A knowledge graph-based scientific and technological achievement supply and demand precise matching method and system

Abstract

The application belongs to the technical field of big data processing and artificial intelligence, and particularly relates to a scientific and technological achievement supply and demand accurate matching method and system based on a knowledge graph, which comprises the following steps: constructing a scientific and technological achievement dynamic knowledge graph containing a time dimension; generating a final maturity level of a to-be-matched scientific and technological achievement based on a technology maturity evaluation operator; constructing a demand-side risk preference model; extracting a plurality of potential matching paths connecting a to-be-matched scientific and technological achievement node and a demand-side enterprise node in the scientific and technological achievement dynamic knowledge graph based on a time sequence path reasoning algorithm, calculating a comprehensive score of the to-be-matched scientific and technological achievement node and the demand-side enterprise node through time axis logic filtering and multi-dimensional weighted scoring, and sorting the to-be-matched scientific and technological achievement according to the comprehensive score and outputting a final matching recommendation list with an explanation report. The application effectively improves the success rate of scientific and technological achievement transformation and the docking efficiency of the supply and demand sides.

Description

Technical Field

[0001] This application belongs to the field of big data processing and artificial intelligence technology, specifically involving a method and system for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs. Background Technology

[0002] With the deepening implementation of the strategy of driving development through scientific and technological innovation, the transformation of scientific and technological achievements has become a key link connecting laboratory research and development with industrial applications. Under the wave of digital transformation, building a multi-field-covering platform for sharing scientific and technological resources and utilizing knowledge graph technology to achieve the relational representation of scientific and technological entities has become an important means to promote the deep integration of industry, academia, and research. Through the digital integration of patents, papers, corporate information, and market trends, a vast network of scientific and technological intelligence can be built, providing underlying data support for the accurate discovery and efficient transfer of scientific and technological achievements.

[0003] Matching the supply and demand of scientific and technological achievements is a key path to achieving precise allocation of scientific and technological resources. Through semantic analysis and association learning, it establishes a mapping relationship between massive amounts of supply-side achievements and demand-side enterprises, aiming to reduce information asymmetry between the two sides. In practical applications, the accuracy of the matching directly affects the technology transfer cycle and the success rate of industrialization.

[0004] In existing technologies, knowledge graph-based matching methods mostly focus on static semantics and co-occurrence relationship analysis, failing to fully consider the lifecycle evolution of scientific and technological achievements over time. This makes it difficult for the system to identify the current R&D process and dynamic activity level of an achievement. Furthermore, traditional matching models generally lack a quantitative evaluation mechanism for technological readiness, failing to effectively distinguish between achievements in the early theoretical research stage and those in the later engineering verification stage. This results in the recommended achievements to requesting companies often not matching their actual engineering implementation capabilities. In addition, existing solutions lack in-depth modeling of the requesting companies' historical conversion records, risk tolerance preferences, and maturity acceptance ranges, making it difficult to conduct cross-dimensional reasoning and fusion scoring based on time-series paths. Consequently, the matching results struggle to achieve a good balance between technological advancement and feasibility of implementation.

[0005] Therefore, how to construct a supply-demand matching method that integrates the dynamic evolution characteristics of technology, the quantitative assessment of technology maturity, and the risk preferences of demand-side enterprises, thereby improving the accuracy and success rate of technology transfer, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the first aspect of this invention provides a method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs, comprising the following steps: S1. Construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time. S2. Define a technology maturity evaluation operator, and based on this evaluation operator, quantitatively evaluate the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generate the final maturity level of the scientific and technological achievements to be matched. The technology maturity evaluation operator identifies the types of entity nodes adjacent to the scientific and technological achievements in the dynamic knowledge graph of scientific and technological achievements, triggers a preset step-by-step score improvement mechanism, and generates the final maturity level of the scientific and technological achievements to be matched. S3. Construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched. S4. Based on the time-series path reasoning algorithm, extract multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. Calculate the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time-axis logical filtering and multi-dimensional weighted scoring. Sort the scientific and technological achievements to be matched according to the comprehensive score and output the final matching recommendation list with an explanation report.

