and technological information retrieval request corresponding to the user input to generate a scientific and technological information retrieval vector corresponding to a technical field, an innovation point and an application scenario; performing retrieval reasoning analysis on the scientific and technological information retrieval vector based on the multi-modal scientific and technological information knowledge network graph to generate an optimal retrieval path of the user scientific and technological information;

[0072] In the embodiment of the application, by inputting the search request "innovative application of new energy vehicle battery endurance improvement technology in 2022-2024" by the user on the science and technology information retrieval platform, the platform uses the BERT pre-training language model to analyze the request, the model analyzes that "new energy vehicle battery" belongs to the field of "electric vehicle technology", "endurance improvement" is the innovation point direction, and "application" points to the actual use scene, extracts the key terms "new energy vehicle battery", "endurance improvement" and "innovative application" in the request, and queries the related expansion information through the knowledge graph. Find the "lithium battery" and "hydrogen fuel cell" sub-nodes associated with "new energy vehicle battery", and the "energy density optimization" and "lightweight design" technical means associated with "endurance improvement", integrate these information into a science and technology information retrieval vector such as [new energy vehicle battery, lithium battery, hydrogen fuel cell, endurance improvement, energy density optimization, lightweight design, innovative application], based on the multi-modal science and technology information knowledge network graph, starting from the "lithium battery" node in the retrieval vector, according to the edge weight and node association relationship, using graph traversal algorithm to find the path, preferentially selecting the edge with high weight, such as the path from "lithium battery" to "high-nickel ternary material" (weight 0.8), and then to "energy density improvement research" (weight 0.7), after calculation and evaluation, a path containing the most relevant nodes and edges is determined as the optimal retrieval path of the user's science and technology information.

[0073] Step S3: based on the user's science and technology information optimal retrieval path, the multi-modal science and technology information knowledge network graph is matched with the user's retrieval individual consultation, and the user's science and technology information individualized retrieval consultation result is obtained.

[0074] In the embodiment of the application, the user's scientific and technological information is retrieved in a multi-modal scientific and technological information knowledge network graph based on an optimal retrieval path of user's scientific and technological information. For example, the optimal path contains the node "lithium battery-high nickel ternary material-energy density improvement research". The information associated with each node in the graph is extracted, including 100 papers on high nickel ternary material for improving the energy density of lithium batteries and 20 related patents. A user's personalized matching model is constructed. The user's interested technical field "lithium battery" and research direction "material optimization" are extracted from the user's historical retrieval records. The matching degree of each retrieval result with the user's interest is calculated. The formula is: MatchScore = α × TF-IDF + β × SemanticSim, where α and β are weight coefficients (set to 0.6 and 0.4), TF-IDF is the keyword frequency-inverse document frequency score, and SemanticSim is the semantic similarity score. For example, a paper "Application of high nickel ternary material in lithium battery", the calculated TF-IDF is 0.7, and the SemanticSim is 0.8. Then MatchScore = 0.6 × 0.7 + 0.4 × 0.8 = 0.74. The retrieval results are sorted according to the matching degree. The top 50 results with high matching degree are used as the user's scientific and technological information personalized retrieval consultation results, which are displayed to the user as the scientific and technological information that best meets the user's needs.

[0075] Step S4: Scientific and technological time sequence analysis is performed on the user's scientific and technological information personalized retrieval consultation results to generate user's retrieval scientific and technological information time sequence arrangement data. Scientific and technological cycle evolution prediction is performed according to the user's retrieval scientific and technological information time sequence arrangement data to generate the user's retrieval scientific and technological information life cycle evolution path.

[0076] In the embodiment of the application, the scientific and technological time sequence analysis is performed on the 50 information in the user's scientific and technological information personalized retrieval consultation results. The publication time of each information is extracted and converted into a timestamp, such as 1640995200 seconds on January 1, 2022. The information is arranged in ascending order of timestamp to form the user's retrieval scientific and technological information time sequence arrangement data. The technical keywords and abstracts of the information are recorded. The scientific and technological cycle evolution prediction is performed by using the user's retrieval scientific and technological information time sequence arrangement data. The grey prediction model GM(1, 1) is used. Taking the technology of "high nickel ternary material for improving the energy density of lithium batteries" as an example, the number of publications of information related to the technology is selected as the original data sequence X (0) =(x (0) (1),x (0) (2),…,x (0) (n)). It is assumed that X (0) =(10, 15, 20, 25, 30) (representing the number of related papers published each year). The original data is accumulated once to obtain X (1) =(x (1) (1),x(1) (2),…,x (1) (n)),wherein i.e. X (1) =(10,25,45,70,100), the differential equation is constructed to solve the development coefficient a and the gray action b, and a prediction model is obtained After the data is substituted and calculated, the number of the technology-related information published in the next three years is 35, 42 and 50 respectively, according to which it is judged that the technology is in a rapid development period, combined with the technology development law, a user retrieval technology information life cycle evolution path diagram with time as the horizontal axis and technology maturity as the vertical axis is drawn, and the stages such as technology germination period, development period and mature period are marked.

[0077] Further, as one embodiment of the present application, referring to Figure 2 , it is a detailed step flow diagram of step S1 in the embodiment, and step S1 in the embodiment includes the following steps: Figure 1

[0078] Step S11: Obtain the structured database corresponding to the technology information;

[0079] In the embodiment of the present application, the structured database corresponding to the technology information is obtained by establishing a data interface with a plurality of data sources, the input and output data of national and local scientific research projects are obtained from the government science and technology statistical department database, the data is presented in the form of a table, and contains fields such as project number, project time, research direction, funding amount, undertaking unit, etc.; the structured metadata of journal papers and conference papers are accessed from the professional science and technology literature database, covering information such as paper ID, author, publication year, journal name, keyword, abstract, etc.; the enterprise internal technology research and development project data are obtained from the enterprise technology research and development management system, containing fields such as project name, field, research and development progress, personnel involved, patent application situation, etc., for example, in a data synchronization operation, 20,000 scientific research project data, 100,000 paper metadata and 10,000 research and development project data of 500 enterprises are obtained in the past 5 years, and are integrated to form a structured database.

[0080] Step S12: Obtain the unstructured literature corresponding to the technology information;

[0081] ​In the embodiment of the present application, by extracting unstructured documents from a plurality of public academic websites, official websites of scientific research institutions, and science and technology forums, extraction rules are set for academic websites, and full texts of papers, full texts of doctoral theses, etc. are extracted page by page according to journal directory levels, such as obtaining full text content of 2000 doctoral theses in the field of computer science in the past 3 years from China Knowledge Network; for official websites of scientific research institutions, technical reports, research results introduction and other documents are extracted, such as obtaining 100 unpublished technical research reports from the official website of a laboratory; in science and technology forums, technical analysis articles and technical discussion posts published by industry experts are extracted, such as extracting 500 discussion articles related to deep learning algorithm optimization from an artificial intelligence forum. The extracted PDF, HTML, TXT and other format documents are uniformly stored, and finally form an unstructured document library corresponding to science and technology information.

[0082] Step S13: obtaining patent text corresponding to science and technology information;

[0083] In the embodiment of the present application, by establishing a data docking channel with the patent database of the State Intellectual Property Office, the patent text corresponding to the science and technology information is obtained, and the data acquisition range is set, such as obtaining the invention patent and utility model patent text in the field of electronic information in the past 10 years. Each time the data is updated, it is screened according to the patent classification number (IPC), for example, 20000 patent text records under the G06F (electronic digital data processing) classification are screened out. The patent text includes patent name, patent number, abstract, claims, specification and other contents. The specification part describes the technical scheme, implementation mode and other information in detail, such as the specific steps and parameter settings of the image recognition algorithm in the specification of a certain image processing patent. These contents provide important materials for subsequent analysis.

[0084] Step S14: science and technology entity recognition and entity attribute analysis are performed on the structured database, unstructured documents and patent text corresponding to the science and technology information to obtain each science and technology information entity and the science and technology attribute between each entity, wherein each science and technology information entity includes technical terms, research institutions, scientific researchers and technical indicators; based on the science and technology attribute between each entity, entity relationship network extraction is performed on each science and technology information entity to generate an entity relationship network between each science and technology entity;

[0085] In the embodiments of the present application, by acquiring structured databases, unstructured literature and patent texts, the named entity recognition technology is used for scientific and technological entity recognition, so as to directly extract technical terms, research institution names, scientific researcher names and technical index values in the table for structured data, such as extracting “quantum communication”, “Tsinghua University”, “Zhang San Researcher” and “transmission rate 100Gbps” from the scientific research project data table. For unstructured literature and patent texts, the sequence labeling model in natural language processing is used to scan the text sentence by sentence, and technical terms (such as “deep learning neural network”), research institutions (such as “Chinese Academy of Sciences Computer Institute”), scientific researchers (such as “Dr. Li Si”) and technical indicators (such as “accuracy reaches 98%”) are labeled. After completing the entity recognition, the entity attributes are analyzed, such as the field of the technical term, the nature of the research institution, the research direction of the scientific researcher, and the measurement unit of the technical index. Based on these attributes, the entity relationship is extracted through syntactic analysis and semantic understanding, for example, from the sentence “Tsinghua University's Zhang San Researcher leads the development of the quantum communication project, and the project transmission rate reaches 100Gbps”, the relationships “Tsinghua University-Undertaking Unit-Quantum Communication Project”, “Zhang San Researcher-Responsible Person-Quantum Communication Project” and “Quantum Communication Project-Technical Index-Transmission Rate 100Gbps” are extracted. All extracted entities and relationships are integrated, and finally the entity relationship network between various scientific and technological entities is generated.