[0008] Furthermore, step S1 specifically includes: Raw data is collected from multi-source heterogeneous databases, and the collected raw unstructured text data is preprocessed. Using a pre-defined named entity recognition model, entity nodes are identified from pre-processed text, and logical relationships between entity nodes are identified through a combination of rule and pattern matching to establish various relationship edges. For each identified entity, a unique entity node is created in the knowledge graph storage layer, and the identified relationships are instantiated as relationship edges; By analyzing the original documents, key time nodes representing the effective time of each relation edge are extracted, and based on the key time nodes, the relation edges are weighted by the initial association strength and the time decay factor. All entity nodes and relation edges carrying time attributes and weights are uniformly stored and associated to form a dynamic knowledge graph of scientific and technological achievements with a time dimension.

[0009] Furthermore, the steps for quantitatively assessing the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements based on the technology maturity assessment operator, and generating the final maturity level of the scientific and technological achievements to be matched, specifically include: S201. Calculate the maturity benchmark value of the scientific and technological achievement to be matched, specifically: S201-1. Set the initial maturity benchmark value for the nodes of scientific and technological achievements to be matched; S201-2. If it is identified that the node of the scientific and technological achievement to be matched is associated with an authorized invention patent and the patent text contains structural keywords, then the initial maturity benchmark value of the node of the scientific and technological achievement to be matched is added to the first improvement increment to obtain the first maturity benchmark value; otherwise, step S201-3 is not executed, and the initial maturity benchmark value of step S201-1 is directly used as the maturity benchmark value of the scientific and technological achievement to be matched and the process jumps to step S202. S201-3. If a production-end entity is found to be associated with the node of the scientific and technological achievement to be matched, the first maturity benchmark value of the node to be matched is added to the second improvement increment to obtain the second maturity benchmark value, and the second maturity benchmark value is used as the maturity benchmark value of the scientific and technological achievement to be matched; otherwise, the first maturity benchmark value in step S201-2 is used as the maturity benchmark value of the scientific and technological achievement to be matched, wherein the production-end entity includes the pilot plant entity, the testing report entity issued by the third-party authoritative institution, the national or industry-level network access license entity, and the industrial application contract or production order entity; S202. Calculate the association density of the scientific and technological achievements to be matched: Calculate the in-degree and out-degree of the nodes of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, normalize them respectively, and then sum them by weight to obtain the association density. S203. Calculate the completeness of the transformation path of the scientific and technological achievement to be matched: Search in the dynamic knowledge graph of scientific and technological achievements for the minimum complete connected path that starts from the original theoretical node, passes through the experimental verification node, and finally reaches the engineering prototype node. Calculate the completeness of the transformation path based on the length of the minimum complete connected path and the coverage of the necessary nodes. S204. Calculate the reputation gain coefficient of the R&D entity: Calculate the reputation gain coefficient of the R&D entity based on the historical technology transfer success rate, the cumulative number of science and technology awards obtained, and the academic status indicators in the relevant sub-technical fields of the R&D entity to which the scientific and technological achievements to be matched belong. S205. Generate initial maturity level: Add the maturity benchmark value with the association density and the completeness of the transformation path, and multiply it by the reputation gain coefficient of the R&D entity to generate the initial maturity level of the scientific and technological achievement to be matched. S206. Generate final maturity level: The initial maturity level of the scientific and technological achievement to be matched is truncated to obtain the final maturity level of the scientific and technological achievement to be matched. Specifically: if the initial maturity level of the scientific and technological achievement to be matched is >9, then the value is 9; if the initial maturity level of the scientific and technological achievement to be matched is <1, then the value is 1; otherwise, the original value is taken, and the final maturity level of the scientific and technological achievement is finally obtained.