[0086] Step S15: The potential association relationships between each scientific and technological entity are predicted and obtained, and the entity relationship network between each scientific and technological entity is connected based on the potential association relationships between each scientific and technological entity, to generate a scientific and technological information initial knowledge network graph; real-time data flow corresponding to the scientific and technological information is obtained, and the scientific and technological information initial knowledge network graph is updated based on the real-time data flow corresponding to the scientific and technological information, to construct a multi-modal scientific and technological information knowledge network graph.

[0087] In the embodiment of the present application, by using the generated entity relationship network, the potential association relationship between scientific and technological entities is predicted by using a graph neural network model, the entity relationship network is converted into graph structure data input model, the model learns the existing relationship mode to predict unknown relationship, for example, the model finds that "Peking University" and "artificial intelligence algorithm research" have frequent association, "Fudan University" is also closely related to "artificial intelligence algorithm research", and then predicts that "Peking University" and "Fudan University" have cooperative relationship in the field of artificial intelligence algorithm, and according to the predicted potential association relationship, the entity relationship network is completed, such as adding a connection edge of "Peking University-cooperative relationship-Fudan University" in the network, to generate an initial knowledge network graph of scientific and technological information. At the same time, real-time data stream corresponding to the scientific and technological information is continuously obtained, including newly published papers, newly applied patents, and newly launched scientific research project information. When new data flows in, the entity recognition and relationship extraction are performed according to the method of step S14, and the new entity and relationship are integrated into the initial knowledge network graph, for example, a paper on "new energy battery materials" published by "Zhejiang University" appears, and after extracting the related entity and relationship, a connection such as "Zhejiang University-research results-new energy battery materials" is added in the graph, realizing the incremental update of the scientific and technological information knowledge network graph, and finally constructing a complete multi-modal scientific and technological information knowledge network graph, providing support for scientific and technological information intelligent retrieval and consultation.

[0088] Further, as an embodiment of the present application, referring to Figure 3 , it is Figure 1 a detailed step flow diagram of step S2 in the embodiment, step S2 in the embodiment includes the following steps:

[0089] Step S21: obtaining a scientific and technological information retrieval request corresponding to user input;

[0090] In the embodiment of the present application, multiple input entrances are set to receive user input, and the user can directly input a text retrieval request through the search box on the platform homepage, or trigger the retrieval function on a specific scientific and technological field classification page. For example, a scientific researcher engaged in the field of artificial intelligence inputs "2023-2024 year innovation application of Transformer model in natural language processing field" in the search box, which is transmitted to the platform background server in the form of a text string in real time; another enterprise technical personnel who pays attention to new energy automobile technology clicks the retrieval button on the new energy special page and inputs "solid state battery in new energy automobile's endurance improvement technology research", which is also accurately obtained by the background. The platform unifies the format of all user input retrieval requests, removes redundant spaces, special symbols, etc., to ensure the consistency and accuracy of the request data, and completes the acquisition of the scientific and technological information retrieval request corresponding to the user input.

[0091] Step S22: constructing a search request semantic understanding module based on the BERT pre-training language model, and performing deep natural language analysis on the corresponding scientific information search request of the user input based on the search request semantic understanding module to obtain a scientific natural language search request corresponding to the user input;

[0092] In the embodiment of the present application, by constructing a search request semantic understanding module based on the BERT pre-training language model, the BERT model has been pre-trained on a large amount of general text and scientific literature corpus, and has strong semantic representation capability. The user input scientific information search request such as "2023-2024 year Transformer model in the field of natural language processing innovative application" is input into the model as a fixed length text segment. The model encodes the context information of each word in depth through a multi-layer bidirectional Transformer encoder, and captures the complex semantic dependency relationship between words. For example, the model can understand the technical application association between "Transformer model" and "natural language processing", and the search emphasis direction expressed by "innovative application". After model processing, a vector representation containing semantic information is output, and the vector representation is converted into a structured semantic understanding result through a post-processing module. The core concept, time range, technical field and other key elements in the user search request are determined, so that a scientific natural language search request such as "search for innovative technical application results of Transformer model in the field of natural language processing during 2023-2024" is obtained, and the depth analysis of the user search intention is realized.

[0093] Step S23: performing word segmentation processing and part-of-speech tagging on the corresponding scientific natural language search request of the user input to generate scientific information basic language features corresponding to the user input search request;

[0094] In the embodiment of the present application, by using a professional Chinese word segmentation tool to process the obtained scientific and technological natural language retrieval request, such as "retrieve the innovative technology application achievements of the Transformer model in the field of natural language processing during the period of 2023-2024", the tool based on dictionary matching and statistical learning algorithm, the text is split into "retrieve" "2023-2024" "period" "Transformer model" "natural language processing" "field" "innovative" "technology" "application" "achievements" and other lexical units. At the same time, the part-of-speech tagging algorithm is used to tag the part-of-speech of each word segmentation result, such as "retrieve" is tagged as a verb, "2023-2024" is tagged as a time noun, "Transformer model" is tagged as a noun, and "innovative" is tagged as an adjective. The word segmentation and part-of-speech tagging results are integrated to form structured data containing words, parts of speech and positional relationships between words, thereby generating scientific and technological information basic language features corresponding to the user input retrieval request. These features clearly present the language structure and semantic composition of the user's retrieval request, providing basic data support for subsequent retrieval analysis.

[0095] Step S24: Based on the multi-modal scientific and technological information knowledge network graph, the scientific and technological information basic language features corresponding to the user input retrieval request are retrieved and semantically expanded to generate scientific and technological information retrieval vectors corresponding to the technical field, innovation points and application scenarios;

[0096] In the embodiment of the present application, based on the constructed multi-modal scientific and technological information knowledge network graph, the scientific and technological information basic language features corresponding to the user input retrieval request are retrieved and semantically expanded. Taking "the innovative technology application achievements of the Transformer model in the field of natural language processing" as an example, the "Transformer model" node is found in the knowledge network graph, and its associated technical field information is obtained, which is found to belong to the "machine learning" and "deep learning" technical branches. The "natural language processing" node is found, and it is clear that it covers "text classification", "machine translation", "question and answer system" and other application scenarios. Combined with "innovative technology application achievements", the innovation points in this field in recent years are mined from the graph, such as "multimodal fusion" and "dynamic adaptive architecture", and the related information such as technical field, innovation point and application scenario extracted from the knowledge network graph is fused with the words in the original retrieval request to generate scientific and technological information retrieval vectors containing rich semantic information, for example, "[machine learning, deep learning, Transformer model, natural language processing, text classification, machine translation, question and answer system, multimodal fusion, dynamic adaptive architecture, innovative application achievements]". The semantics of the retrieval request is fully expanded and refined, and the information in the knowledge network graph is more accurately matched.

[0097] Step S25: Based on the multi-modal scientific and technological information knowledge network graph, the scientific and technological information retrieval vector is analyzed by retrieval reasoning to generate the optimal scientific and technological information retrieval path of the user.

[0098] In the embodiment of the application, by analyzing the generated scientific and technological information retrieval vector based on the multi-modal scientific and technological information knowledge network graph, taking the "[machine learning, deep learning, Transformer model, natural language processing, text classification, machine translation, question and answer system, multi-modal fusion, dynamic adaptive architecture, innovative application results]" retrieval vector as an example, each element in the vector is taken as a node clue, and the knowledge network graph is searched. Starting from the "Transformer model" node, along the connection edge of the node with "innovative application results" and "natural language processing", the associated scientific information resource nodes such as papers, patents and technical reports are searched. In the search process, the advantages and disadvantages of different retrieval paths are evaluated according to the association weight and path length between nodes in the knowledge network graph. For example, the path with high connection edge weight (indicating close association) and short path length (indicating direct information association) is preferentially selected. Through the combination of depth-first search and breadth-first search strategy, the entire knowledge network graph is traversed, and finally a path containing the most relevant scientific information resource nodes is determined, and the optimal scientific information retrieval path of the user is generated. For example, starting from the "Transformer model" node, through the "natural language processing-machine translation" branch, a high-impact paper node about the application innovation of Transformer model in machine translation published in 2024 and related patent technology nodes are found, forming a complete and optimal retrieval path that can meet the user's retrieval requirements, providing precise navigation for the user to quickly obtain the required scientific information.

[0099] Further, step S24 includes the following steps:

[0100] The scientific and technological information basis language features corresponding to the user input retrieval request are identified to identify the scientific and technological terms, research institutions and scientific research personnel term entities corresponding to the user input retrieval request;

[0101] In the embodiment of the present application, the scientific and technical language entity recognition is performed on the corresponding scientific and technical information basic language features of the user input search request. Taking the search request "2023-2024, the research progress of quantum error correction codes in the field of quantum computing by the Zhang San team of Tsinghua University" as an example, the basic language features thereof have been segmented and tagged with parts of speech, including "2023-2024" (time noun), "Tsinghua University" (noun), "Zhang San" (noun), "quantum computing" (noun), "quantum error correction code" (noun), "research progress" (noun) and other lexical units. The named entity recognition model based on deep learning is used for scientific and technical language entity recognition. The model has been trained on a large amount of scientific literature corpus and can accurately identify technical terms, research institutions and researchers. The model scans each word in the basic language features in order, analyzes its context information through a multi-layer neural network, and determines the entity category to which the word belongs. For example, for "Tsinghua University", the model identifies it as a research institution entity according to its semantic association and context information in the field of science and technology; for "Zhang San", it identifies it as a researcher entity in combination with the information such as "team" before and after it; for "quantum computing" and "quantum error correction code", they are identified as technical term entities according to their semantic features in the professional field. Finally, the term entities "Tsinghua University" (research institution), "Zhang San" (researcher), "quantum computing" and "quantum error correction code" (technical terms) are identified from the search request, and the scientific and technical language entity recognition process is completed.

[0102] Preferably, the scientific and technical, research institution and researcher term entities corresponding to the user input search request are semantically expanded and mapped based on the multi-modal scientific information knowledge network graph, so as to map the term entities in the user input search request to the concept nodes in the knowledge graph through semantic expansion and mapping, wherein the concept nodes include the technical field label, the innovation point feature vector and the application scenario description corresponding to the user input search request;

[0103] In the embodiments of the present application, the identified scientific and technological technology, research institution and scientific research personnel term entity is mapped and expanded in semantics based on the constructed multi-modal scientific and technological information knowledge network graph, and the term entity such as “Tsinghua University”, “Zhang San”, “quantum computing”, “quantum error correcting code” identified in the previous step is taken as an example. The semantic expansion mapping is performed in the knowledge network graph, for the research institution entity of “Tsinghua University”, the corresponding node of the research institution entity is found in the knowledge graph, all attributes and relationships associated with the node are obtained, and it is found that the research institution entity is closely associated with the technical field label such as “computer science and technology” and “physics” in the knowledge graph, and is also associated with subordinate institution information such as “quantum information research center”. For the scientific research personnel entity of “Zhang San”, the node of the scientific research personnel entity is found in the knowledge graph, it is found that the scientific research personnel entity is associated with the technical term node such as “quantum computing” and “quantum error correcting code” and is also associated with the innovation point feature vector such as “quantum computing field international authority expert”. For the technical term entity of “quantum computing” and “quantum error correcting code”, it is found that they belong to the technical field label of “quantum information technology” in the knowledge graph, and are also associated with more specific technical branches such as “topological quantum error correction” and “surface code”, and innovation point feature vectors such as “fault-tolerant quantum computing” and “quantum bit stability improvement”, and are also associated with application scenario descriptions such as “quantum communication” and “quantum simulation”. The technical field label, innovation point feature vector and application scenario description obtained from the knowledge graph are mapped and associated with the term entity in the original search request, for example, “Tsinghua University” is mapped to the technical field label of “computer science and technology” and “physics”, “Zhang San” is mapped to the innovation point feature vector of “quantum computing field international authority expert”, and “quantum error correcting code” is mapped to the application scenario description of “fault-tolerant quantum computing”, and finally the semantic expansion mapping process is completed.

[0104] Preferably, the technical field label, innovation point feature vector and application scenario description corresponding to the user input search request are retrieved and combined to generate scientific and technological information search vectors corresponding to the technical field, innovation point and application scenario.

[0105] In the embodiments of the present application, the technical field label corresponding to the user input retrieval request, the innovation point feature vector and the application scenario description are combined into a retrieval vector. For example, the semantic expansion mapping result obtained in the foregoing steps is taken as an example, the technical field label includes "quantum information technology", "computer science and technology", and "physics"; the innovation point feature vector includes "international authority expert in the field of quantum computing", "topological quantum error correction", and "fault-tolerant quantum computing"; and the application scenario description includes "quantum communication" and "quantum simulation". When the retrieval vector is combined, first, the features of each semantic category are integrated. For the technical field label, it is converted into a field vector representation. For example, "quantum information technology" is assigned a high value at a specific position in the vector dimension, and "computer science and technology" and "physics" are also assigned appropriate values at the corresponding dimensions. For the innovation point feature vector, the features such as "international authority expert in the field of quantum computing", "topological quantum error correction", and "fault-tolerant quantum computing" are converted into a numerical feature vector, and each feature occupies a specific dimension in the vector. For the application scenario description, the scenarios such as "quantum communication" and "quantum simulation" are converted into a scenario vector representation. Then, the vectors of the three semantic categories are combined, and the technical field vector, the innovation point feature vector and the application scenario vector are sequentially connected into a longer vector in a splicing manner. For example, the technical field vector has 10 dimensions, the innovation point feature vector has 15 dimensions, and the application scenario vector has 8 dimensions. The combined vector has 33 dimensions. In the combination process, appropriate weights are assigned to each semantic category to balance the importance of different categories in retrieval. For example, the weight of the technical field is set to 0.4, the weight of the innovation point feature is set to 0.3, and the weight of the application scenario is set to 0.3. The final generated technology information retrieval vector is "[quantum information technology, computer science and technology, physics, international authority expert in the field of quantum computing, topological quantum error correction, fault-tolerant quantum computing, quantum communication, quantum simulation,...]". The vector contains rich semantic information of the technical field, innovation point and application scenario of the user retrieval request, and provides comprehensive and accurate retrieval basis for subsequent retrieval reasoning analysis.

[0106] Further, step S25 includes the following steps:

[0107] Step S251: deconstructing and reconstructing the corresponding different modal technology information in the multi-modal technology information knowledge network graph to take the concept nodes corresponding to the different modal technology information as topological nodes, and constructing the connection edges between the topological nodes according to the semantic association, logical deduction relationship and structural similarity between the technology information, to generate a multi-modal technology semantic topological weaving body.

[0108] In this embodiment of the invention, a multimodal scientific and technological information knowledge network graph in a certain scientific and technological information retrieval platform contains scientific and technological information of different modalities, such as text (e.g., academic papers, patent documents), images (e.g., technical diagrams, experimental flowcharts), and data tables (e.g., experimental data, statistical reports). This information of different modalities is deconstructed and reconstructed. Taking an academic paper on "new energy battery materials," its corresponding battery structure diagram, and related performance test data tables as an example, conceptual nodes such as "lithium-ion battery," "cathode material," and "energy density" are extracted from the academic paper; concepts such as "electrode structure" and "separator" are identified from the images; and "charge-discharge cycle" is obtained from the data tables. Concepts such as "cycle count" and "voltage plateau" are used as topological nodes. Based on the semantic relationships between scientific and technological information, such as the compositional relationship between "lithium-ion battery" and "cathode material"; logical deduction relationships, such as the impact of "energy density" and "charge-discharge cycle count" on battery performance; and structural similarities, such as common components in different battery structure diagrams, connection edges are constructed between topological nodes. If two nodes have a close semantic relationship and a direct logical deduction, the weight of the connection edge is set to 0.8; if the relationship is weak, it is set to 0.2. Finally, the concept nodes and connection edges corresponding to all different modal scientific and technological information are integrated to generate a multimodal scientific and technological semantic topology weave, which intuitively shows the inherent connections between different modal scientific and technological information.

[0109] Step S252: Take the science and technology information retrieval vector as the attraction source entity, and perform retrieval attraction field mapping calculation between each topological node in the multimodal science and technology semantic topology weave and the attraction source entity, so as to calculate the attraction degree between the retrieval vector and the corresponding different science and technology information nodes in the graph, and construct the retrieval vector attraction field mapping matrix.