[0010] Furthermore, the specific steps in constructing the demand-side risk preference model include: The interface extracts the historical technology introduction records, R&D investment scale, existing intellectual property database and industry attributes of the demanding enterprise from the external enterprise information platform. The industry attributes include industry background, registered capital and business scope. Based on historical technology import records, each technology project previously imported or independently developed by the demanding enterprise is traced back. Using the technology maturity assessment operator defined in step S2, the maturity level at the import or project initiation time point is obtained. Then, the arithmetic mean of the maturity scores of all valid projects at the import or project initiation time point is calculated to obtain the historical average maturity level of the demanding enterprise's projects. Specifically: if a technology project can successfully obtain its maturity level through the dynamic knowledge graph of scientific and technological achievements, it is defined as a valid project and included in the statistics; if the maturity level cannot be obtained due to data loss, the absence of the project or its related entities in the knowledge graph, etc., it is defined as an invalid project and excluded from the statistics. The risk tolerance level of the demand-side enterprise is determined based on the average maturity level of its historical projects. The basic maturity range of demand-side enterprises is determined based on their risk tolerance level. Then, the basic maturity range of demand-side enterprises is revised based on two factors: enterprise size and industry position, to obtain the expected range of technology maturity of demand-side enterprises. The technology proximity score is calculated based on the vector of the technology profile of the demand-side enterprises and the supply vector of each technology achievement to be matched.

[0011] Furthermore, step S4 specifically includes: In the dynamic knowledge graph of scientific and technological achievements, multiple potential matching paths connecting the nodes of scientific and technological achievements to be matched with the nodes of demanding enterprises are extracted. During the search process, paths with time inversion are discarded, and the path activity score of paths with long-term stagnation logic breakpoints is reduced. Based on the vector of the technology profile of the demand-side enterprises and the supply vector of each technology achievement to be matched, the similarity score of the technology field is calculated. The consistency score of time evolution trend is calculated based on the duration of the scientific and technological achievement to be matched, its current maturity level, and the historical evolution pattern of other scientific and technological achievements in the technical field to which the scientific and technological achievement belongs. The technology maturity matching score is calculated based on the final maturity level of the technology achievement to be matched and the expected range of technology maturity of the demanding enterprise. The similarity scores in the technical field, the consistency scores in the time evolution trend, the matching scores in the technology maturity, and the activity scores in the path of the scientific and technological achievements to be matched are weighted and integrated to obtain a comprehensive score. The scientific and technological achievements to be matched are ranked according to the comprehensive score, and then the final matching recommendation list with an explanation report is output.

[0012] Furthermore, the technical field similarity score is calculated as follows: the text of all valid patents is extracted from the intellectual property database of the demand-side enterprises, and a vector representing the technical profile of the demand-side enterprises is generated using a pre-trained BERT model; for the descriptive text of each scientific and technological achievement to be matched, a supply vector is generated using the same pre-trained BERT model; the initial similarity is calculated using a cosine similarity algorithm based on vector space, and then the negative value is set to zero using the ReLU function to obtain the technical field similarity score.

[0013] Furthermore, the weight coefficients in the multi-dimensional weighted fusion calculation are dynamically configured according to the search intent set by the demand-side enterprise in the interactive interface, including at least three preset weight combinations: forward-looking layout mode, instant output mode, and balanced matching mode.

[0014] Furthermore, if the demand-side enterprise is a new entrant lacking historical technology import records, and its average maturity level for historical projects is unavailable, the risk tolerance level and expected technology maturity range of this new demand-side enterprise are determined based on the risk preference model parameters of similar demand-side enterprises. Specifically, a collaborative filtering-based method is used to retrieve the most similar enterprises based on three fields: industry background, registered capital, and business scope, as recorded in the business registration. For each type of demand-side enterprise, the risk preference model parameters of these enterprises are integrated and calculated to obtain the average risk tolerance level and the average expected range of technology maturity. These parameters are then used as the risk tolerance level and expected range of technology maturity for new demand-side enterprises. The data will be updated based on actual data after the new demand-side enterprises subsequently engage in actual technology introduction activities. The risk preference model parameters include the risk tolerance level and the expected range of technology maturity.

[0015] Furthermore, it also includes a technology maturity evolution prediction mechanism: extracting samples of scientific and technological achievements with complete evolution trajectories from the constructed dynamic knowledge graph of scientific and technological achievements, and training a long short-term memory neural network model; inputting the maturity score sequence of the target scientific and technological achievement to be predicted and in the middle of the research and development stage at several time nodes into the trained long short-term memory neural network model, and estimating the expected time required for the technology to reach a higher maturity level in a rolling prediction manner.

[0016] A second aspect of this invention provides a knowledge graph-based system for precise matching of supply and demand of scientific and technological achievements, used to execute a knowledge graph-based method for precise matching of supply and demand of scientific and technological achievements, including: The knowledge graph construction module is used to construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time. The achievement evaluation module uses a technology maturity evaluation operator to quantitatively assess the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generates the final maturity level of the scientific and technological achievements to be matched. The demand modeling module is used to construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched. The matching and recommendation module, based on the time-series path reasoning algorithm, extracts multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. It calculates the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time-axis logical filtering and multi-dimensional weighted scoring. Based on the comprehensive score, it sorts the scientific and technological achievements to be matched and outputs the final matching recommendation list with an explanation report. The maturity evolution prediction module is used to train a long short-term memory neural network model based on the historical evolution trajectory of scientific and technological achievements, and to predict the expected time required for the target scientific and technological achievements to reach a higher maturity level.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: 1. By constructing a dynamic knowledge graph of scientific and technological achievements that includes a time dimension, this invention can not only achieve static association representation of technical entities, but also deeply reveal the evolution law of scientific and technological achievements throughout their entire life cycle. In addition, by introducing time attributes, the system can identify the latest developments and historical context of technology, effectively avoiding matching outdated technologies or research to demanding enterprises, and significantly improving the timeliness of matching results.

[0018] 2. This invention solves the problem of missing perception of the technology transformation stage in traditional matching models by defining a technology maturity evaluation operator. At the same time, by comprehensively identifying multiple entity nodes such as papers, patents, pilot production bases, and test reports, it can not only accurately quantify the readiness of scientific and technological achievements, but also ensure that the matched technology is consistent with the engineering capabilities of the demanding enterprise. This is of great significance for reducing the technology introduction risk of demanding enterprises and improving the success rate of scientific and technological achievements transformation.

[0019] 3. The demand-side risk preference model constructed in this invention can deeply mine the personalized needs of demand-side enterprises. By modeling historical behavioral data, the system can automatically infer the expected range of technology maturity for demand-side enterprises, thereby realizing a shift from simple keyword search to a deeper understanding of intent. This personalized matching mechanism significantly reduces the screening costs for demand-side enterprises in the face of massive amounts of information.

[0020] 4. The temporal path reasoning algorithm used in this invention achieves weighted fusion matching of four dimensions: technology similarity, time evolution trend, maturity matching, and path activity. This multi-dimensional reasoning capability makes the matching results more scientific and comprehensive, taking into account not only the similarity of the technology itself, but also the synchronicity between the pace of technology development and the needs of enterprises. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution for a knowledge graph-based method for precise matching of supply and demand of scientific and technological achievements. Figure 2 This is a schematic diagram illustrating the core principle of precise supply and demand matching based on time-series path reasoning; Figure 3 It is a logical flowchart for constructing a dynamic knowledge graph of scientific and technological achievements that includes a time dimension; Figure 4 It is a logical flowchart of the demand-side risk preference model construction and feature vector extraction; Figure 5 This is a schematic diagram illustrating the core principle of technology maturity assessment based on entity node association density and transformation path. Detailed Implementation

[0022] Example 1

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 5 The present invention will be further described in detail below with reference to specific embodiments.

[0024] Firstly, this application discloses a method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs, which is implemented according to the following steps: Step S1: Construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time.

[0025] This dynamic knowledge graph of scientific and technological achievements aims to carry multi-dimensional information about scientific and technological achievements and depict their life cycle evolution. The specific construction process is as follows.