[0110] In this embodiment of the invention, assuming a user inputs the search request "Research progress of high energy density new energy battery materials", a scientific and technological information retrieval vector is generated after processing. This vector is used as the attraction source entity. In the multimodal scientific and technological semantic topology weaving, the retrieval attraction field mapping calculation is performed between each topology node and the attraction source entity. The degree of attraction is calculated using the following formula: Where A ij W represents the attraction between retrieval vector i and topological node j. ij To determine the semantic association weight between retrieval vector i and topological node j (determined by calculating the keyword overlap between the two, with a value ranging from 0 to 1), S ij The importance score for topology node j is calculated based on the number and weight of its connecting edges in the topology weave; more connecting edges and higher weights result in a higher score. n is the total number of nodes in the topology weave. For example, for the topology node "lithium-ion battery cathode material", the semantic association weight W with the retrieval vector is... ij =0.7, its importance score Sij =8; calculate its attraction degree A with the retrieval vector ij , and the same calculation is performed on all nodes in the woven body, and the calculation results are arranged into a retrieval vector attraction field mapping matrix, each row of the matrix corresponds to a retrieval vector, each column corresponds to a topology node, and the element value is the attraction degree of the two, which clearly presents the attraction degree relationship between the retrieval vector and the different technology information nodes in the graph.

[0111] Step S253: According to the retrieval vector attraction field mapping matrix, the node with the maximum attraction strength is taken as the starting point, and based on the starting point, the retrieval conduction path of the technology information retrieval vector in the multi-modal technology information knowledge network graph is simulated, so as to follow the corresponding technology information logical relationship and field knowledge context in the multi-modal technology information knowledge network graph, and the conduction path is continuously extended after passing through each node according to the weight of the node and the connection relationship with the adjacent node, and a user technology information retrieval conduction path tree is generated, wherein the branches and nodes of the tree reflect the corresponding potential retrieval propagation path and importance of the technology information retrieval vector in the multi-modal technology information knowledge network graph.

[0112] In the embodiment of the application, according to the previously obtained retrieval vector attraction field mapping matrix, the node with the maximum attraction strength is selected as the starting point, and it is assumed that the "high-nickel ternary positive electrode material" node has the maximum attraction strength. The retrieval conduction path of the technology information retrieval vector in the multi-modal technology information knowledge network graph is simulated based on the starting point, and the technology information logical relationship and field knowledge context in the graph are followed. After passing through each node, the conduction path is extended according to the weight (node importance score) of the node and the connection relationship with the adjacent node. If the weight of the current node "high-nickel ternary positive electrode material" is 7 and the connection edge weight between the adjacent node "high-nickel ternary positive electrode material" and the adjacent node "improving energy density method" is 0.6, the adjacent node is selected as the next conduction node. In the extension process, the nodes and path branches passed through are recorded to form a user technology information retrieval conduction path tree. The root node of the tree is the starting point, the branches represent different extension directions, and the size and color of the node represent the importance (the higher the importance, the larger the node and the darker the color). For example, the final generated conduction path tree contains multiple branch paths extending from "high-nickel ternary positive electrode material" to "material preparation process" and "performance optimization strategy", which clearly reflects the potential retrieval propagation path of the retrieval vector in the graph and the importance of each path and node.

[0113] Step S254: The user technology information retrieval conduction path tree is optimized by retrieval reasoning constraint to generate an optimal user technology information retrieval path.

[0114] In the embodiments of the present application, the generated user technology information retrieval conduction path tree is retrieved and reasoned to optimize constraints, for example, setting optimization constraints such as the length of the retrieval path being no more than 5 nodes, the average weight of the nodes in the path being greater than 4, calculating the comprehensive score of each path in the path tree, and the formula is: Score = a x W avg + b x L + g x S end , wherein Score is the comprehensive score of the path, W avg is the average weight of the nodes in the path, L is the length of the path (the reciprocal is taken to ensure that the shorter the length, the higher the score), S end is the importance score of the end node of the path, a, b, g are weight coefficients (set to 0.4, 0.3, 0.3 respectively), for example, path 1 contains 3 nodes, the node weights are 5, 6, and 7 respectively, and the end node importance score is 7, the comprehensive score is calculated as Score = 0.4 x (5+6+7) / 3 + 0.3 x 1 / 3 + 0.3 x 7 ≈ 5.1; all paths are calculated, and the path with the highest comprehensive score that meets the constraint condition is selected as the optimal retrieval path of the user technology information, for example, path 1 has the highest score and meets the constraint, so it is determined as the optimal retrieval path, which provides accurate guidance for user retrieval.

[0115] Further, the step S254 comprises the following steps:

[0116] Obtain the corresponding professional logical rules and subject knowledge system in the field of science and technology, and obtain the corresponding retrieval request constraint conditions from the user input corresponding science and technology information retrieval request;

[0117] In the embodiments of the present application, the professional logical rules and subject knowledge system are constructed in advance for different fields of science and technology, for example, in the field of artificial intelligence, the professional logical rules include logical relationships such as “deep learning belongs to the branch field of machine learning” and “natural language processing is an important application direction of artificial intelligence”; the subject knowledge system covers a complete knowledge architecture from basic theory (such as neural network structure, algorithm principle) to application technology (such as machine translation, intelligent question answering system). At the same time, the retrieval request constraint conditions are extracted from the user input science and technology information retrieval request, for example, the user inputs “2022-2024 year in the top conference published papers on the application of large language model in medical question answering system”, wherein “2022-2024 year” is a time constraint condition, “top conference published” is an authority constraint condition, and “large language model” and “medical question answering system” are technical field and application scenario constraint conditions. The professional logical rules, subject knowledge system corresponding to the science and technology field, and the constraint conditions extracted from the retrieval request are integrated to provide a basis for subsequent retrieval path optimization.

[0118] Preferably, the user's science and technology information retrieval conduction path tree is pruned and optimized based on the corresponding professional logic rules in the science and technology field and the discipline knowledge system and the retrieval request constraint condition, to remove the path branches that do not meet the logic rules, deviate greatly from the retrieval target, or correspond to redundant information, while the remaining path branches are weight-adjusted and optimized sorted, to find the path branch combination that best meets the logic and retrieval request in the pruned user's science and technology information retrieval conduction path tree by introducing the shortest path algorithm in graph theory, so as to generate a logic constraint pruned and optimized retrieval path tree;

[0119] In the embodiment of the present application, the user's science and technology information retrieval conduction path tree is pruned and optimized based on the previously acquired professional logic rules, discipline knowledge system and retrieval request constraint condition. Assuming that the user's science and technology information retrieval conduction path tree is generated in the multi-modal science and technology information knowledge network graph according to the user's retrieval request, it contains numerous path branches, each of which represents a possible retrieval path. First, according to the professional logic rules and the discipline knowledge system, the path branches that do not meet the logic are removed. For example, if a path branch extends from the "computer graphics" node to the "application of large language models in medical question answering systems", since "computer graphics" and "application of large language models in medical question answering systems" have no direct connection in professional logic, this path branch is determined to be removed because it does not meet the logic rules. Then, according to the retrieval request constraint condition, the path branches that deviate greatly from the retrieval target are removed, such as a path branch that points to a paper on large language models published before 2021, which does not meet the "2022-2024" time constraint condition, so it is deleted. For path branches with redundant information, if multiple branches point to the same paper or similar research content, only one of them is retained. After removing the path branches that do not meet the requirements, the remaining path branches are weight-adjusted and optimized sorted, and the Dijkstra shortest path algorithm in graph theory is introduced to calculate the "distance" of each path branch to the target node, taking the matching degree of the nodes contained in the path branch with the retrieval request constraint condition and the close degree of association in the knowledge network graph as the measurement standard of path length. For example, a path branch starts from the "large language model" node, passes through the "medical field application" node, and finally points to a paper node that meets the time and authority requirements. The connection edge weight in the graph is high, and the matching degree with the retrieval request constraint condition is high. Through algorithm calculation, the "distance" of this path branch is shorter, and it is given a higher weight and placed in a front position. After the above operation, a logic constraint pruned and optimized retrieval path tree is finally generated.

[0120] Preferably, based on the technology information retrieval vector, each path in the logical constraint pruning and optimization retrieval path tree is optimized for retrieval reasoning constraint optimization, and each path in the logical constraint pruning and optimization retrieval path tree is assigned a dynamic weight, which is determined by the timeliness, authority and matching degree of the technology information with the technology information retrieval vector through weighted calculation, and the highest comprehensive score is screened out from the technology information retrieval vector, to clearly indicate the optimal retrieval path for obtaining the most relevant and high-quality technology information in the multimodal technology information knowledge network graph from the technology information retrieval vector, to generate the user technology information optimal retrieval path.