[0026] Step S101: Data Acquisition and Preprocessing: Collect raw data from multi-source heterogeneous databases and preprocess the collected raw unstructured text data.

[0027] Distributed web crawling tools were used to collect raw data from diverse and heterogeneous databases, including patent literature databases, full-text academic journal databases, enterprise registration and qualification databases, government technology transfer disclosure databases at all levels, and third-party technology consulting report databases. These distributed crawling tools possess high concurrency processing capabilities and can dynamically adjust request frequency and header information based on the anti-crawling strategies of different websites.

[0028] The system performs preprocessing operations on the collected raw unstructured text data, including deduplication, cleaning, and format normalization. Deduplication involves identifying and removing duplicate records by calculating text hash values ​​or feature fingerprints; cleaning involves removing garbled characters, invalid null values, and interfering information unrelated to the scientific and technological achievements; format normalization involves converting all data from all sources into a preset standardized data format and storing it in a distributed file system or data lake. This step completes the aggregation and standardization of multi-source heterogeneous scientific and technological achievement information, providing a clean and uniformly formatted corpus foundation for subsequent information extraction.

[0029] Step S102, Entity Recognition and Relationship Extraction: Using natural language processing technology, a deep learning-based named entity recognition model is deeply integrated to complete entity recognition and create a corresponding entity node for each unique entity. Then, a combination of rule and pattern matching is used to identify the logical relationships between entity nodes in order to establish various relationship edges.

[0030] It is important to note that the named entity recognition model needs to be trained before implementation.

[0031] The training data is constructed as follows: text fragments such as patent abstracts, paper abstracts, and project introductions are collected from the above-mentioned multi-source databases. Domain experts then label the boundaries and types of entities according to categories such as core technical terms, R&D team names, affiliated institutions, experimental equipment, project names, funding plans, and application scenarios, forming a labeling sequence.

[0032] Specifically, the annotation pattern adopts the BIOES scheme, assigning a label to each character, where B represents the start character of an entity, I represents an internal character of an entity, E represents the end character of an entity, S represents an entity consisting of a single character, and O represents a non-entity character. The annotated data is divided into training and validation sets in an 8:2 ratio.

[0033] This named entity recognition model employs a sequence labeling architecture combining a bidirectional long short-term memory network (LSTM) and a conditional random field (CRF). The model's input layer receives character sequences from text, where each character is first mapped to a corresponding domain-pre-trained word vector with a dimension of 300. The bidirectional LSTM network layer comprises two LSTM networks, forward and backward, with each direction's hidden layer having a dimension of 256, used to capture contextual semantic dependencies from both the forward and backward directions of the sentence.

[0034] The outputs of the bidirectional Long Short-Term Memory (LSTM) network layers are concatenated into a 512-dimensional vector and fed into a fully connected layer. This vector converts the dimensions to the number of label categories, forming an emission score matrix. This emission score matrix represents the score for each label at each position given an input character.

[0035] The Conditional Random Field (CRF) layer incorporates a trainable label transition parameter matrix, the dimension of which is the product of the number of label categories, used to model the transition constraints between labels. For example, in the BIOES annotation scheme, the transition score of "B-Technology" followed by "I-Technology" should be higher than that of "O" followed by "I-Technology".

[0036] It should be noted that the label transition parameter matrix is ​​automatically learned through backpropagation during the training of the named entity recognition model. The conditional random field layer only defines the existence of the label transition parameter matrix and uses it for global path decoding.

[0037] During decoding, the Conditional Random Field layer combines the emission score matrix and the label transition parameter matrix to globally normalize the joint score of the entire label sequence, and uses the Viterbi algorithm to dynamically search for the globally optimal label sequence, thereby ensuring that the output sequence is both reasonable in terms of contextual semantics and label transition logic.

[0038] During training, the loss function is defined as the negative log-likelihood of the conditional random field output, and the calculation formula is: . in, This represents the number of samples in a training batch during the training of the named entity recognition model. Indicates the first A sequence of input characters, express The corresponding real standard label sequence; Indicates that in a given input sequence Under these conditions, the model predicts the correct label sequence. The conditional probability.

[0039] The goal of training is to minimize this loss function. The goal is to maximize the probability that the named entity recognition model outputs the correct label sequence on the training set. The model parameters are updated using a stochastic gradient descent optimization algorithm, with an initial learning rate of 0.01 that decays with each training epoch. Training stops when the named entity recognition model achieves an F1 score of 0.95 on the validation set, resulting in a usable named entity recognition model.

[0040] When applying this named entity recognition model, the preprocessed text is first finely segmented and labeled with parts of speech. Then, the text character sequence is input into the model, where the feature extraction layer extracts the semantic features of the text, ultimately outputting an entity label for each word. According to preset merging rules, the system merges consecutive word sequences with the same entity type into a single entity mention. After deduplication and standardization, a corresponding entity node is created in the graph database for each unique entity mention. For example, the tag sequence "B-TECH" followed by "I-TECH" is merged to obtain the technical term entity "deep neural network," and a corresponding entity node is created. Similarly, "B-ORG" followed by "I-ORG" is merged to obtain the organizational entity "Baidu Company," and a corresponding entity node is created. Through this process, various entity nodes can be identified, including core technical terms, R&D team names, affiliated institutions, experimental equipment, project names, funding plans, and application scenarios.

[0041] In the relationship extraction stage, a combination of rule and pattern matching methods is used to identify deep logical ties between entity nodes. For example, by analyzing the co-occurrence relationships or specific predicate structures between patent applicants, paper authors, and R&D institutions, affiliation or cooperation relationships are identified, thereby establishing the R&D entity affiliation edge; by parsing the citation and cited information between patents and the reference links between papers, technology evolution edges are established; and by extracting application scenario descriptions from results specifications or news reports, relationship edges between technology and industry scenarios are established.

[0042] Step S103: Construction and storage of knowledge graph entity nodes and relation edges: For each identified entity, the system creates a unique node identifier in the knowledge graph storage layer, i.e., the high-performance graph database, and configures an attribute set for it. The attribute set includes at least the entity's name, category, description text, and source record.

[0043] The various relationships extracted in step S102 are instantiated as relation edges in the graph database. Steps S102 and S103 together complete the process of extracting structured entities and relationships from unstructured scientific and technological texts and establishing entity nodes and relation edges in the graph database, laying the foundation for the nodes and edges of the knowledge graph.

[0044] Step S104: Assign temporal attributes and weights to relation edges: By parsing the original literature, extract a key time node representing the effective time of each relation edge, and assign weights to the relation edges based on the key time node through the initial association strength and time decay factor.

[0045] In this application, the original documents include, but are not limited to, patent specifications, full texts of papers, project acceptance reports, etc.

[0046] For patent-related relationships, the authorization announcement date is used first; if there is no authorization announcement date, the priority date or publication date is used. For paper-related relationships, the publication date is used; for technology transfer-related relationships, the contract signing date is used. This time point is mapped to the attribute fields of the relationship edge for subsequent time decay calculations, forming a technical logic network with temporal evolution characteristics.

[0047] It should be noted that the authorization announcement date is the date when the patent obtains legal protection, the technical solution is officially disclosed, and the rights are confirmed. It better reflects the actual time node when the technical achievement enters an applicable state and is more reasonable for time decay calculation. The priority date is mainly used to judge novelty and does not mean that the technology has matured and been disclosed.