[0121] In the embodiment of the application, based on the previously generated technology information retrieval vector, each path in the logical constraint pruning and optimization retrieval path tree is optimized for retrieval reasoning constraint optimization. Taking the retrieval vector related to "Application Research of Large Language Model in Medical Question Answering System" as an example, the vector contains semantic information such as technical field, innovation point and application scenario, and each path in the logical constraint pruning and optimization retrieval path tree is assigned a dynamic weight. The determination of the weight comprehensively considers the timeliness, authority and matching degree of the technology information with the technology information retrieval vector. In terms of timeliness, the weight coefficient of technology information in the past three years is set to 0.3. If a paper pointed to by a path is published in 2023, it will get a higher score in the timeliness dimension. In terms of authority, the impact factor of the journal where the paper is published and the conference level are used as the measurement standard. Papers published in "top conferences" have a weight coefficient of 0.4 in the authority dimension. The matching degree is determined by calculating the coincidence and correlation strength of the nodes involved in the path and the semantic information in the retrieval vector. The path with high matching degree has a weight coefficient of 0.3 in this dimension. For example, path A points to a paper published in a top conference in 2023, which matches the retrieval vector well. Its timeliness score is 0.9 (full score 1), authority score is 0.8, and matching degree score is 0.8. The comprehensive score is calculated as 0.3x0.9+0.4x0.8+0.3x0.8=0.83. Path B points to a paper published in a general journal in 2022, which has an average matching degree with the retrieval vector. Its timeliness score is 0.8, authority score is 0.4, and matching degree score is 0.6. The comprehensive score is calculated as 0.3x0.8+0.4x0.4+0.3x0.6=0.58. By comparing the comprehensive scores of all paths, the path with the highest score, path A, is selected. This path clearly indicates the optimal retrieval path for obtaining the most relevant and high-quality technology information in the multimodal technology information knowledge network graph from the technology information retrieval vector, and finally generates the user technology information optimal retrieval path.

[0122] Further, step S3 includes the following steps:

[0123] Step S31: Based on the optimal retrieval path of user scientific and technological information, perform user intent retrieval processing on the multimodal scientific and technological information knowledge network graph to obtain the user's optimal scientific and technological path intent retrieval results;

[0124] In this embodiment of the invention, user intent retrieval processing is performed on a multimodal scientific and technological information knowledge network graph based on the optimal retrieval path of user scientific and technological information in a certain scientific and technological information retrieval platform. Taking the optimal retrieval path generated by the user's search request "research progress of high energy density new energy battery materials" as an example, this path includes nodes such as "high-nickel ternary cathode material → material preparation process → performance optimization strategy → latest research results". The knowledge network graph is searched along this path, and the scientific and technological information associated with each node is filtered and matched. In the "high-nickel ternary cathode material" node, information such as academic papers and patent documents directly connected to it and with a weight higher than 0.7 is extracted; in the "performance optimization strategy" node, research on specific methods for improving battery energy density published in the past three years is collected. For each retrieved piece of information, its semantic relevance to the path node is calculated using the following formula: Where R ij C represents the degree of association between information j and node i. ij C represents the number of keywords in information j that overlap with those in node i. i总 Let S be the total number of keywords for node i. For example, if the keyword overlap between a paper and the node "performance optimization strategy" is 0.8, then the correlation between the paper and that node is 0.8. After calculating the correlation of all information associated with nodes along the path, information with a correlation greater than 0.6 is selected as candidate results. Then, the information is weighted and ranked according to its timeliness (the more recent the publication time, the higher the weight) and authority (journal impact factor, number of patent citations, etc.). The formula is: S j =α×R j +β×T j +γ×A j S j R is the overall score for information j. j To score the relevance, T j For timeliness score (value 0-1), A j The authority score (value 0-1) is used, and α, β, and γ are weighting coefficients (set to 0.5, 0.3, and 0.2 respectively). Finally, the top 50 information with the comprehensive score are selected to form the user's optimal technology path intent retrieval results.

[0125] Step S32: Obtain the user's historical search behavior and corresponding historical browsing records, and construct a user's science and technology information interest model based on the user's historical search behavior and corresponding historical browsing records;

[0126] In the embodiment of the present application, the user's historical search behavior and the corresponding historical browsing records are acquired to construct a user's science and technology information interest model. It is assumed that the user has performed 20 times of science and technology information retrieval in the past 3 months, and has browsed 100 academic papers, 30 patents and 20 technical reports. The keywords such as "solid-state battery", "electrolyte material" and "cycle life" are extracted for each retrieval request. The keywords in the title and abstract of each browsed science and technology information are extracted, and different weights are given according to the information type. The keyword weight of the academic paper is 0.8, the keyword weight of the patent is 0.7, and the keyword weight of the technical report is 0.6. The frequency of each keyword is counted to form an initial keyword frequency vector, and the interest score of the keyword is calculated. The formula is: Wherein I k is the interest score of the keyword k, F k,i is the frequency of the keyword k in the i-th information, W i is the type weight of the i-th information, T i is a time decay factor (recent information weight is high, such as 1 month information T i = 1, 2 months information T i = 0.8, 3 months information T i = 0.6). For example, the keyword "solid-state battery" appears in 5 academic papers within 1 month with frequencies of 0.3, 0.4, 0.2, 0.5 and 0.3, respectively. Its interest score I 固态电池 = (0.3 x 0.8 x 1 + 0.4 x 0.8 x 1 + 0.2 x 0.8 x 1 + 0.5 x 0.8 x 1 + 0.3 x 0.8 x 1) = 1.36. After calculating the interest scores of all keywords, the user's science and technology information interest model is formed, which clearly shows the user's focus on the science and technology field (such as "solid-state battery" field) and research direction (such as "electrolyte material optimization").

[0127] Step S33: The user's science and technology field and research direction of interest are obtained through the user's science and technology information interest model, and each science and technology information retrieval result in the user's science and technology optimal path intention retrieval result is sorted and matched for personalized consultation based on the user's science and technology field and research direction of interest, to obtain the user's science and technology information personalized retrieval consultation result.

[0128] In the embodiment of the present application, the user's science and technology information interest model is used to sort and match each science and technology information retrieval result in the user's science and technology optimal path intention retrieval result for personalized consultation. The keywords and their interest scores in the user's interest model are compared with the keywords in the retrieval result, the matching degree of the retrieval result and the user's interest is calculated, and the formula is: Wherein M jM is the matching degree of the retrieval result j to the user interest, I k R is the interest degree score of the keyword k jk W is the weight of the keyword k in the retrieval result j (different weights are given according to the position of the keyword in the title, abstract and body, the weight in the title is 0.8, the weight in the abstract is 0.5, and the weight in the body is 0.3). For example, a paper contains the keyword "solid-state battery" (interest degree score 1.2) in the title and the keyword "electrolyte material" (interest degree score 0.9) in the abstract, then the matching degree M j of the paper to the user interest is 1.2*0.8+0.9*0.5=1.41. After calculating the matching degree of all retrieval results, the final ranking score is formed by combining the comprehensive score S j of the retrieval result, and the formula is: FinalScore j =δ*S j +∈*M j where δ and ∈ are weight coefficients (0.6 and 0.4 respectively), and the retrieval results are arranged in descending order according to the final ranking score, and the results with higher scores are displayed first. For example, a paper on "Application of Solid-state Electrolyte Materials in High-energy-density Batteries" has a comprehensive score S j =0.8 and a matching degree M j =1.6, so the final ranking score FinalScore j =1.12, and it is ranked in the front of the result list. In this way, the user's personalized retrieval consulting results of scientific and technological information are obtained, and the scientific and technological information that meets the user's interest and needs is ranked and displayed.

[0129] Further, step S4 includes the following steps:

[0130] Step S41: performing scientific time sequence analysis on each retrieval recommended scientific and technological information in the user's personalized retrieval consulting results of scientific and technological information to generate user retrieval scientific and technological information time sequence arrangement data;

[0131] In the embodiment of the application, by taking the personalized search results of "high energy density new energy battery materials" obtained by the user on the science and technology information retrieval platform as an example, the results contain 50 academic papers, patents and technical reports and other science and technology information, the publication or disclosure time of each search recommended science and technology information is extracted, such as the paper "Interface optimization research of high-nickel ternary positive electrode materials" published in 2022, and the patent "Preparation method of high energy density solid state battery" disclosed in 2023, these information is arranged in chronological order to form the user search science and technology information time sequence arrangement data, in order to further quantify the time characteristics, the time stamp conversion method is used to convert the date format data into the number of seconds since a specific starting date (such as January 1, 2000, 0:00:00), assuming that the time stamp corresponding to January 1, 2022, 0:00:00 is t1 = 693561600 seconds, and the time stamp corresponding to January 1, 2023, 0:00:00 is t2 = 725097600 seconds, for the above paper and patent, the time stamps t_paper and t_patent are respectively assigned, and the time stamps are sorted according to the size, and the key content abstract, involved technical points and other attributes of each information are recorded, and finally a structured user search science and technology information time sequence arrangement data table is formed, containing time stamp, information type, technical keyword, content abstract and other fields.