[0048] In this application, when assigning weights to each relation edge, the weight of each relation edge is... From the initial correlation strength With time decay factor The decision was made jointly, and the calculation formula is as follows:

[0049] The initial association strength W0 employs different quantitative indicators depending on the type of relationship edge. For example, for citation or collaboration relationships between patents and papers, indicators such as citation frequency and collaboration depth are used; for collaboration relationships between institutions, indicators such as the number of jointly published papers, the number of jointly filed patent applications, and the number of collaborative projects are used; for associations between institutions and equipment, indicators such as equipment purchase amount, usage frequency, and the level of jointly constructed laboratories are used; and for associations between equipment and technology, the number of times the equipment is mentioned in relevant patents or papers is used. The raw values ​​of all indicators are subjected to Min-Max normalization, mapping each indicator value to the [0,1] interval. The normalized indicator values ​​are then weighted and summed to obtain the initial association strength, thus ensuring that the initial association strength is non-negative. Time decay factor. The calculation formula is: , in, It is a natural constant. The decay rate parameter is set to 0.2. This represents the time interval between the current system time and the timestamp of the associated edge, in years. This mechanism ensures that the knowledge graph prioritizes the latest research and development updates when performing subsequent matching calculations.

[0050] Step S105: Construct a dynamic knowledge graph of scientific and technological achievements: Store and associate all the entity nodes and relation edges that carry time attributes and weights in a unified manner to form a dynamic knowledge graph of scientific and technological achievements with a time dimension.

[0051] In summary, the knowledge graph constructed in step S1 not only achieves a static representation of technological entities and their associations, but also, through the embedded timeline of technological evolution and dynamic decay mechanism, gives greater weight to recent associations. This structured data foundation and time-series awareness directly provide timeline support and a computable graph environment for defining technology maturity evaluation operators and quantitatively assessing the readiness of scientific and technological achievements in subsequent step S2.

[0052] Step S2: Define a technology maturity assessment operator, and based on this assessment operator, quantitatively evaluate the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generate the final maturity level of the scientific and technological achievements to be matched. The technology maturity assessment operator identifies the types of entity nodes adjacent to the scientific and technological achievements in the dynamic knowledge graph of scientific and technological achievements, triggers a preset step-by-step score improvement mechanism, and generates the final maturity level of the scientific and technological achievements to be matched.

[0053] Based on the aforementioned dynamic knowledge graph of scientific and technological achievements that includes a time dimension, step S2 executes the operation of defining technology maturity assessment operators to establish a quantitative evaluation system based on the association density of entity nodes and their attribute characteristics, covering the entire life cycle from theoretical conception to commercial application, with multiple maturity levels preset from level 1 to level 9. The specific process is as follows.

[0054] Step S201: Calculate the maturity benchmark value of the scientific and technological achievement to be matched, including the following steps: Step S201-1: Set the initial maturity benchmark value of the technology achievement node to be matched: The system scans the adjacency relationship of the technology achievement node to be matched in the knowledge graph. If it finds that the node only has a citation or publication relationship with academic paper nodes and is not associated with any patent, pilot plant or production order entity, then the technology is determined to be in the basic research stage and its initial maturity benchmark value is set to 2.

[0055] Step S201-2: Keyword identification and score enhancement based on patent structural features: When a node of the scientific and technological achievement to be matched is identified as being associated with an authorized invention patent, the system calls a pre-trained natural language understanding engine to perform semantic analysis on the claims and specification text of the patent, and determines whether it contains structural keywords that characterize the engineering essence of the technology. If the natural language understanding engine determines that the patent text contains structural keywords, it means that the technology has entered the laboratory verification or prototype development stage. The initial maturity benchmark value of the node of the scientific and technological achievement to be matched is added to the first enhancement increment to obtain the first maturity benchmark value (the first enhancement increment is set to 2; if the initial maturity benchmark value is 2, the first maturity benchmark value is 4 at this time), and step S201-3 is executed. Otherwise, step S201-3 is not executed, and the initial maturity benchmark value of step S201-1 is directly used as the maturity benchmark value of the scientific and technological achievement to be matched and the process jumps to step S202. The structural keywords include devices, systems, structural components, specific control circuits, and industrial production processes, etc.

[0056] This natural language understanding engine is fine-tuned based on a pre-trained language model. The training data is constructed as follows: the full text of the claims and specifications of authorized invention patents are collected from the patent database as text samples. Domain experts annotate each text. If any of the above-mentioned structural keywords appear in the text and the context clearly points to a specific implementation scheme, it is labeled as a positive sample with a label of 1.