[0132] Step S42: performing technical time sequence development mining analysis on the user search science and technology information time sequence arrangement data to obtain user search science and technology information technology development rule data;

[0133] In the embodiment of the application, based on the previously generated user search science and technology information time sequence arrangement data, technical time sequence development mining analysis is performed, taking the technical field of "high energy density new energy battery materials" as an example, the frequency and time distribution of the appearance of technical keywords are extracted from the time sequence arrangement data, such as the keywords "high-nickel ternary material", "solid state electrolyte" and "nano structure", the appearance frequency change rate of each keyword in different time intervals is calculated, and the formula is: Wherein R k,r is the frequency change rate of keyword k in time interval t, F k,t is the appearance frequency of keyword k in time interval t, F k,t-1The frequency change rate of "solid-state electrolyte" is 0.5, that is, an increase of 50%, by analyzing the frequency change trend of a plurality of keywords, combining the technical content summary, summarizing the technology development law, finding that the research on "high-nickel ternary material" was mainly focused on performance improvement from 2018 to 2020; the research on "solid-state electrolyte" rapidly increased from 2021 to 2023 and gradually became a hotspot; the application of "nano structure" in battery materials began to be explored after 2023. These laws are arranged into technology development law data, including technology hotspots, technology evolution direction, key technology breakthrough points and other information in different time periods.

[0134] Step S43: Based on the user search technology information technology development law data, the technology development stagnation point of the user search technology information time sequence arrangement data is identified to obtain the user search technology information technology development time sequence stagnation point.

[0135] In the embodiment of the application, based on the previously obtained user search technology information technology development law data, the technology development stagnation point of the user search technology information time sequence arrangement data is identified, and the stagnation point judgment standard is set. When the frequency change rate of a certain technology keyword is less than 0.1 and lower than the average change rate of the technology field for two consecutive time intervals, it is determined that the technology development in the time period appears stagnation. Taking the "high-nickel ternary material" technology as an example, assuming that its frequency change rate in 2023-2024 is 0.08, and the average change rate of the technology field is 0.2, which meets the stagnation point judgment standard, and it is determined that 2023-2024 is the stagnation point of the development of "high-nickel ternary material" technology. The above judgment is made for all technology keywords in the time sequence arrangement data, and the time interval, involved technology name and related information of each stagnation point are recorded. In addition to the frequency change rate, the innovation of the technical research content and the influence of the research results are also considered to assist in the judgment. If the research on a certain technology in a certain time period is mostly repetitive verification experiment and lacks innovative breakthrough, it is also regarded as a technology development stagnation point. Finally, the user search technology information technology development time sequence stagnation point list is formed, and the stagnation stage in the technology development process is clear.

[0136] Step S44: Based on the user search technology information technology development time sequence stagnation point, the technology cycle evolution prediction of the user search technology information time sequence arrangement data is performed to generate the user search technology information life cycle evolution path.

[0137] In the embodiments of the present application, the technology cycle evolution prediction of the user search technology information time sequence arrangement data is performed based on the previously obtained user search technology information technology development time sequence stagnation point, the method of combining analogy analysis and trend extrapolation is adopted, the technology of "high energy density new energy battery material" is taken as an example, the development cycle of similar technologies in history (such as the development cycle of lithium ion battery technology from early research and development to mature application) is referred to, the current technology development stagnation point and law are combined, the future development stage is predicted, it is assumed that the lithium ion battery technology has experienced 20 years from the concept to the large-scale application, among which the early research and development (more technical stagnation points) accounts for 8 years, the rapid technology development period accounts for 10 years, and the mature period accounts for 2 years. Compared with the technology of "high energy density new energy battery material", which has developed for 8 years and is currently in the technical development stagnation stage, it is predicted that it will still be in the technical bottleneck breakthrough period for 2-3 years, and then enter the rapid development stage for 5-7 years, and finally reach the technology maturity. In order to quantify the prediction result, the technology maturity index M t is set, and the calculation formula is: Wherein T progress is the developed time of technology, T total is the predicted complete development cycle time of technology, it is assumed that the complete development cycle of the predicted "high energy density new energy battery material" technology is 15 years, and it has developed for 8 years, so the current technology maturity M t ≈0.53, according to the information of technology maturity, technology hotspot change and stagnation point occurrence rule in different stages, the user search technology information life cycle evolution path diagram is generated, taking time as the horizontal axis and technology maturity as the vertical axis, the stages of technology germination period, development period, stagnation period, rapid growth period and mature period are marked, and the complete context and future trend of technology development are displayed for the user.

[0138] Further, the step S44 includes the following steps:

[0139] The potential technical bottleneck of the user search technology information technology development time sequence stagnation point is mined and analyzed, and the potential technical bottleneck point corresponding to the user search technology information time lag is obtained.

[0140] In the embodiments of the present application, by taking the technical field of "high energy density new energy battery materials" as an example, based on the identified "high nickel ternary material" in 2023-2024 technical development stagnation, the potential technical bottleneck of the stagnation is analyzed, from the user retrieval technology information time sequence arrangement data, the technical points, research content abstract and experimental result data of the related technology information in this period are extracted, through natural language processing technology, the semantic analysis of the technical description in a large number of papers and patents is carried out, the key word groups appearing frequently and pointing to technical difficulties are identified, such as "poor interface stability" "fast cycle life attenuation" "energy density improvement is limited" and the like, combined with the research comments of experts in the field and industry technical reports, the technical difficulties appearing more than 1.5 times of the average frequency of all technical keywords in this period and mentioned in multiple core literatures are screened out to determine the potential technical bottleneck point. For example, the problem of "poor interface stability of high nickel ternary material in high temperature environment" is discussed in 10 core papers in this period, and the frequency is higher than the average level, which is determined as the potential technical bottleneck point corresponding to the time lag, and finally a list containing multiple potential technical bottleneck points is formed.

[0141] Preferably, the potential technical bottleneck influence degree corresponding to the user retrieval technology information time lag is obtained by performing technical bottleneck influence evaluation according to the potential technical bottleneck point corresponding to the user retrieval technology information time lag;

[0142] In the embodiments of the present application, according to the previously obtained potential technical bottleneck points such as "poor interface stability of high nickel ternary material in high temperature environment", technical bottleneck influence evaluation is performed, and an evaluation system is constructed from three dimensions: technical development obstruction degree, industrial application limitation degree and market competitiveness weakening degree. For the technical development obstruction degree, the proportion of the number of research projects affected by the bottleneck to the total number of research projects is calculated, and the formula is: Suppose that 15 research projects are affected by "poor interface stability", and the total number of research projects is 20, then I tech = 0.75, the industrial application limitation degree is measured by calculating the proportion of the number of technical achievements that cannot be industrialized due to the bottleneck to the total number of related technical achievements, and the formula is: If 8 achievements cannot be industrialized due to the problem, and the total number of related achievements is 10, then I indus = 0.8, the market competitiveness weakening degree is evaluated according to the proportion of market share decline of enterprises due to the bottleneck, and the formula is: Suppose that the market share of a certain enterprise decreases from 20% to 12%, then I market = 0.4, the scores of the three dimensions are combined, and the weighted average is used to calculate the potential technical bottleneck influence degree of technology, and the formula is: I total = 0.4 x I tech + 0.3 x I indus+0.3xI market , substituting the data can be I total =0.4x0.75+0.3x0.8+0.3x0.4=0.66, that is, the influence degree of the potential technical bottleneck on the development of "high nickel ternary material" technology is 0.66, and the influence degree list corresponding to each potential technical bottleneck point is formed.

[0143] Preferably, based on the user searching for scientific and technological information technical development time lag point, the user searches for scientific and technological information time sequence sorting data corresponding to the time lag time node is searched for scientific and technological competition detection, so as to obtain the scientific and technological competition feedback data corresponding to the user searching for scientific and technological information time lag;

[0144] In the embodiment of the application, based on the technical development lag point of "high nickel ternary material" in 2023-2024, the user searches for scientific and technological information technical competition detection is carried out at the time node of 2023-2024 corresponding to the user searching for scientific and technological information time sequence sorting data, the research and development investment data, patent application quantity, new product release situation and other information of related enterprises and research institutions in the global range in this period are collected, and the competitiveness index of each competitor is calculated, and the research and development investment intensity index is: Suppose that enterprise A invests 50 million yuan in research and development in this period, and the total income is 500 million yuan, then C rd =0.1; the patent application competitiveness index is: If enterprise A applies for 20 patents, and the total number of global applications is 100, then C patent =0.2, the comprehensive competitiveness score of each competitor is obtained by comprehensively considering multiple indexes, and the formula is: C total =0.4xC rd +0.6xC patent , the difference between each competitor and its own technology development is compared and analyzed, and the competition gap coefficient is calculated, and the formula is: If the comprehensive competitiveness score of enterprise A is 0.6, and the score of competitor enterprise B is 0.8, then G gap =(0.8-0.6) / 0.6≈0.33, all the competition gap coefficients are summarized to form the scientific and technological competition feedback data corresponding to the user searching for scientific and technological information time lag, and the competition situation faced in this time lag stage is directly displayed.

[0145] Preferably, the user searches for scientific and technological competition threat coefficient corresponding to the user searching for scientific and technological competition feedback data is obtained, and the scientific and technological life cycle attenuation quantization is carried out at the time lag time node corresponding to the user searching for scientific and technological information time sequence sorting data based on the scientific and technological potential technical bottleneck influence degree corresponding to the user searching for scientific and technological information time lag and the user searching for scientific and technological competition threat coefficient, so as to obtain the scientific and technological life cycle attenuation amplitude corresponding to the user searching for scientific and technological information time lag;

[0146] In the embodiment of the present application, by obtaining the scientific and technological competition feedback data, the user retrieves the scientific and technological competition threat coefficient, selects the top three competition subjects with the largest competition gap coefficient, calculates the weighted average gap coefficient as the competition threat coefficient, and the formula is: Assuming that the gap coefficients of the three competition subjects are 0.3, 0.25, and 0.2, the weights w1=0.5, w2=0.3, and w3=0.2, then T threat =(0.5*0.3+0.3*0.25+0.2*0.2) / (0.5+0.3+0.2)=0.265, combined with the previously obtained scientific and technological potential technical bottleneck influence degree, the scientific and technological life cycle attenuation quantification is performed on the user retrieves the scientific and technological information time sequence sorting data at the time lag time node in 2023-2024, and the formula is: D decay =0.6*I total +0.4*T threat , for example, I total =0.66, T threat =0.265, D decay =0.6*0.66+0.4*0.265=0.514, that is, the scientific and technological life cycle attenuation amplitude corresponding to the time lag of "high-nickel ternary material" technology in 2023-2024 is 0.514, which quantifies the negative influence degree of the time lag on the technical life cycle.

[0147] Preferably, based on the scientific and technological life cycle attenuation amplitude corresponding to the user retrieves the scientific and technological information time lag and combined with the user retrieves the scientific and technological information technology development law data, the scientific and technological cycle evolution prediction is performed on the user retrieves the scientific and technological information time sequence arrangement data to generate the user retrieves the scientific and technological information life cycle evolution path.

[0148] In the embodiment of the present application, based on the previously obtained scientific and technological life cycle attenuation amplitude 0.514 corresponding to the time lag of "high-nickel ternary material" technology in 2023-2024, combined with the previously obtained technology development law data, the scientific and technological cycle evolution prediction is performed on the user retrieves the scientific and technological information time sequence arrangement data, on the basis of the original technology development cycle prediction, the prediction result is adjusted considering the attenuation amplitude, assuming that the original prediction is that the technology enters the rapid development period after 2-3 years after the lag point, due to the existence of the attenuation amplitude of 0.514, the starting time of the rapid development period is delayed by 0.514*12=6.168 months (converted according to 12 months per year), that is, about 6 months. At the same time, the technology maturity time is reevaluated, if the original prediction is 15 years to reach maturity, considering the attenuation, the maturity time is expected to be extended to 15+0.514 / 0.66*2≈16.56 years (adjusted according to the relationship between the technology bottleneck influence degree and the original cycle), taking time as the horizontal axis and technology maturity as the vertical axis, according to the adjusted time node and technology maturity index M tThe user retrieval technology information life cycle evolution path diagram is drawn, and stages such as a technology budding period, a development period, a stagnation period, an adjusted rapid growth period and a mature period affected by time lag are marked in the diagram, so that a life cycle evolution path more accurately reflecting the actual situation of technology development is provided for the user, and the user is assisted in judging and making decisions on the technology development trend.

[0149] Further, the application also provides a big data-based technology information intelligent retrieval consulting system for executing the big data-based technology information intelligent retrieval consulting method.

[0150] The technology knowledge network construction module is used for acquiring the structured database, the unstructured literature and the patent text corresponding to the technology information, and performing technology knowledge semantic association according to the structured database, the unstructured literature and the patent text corresponding to the technology information, so as to construct a multi-modal technology information knowledge network graph.

[0151] The user request retrieval reasoning module is used for acquiring the technology information retrieval request corresponding to the user input, and performing retrieval art analysis on the technology information retrieval request corresponding to the user input, so as to generate a technology information retrieval vector including a technical field, an innovation point and an application scenario; the technology information retrieval vector is subjected to retrieval reasoning analysis based on the multi-modal technology information knowledge network graph, so as to generate an optimal retrieval path of the user technology information.

[0152] The user retrieval consulting module is used for performing user retrieval individual consulting matching on the multi-modal technology information knowledge network graph based on the optimal retrieval path of the user technology information, so as to obtain a user technology information individualized retrieval consulting result.

[0153] The consulting technology cycle evolution module is used for performing technology time sequence analysis on the user technology information individualized retrieval consulting result, so as to generate user retrieval technology information time sequence arrangement data; and performing technology cycle evolution prediction according to the user retrieval technology information time sequence arrangement data, so as to generate a user retrieval technology information life cycle evolution path.

[0154] The above is only a specific embodiment of the application, which enables those skilled in the art to understand or implement the application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent retrieval and consultation of scientific and technological information based on big data, characterized in that, Includes the following steps: Step S1: Obtain the structured database, unstructured documents, and patent texts corresponding to the scientific and technological information, and perform semantic association of scientific and technological knowledge based on the structured database, unstructured documents, and patent texts corresponding to the scientific and technological information to construct a multimodal scientific and technological information knowledge network graph. Step S2: Obtain the science and technology information retrieval request corresponding to the user input, and perform retrieval general analysis on the science and technology information retrieval request corresponding to the user input to generate science and technology information retrieval vectors including technical fields, innovation points and application scenarios; perform retrieval reasoning analysis on the science and technology information retrieval vectors based on the multimodal science and technology information knowledge network graph to generate the optimal retrieval path for the user's science and technology information; Step S3: Based on the optimal retrieval path of user's scientific and technological information, perform personalized consultation matching on the multimodal scientific and technological information knowledge network graph to obtain personalized retrieval consultation results for user's scientific and technological information; Step S4: Perform time-series analysis on the personalized search results for user science and technology information to generate time-series data of user-searched science and technology information; Based on the time-series data of user-retrieved scientific and technological information, we can predict the evolution of the scientific and technological cycle to generate the life cycle evolution path of user-retrieved scientific and technological information.

2. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain the structured database corresponding to the scientific and technological information; Step S12: Obtain the unstructured documents corresponding to the scientific and technological information; Step S13: Obtain the patent text corresponding to the scientific and technological information; Step S14: Perform scientific and technological entity identification and entity attribute analysis on the structured database, unstructured documents and patent texts corresponding to the scientific and technological information to obtain the scientific and technological attributes of each scientific and technological information entity and the relationships between them. Each scientific and technological information entity includes technical terms, research institutions, researchers and technical indicators. Based on the scientific and technological attributes between each entity, extract the entity relationship network of each scientific and technological information entity to generate the entity relationship network between each scientific and technological entity. Step S15: Obtain the potential relationships between various scientific and technological information entities through prediction, and perform scientific and technological knowledge completion and connection on the entity relationship network between various scientific and technological entities based on the potential relationships between them to generate an initial knowledge network graph of scientific and technological information; obtain the real-time data stream corresponding to the scientific and technological information, and perform scientific and technological incremental updates on the initial knowledge network graph of scientific and technological information based on the real-time data stream corresponding to the scientific and technological information to construct a multimodal knowledge network graph of scientific and technological information.

3. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Obtain the science and technology information retrieval request corresponding to the user's input; Step S22: Construct a retrieval request semantic understanding module based on the BERT pre-trained language model, and perform deep natural language parsing on the scientific and technological information retrieval request corresponding to the user input based on the retrieval request semantic understanding module to obtain the scientific and technological natural language retrieval request corresponding to the user input. Step S23: Perform word segmentation and part-of-speech tagging on the scientific and technological natural language retrieval request corresponding to the user input, so as to generate the basic language features of scientific and technological information corresponding to the user input retrieval request; Step S24: Based on the multimodal science and technology information knowledge network graph, perform retrieval general semantic expansion on the basic language features of science and technology information corresponding to the user input retrieval request to generate science and technology information retrieval vectors including technical fields, innovation points and application scenarios. Step S25: Based on the multimodal science and technology information knowledge network graph, perform retrieval reasoning analysis on the science and technology information retrieval vector to generate the optimal retrieval path for user science and technology information.

4. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 3, characterized in that, Step S24 includes the following steps: The system performs scientific and technological term entity recognition on the basic language features of scientific and technological information corresponding to the user's input search request, so as to identify the scientific and technological, research institutions and scientific researchers term entities corresponding to the user's input search request. Based on the multimodal science and technology information knowledge network graph, the terminology entities of science and technology, research institutions and researchers corresponding to the user input retrieval request are semantically extended and mapped to concept nodes in the knowledge graph. The concept nodes include technical field tags, innovation point feature vectors and application scenario descriptions corresponding to the user input retrieval request. The search vector is generated by combining the technical field tags, innovation feature vectors, and application scenario descriptions corresponding to the user's input search request to form a scientific and technological information search vector that includes the technical field, innovation points, and application scenarios.

5. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 3, characterized in that, Step S25 includes the following steps: Step S251: Deconstruct and reconstruct the different modal scientific and technological information in the multimodal scientific and technological information knowledge network graph, so as to take the concept nodes corresponding to the different modal scientific and technological information as topological nodes, and construct the connection edges between the topological nodes according to the semantic association, logical deduction relationship and structural similarity between the scientific and technological information, so as to generate a multimodal scientific and technological semantic topological weave. Step S252: Take the science and technology information retrieval vector as the attraction source entity, and perform retrieval attraction field mapping calculation between each topological node in the multimodal science and technology semantic topology weave and the attraction source entity, so as to calculate the attraction degree between the retrieval vector and the corresponding different science and technology information nodes in the graph, and construct the retrieval vector attraction field mapping matrix. Step S253: Based on the attraction field mapping matrix of the retrieval vector, take the node with the strongest attraction as the starting point, and simulate the retrieval transmission path of the scientific and technological information retrieval vector in the multimodal scientific and technological information knowledge network graph based on the starting point. Follow the corresponding scientific and technological information logical relationship and domain knowledge context within the multimodal scientific and technological information knowledge network graph. After passing through each node, extend the transmission path continuously according to the weight of the node and the connection relationship with adjacent nodes to generate a user scientific and technological information retrieval transmission path tree. The branches and nodes of the tree reflect the potential retrieval propagation path and importance of the scientific and technological information retrieval vector in the multimodal scientific and technological information knowledge network graph. Step S254: Optimize the retrieval reasoning constraints of the user's scientific and technological information retrieval path tree to generate the optimal retrieval path for the user's scientific and technological information.

6. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 5, characterized in that, Step S254 includes the following steps: It acquires the professional logic rules and subject knowledge system corresponding to the field of science and technology, and obtains the corresponding search request constraints from the user's input of the corresponding science and technology information search request; Based on the professional logical rules corresponding to the science and technology field, as well as the subject knowledge system and retrieval request constraints, the user science and technology information retrieval transmission path tree is pruned and optimized to remove path branches that do not conform to logical rules, deviate significantly from the retrieval target, or contain redundant information. At the same time, the retained path branches are weighted and optimized and sorted. By introducing the shortest path algorithm in graph theory, the most logical and retrieval request-compliant path branch combination is found in the pruned user science and technology information retrieval transmission path tree, thereby generating a logically constrained pruned and optimized retrieval path tree. Based on the science and technology information retrieval vector, the retrieval path tree is optimized by pruning logical constraints. Each path in the retrieval path tree is assigned a dynamic weight, which is determined by a weighted calculation based on the timeliness, authority, and matching degree of the science and technology information with the science and technology information retrieval vector. The retrieval path with the highest comprehensive score is selected to clearly indicate the optimal retrieval path that starts from the science and technology information retrieval vector and obtains the most relevant and highest quality science and technology information in the multimodal science and technology information knowledge network graph, thus generating the optimal retrieval path for the user's science and technology information.

7. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Based on the optimal retrieval path of user scientific and technological information, perform user intent retrieval processing on the multimodal scientific and technological information knowledge network graph to obtain the user's optimal scientific and technological path intent retrieval results; Step S32: Obtain the user's historical search behavior and corresponding historical browsing records, and construct a user's science and technology information interest model based on the user's historical search behavior and corresponding historical browsing records; Step S33: Obtain the technology fields and research directions that the user is interested in through the user's technology information interest model, and perform personalized consultation ranking and matching on each technology information retrieval result in the user's optimal technology path intent retrieval result based on the technology fields and research directions that the user is interested in, to obtain the user's personalized technology information retrieval consultation result.

8. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Perform time series analysis on each recommended scientific and technological information in the personalized search and consultation results of user scientific and technological information to generate time-series arrangement data of user searched scientific and technological information; Step S42: Perform technological time-series development mining analysis on the time-series data of user-retrieved scientific and technological information to obtain data on the development patterns of user-retrieved scientific and technological information technology; Step S43: Based on the data on the development patterns of science and technology information technology retrieved by users, identify the technological development stagnation points in the time-series data of science and technology information retrieved by users, so as to obtain the time-series stagnation points in the development of science and technology information technology retrieved by users; Step S44: Based on the time lag points of the development of user-retrieved scientific and technological information technology, predict the evolution of the scientific and technological cycle of user-retrieved scientific and technological information time sequence data, so as to generate the life cycle evolution path of user-retrieved scientific and technological information.

9. The intelligent retrieval and consultation method for scientific and technological information based on big data according to claim 8, characterized in that, Step S44 includes the following steps: By mining and analyzing the potential technical bottlenecks at the time lag points in the development of science and technology information retrieved by users, the potential technical bottleneck points corresponding to the time lag in users' retrieval of science and technology information can be obtained. Based on the potential technical bottlenecks corresponding to the time lag in users' retrieval of scientific and technological information, an assessment of the impact of potential technical bottlenecks is conducted to obtain the degree of impact of the time lag in users' retrieval of scientific and technological information. Based on the time lag points of the development of science and technology information technology retrieved by users, the technology competition detection is performed at the corresponding time lag points on the time-series sorting data of the technology information retrieved by users, so as to obtain the technology competition feedback data corresponding to the time lag of the technology information retrieved by users. By obtaining the corresponding technological competition threat coefficient of users' search technology information based on the technological competition feedback data corresponding to the time lag of users' search technology information, and quantifying the technological life cycle decay at the corresponding time lag time node on the time-series sorting data of users' search technology information based on the degree of impact of potential technological bottlenecks corresponding to the time lag of users' search technology information and the technological competition threat coefficient of users' search technology information, the technological life cycle decay magnitude corresponding to the time lag of users' search technology information is obtained. Based on the decay of the technology life cycle corresponding to the time lag of user-retrieved technology information, and combined with data on the development patterns of user-retrieved technology information, the evolution of the technology cycle is predicted by analyzing the time-series data of user-retrieved technology information, in order to generate the evolution path of the user-retrieved technology information life cycle.

10. A big data-based intelligent retrieval and consultation system for scientific and technological information, characterized in that: For executing the big data-based intelligent retrieval and consultation method for scientific and technological information as described in claim 1, the big data-based intelligent retrieval and consultation system for scientific and technological information comprises: The technology knowledge network construction module is used to acquire structured databases, unstructured documents, and patent texts corresponding to technology information, and to perform semantic association of technology knowledge based on the structured databases, unstructured documents, and patent texts corresponding to technology information, so as to construct a multimodal technology information knowledge network graph. The user request retrieval reasoning module is used to obtain the scientific and technological information retrieval request corresponding to the user input, and to perform retrieval general analysis on the scientific and technological information retrieval request corresponding to the user input to generate scientific and technological information retrieval vectors including technical fields, innovation points and application scenarios; and to perform retrieval reasoning analysis on the scientific and technological information retrieval vectors based on the multimodal scientific and technological information knowledge network graph to generate the optimal retrieval path for the user's scientific and technological information. The user retrieval and consultation module is used to perform personalized consultation matching for users based on the optimal retrieval path of user scientific and technological information and the multimodal scientific and technological information knowledge network graph, so as to obtain personalized retrieval and consultation results for user scientific and technological information. The consultation technology cycle evolution module is used to perform technology time series analysis on the personalized technology information retrieval consultation results of users to generate time-series data of user-retrieved technology information; based on the time-series data of user-retrieved technology information, it performs technology cycle evolution prediction to generate the life cycle evolution path of user-retrieved technology information.

Images