[0057] Conversely, if the entire text only provides a functional or methodological description without specific structural components or process implementation details, it is labeled as a negative sample with a label of 0. The labeled samples are then divided into training and validation sets in an 8:2 ratio.

[0058] This natural language understanding engine uses a pre-trained BERT model as the text encoder, with a fully connected classification layer added at the top. During input, the patent text is truncated or processed into 512-word segments. The BERT model then generates a hidden vector with the [CLS] marker, which has 768 dimensions. This vector is further mapped to a 2D output through the fully connected layer, and the predicted probabilities of positive and negative classes are obtained using the Softmax function. During training, the cross-entropy loss function is defined, and its calculation formula is as follows: , Where K is the number of samples in a training batch during the training of the natural language understanding engine. This represents the true label of the j-th sample, with a value of either 0 or 1. This represents the probability that the model predicts the sample to be of the positive class. The training objective is to minimize this loss function. The goal is to maximize the accuracy of label prediction on the training set. The AdamW optimizer is used, with an initial learning rate of 2e-5. During training, the F1 score on the validation set is monitored, and training stops when the F1 score reaches 0.92, resulting in a usable natural language understanding engine.

[0059] Step S201-3, Score Leap Based on Production-End Entity Association: Continue to check the graph neighborhood of the node to be matched. If the node is found to be associated with a production-end entity, it is determined that the technology has the foundation for engineering application. The first maturity benchmark value of the node to be matched is added to the second improvement increment to obtain the second maturity benchmark value, and the second maturity benchmark value is used as the maturity benchmark value of the technology to be matched (the second improvement increment is set to 2; if the first maturity benchmark value is 4, then the second maturity benchmark value is 6). Otherwise, the first maturity benchmark value of step S201-2 is used as the maturity benchmark value of the technology to be matched. The production-end entities include pilot-scale base entities, entities with testing reports issued by third-party authoritative institutions, entities with national or industry-level network access license certificates, and entities with industrial application contracts or production orders.

[0060] Furthermore, if the associated production entity shows that the technology to be matched is operating stably on a large production line, the duration field of the production order entity is parsed to calculate the number of months of stable operation. Each time a preset time period is completed, the maturity benchmark value of the technology to be matched increases by 0.5, up to a maximum of 9. For example, if the maturity benchmark value of the technology to be matched is 6.0, and the preset time period is 12 months, it increases to 6.5 after 12 months, to 7.0 after 24 months, and so on, until it reaches 9.0 after 72 months. At this point, the score has reached its upper limit, and even if the preset time period is completed again, the maturity benchmark value will not increase further and will remain at 9.0.

[0061] Step S202: Calculate the association density of the scientific and technological achievements to be matched: Calculate the in-degree and out-degree of the nodes of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements. The in-degree counts the number of relational edges pointing to the node, reflecting the frequency with which the technology is cited in subsequent patents and associated with application scenario nodes; the out-degree counts the number of relational edges originating from the node, reflecting the breadth of the technology's support for other technologies. For the in-degree value... and out-degree value After performing Min-Max normalization to the [0,1] interval, the association density of the scientific and technological achievements to be matched is obtained by weighted summation, as shown in the following expression: , in, The correlation density of the scientific and technological achievements to be matched; and These are weighting coefficients, all defaulting to 0.5; norm(·) represents the Min-Max normalization function. The upper limit for this bonus item is set to 1 to avoid excessively influencing the overall grade.

[0062] Step S203: Calculate the completeness of the transformation path of the scientific and technological achievement to be matched: Search in the dynamic knowledge graph of scientific and technological achievements for all complete connected paths starting from the original theoretical node, passing through the experimental verification node, and finally reaching the engineering prototype node. Select the shortest path as the minimum path, and calculate the completeness of the transformation path based on the length of the minimum path and the coverage of the necessary nodes. The coverage of the necessary nodes is the ratio of the number of types of necessary nodes actually covered to the total number of types of necessary nodes expected to be covered.

[0063] In this application, the path length is determined by the number of relational edges contained in the path, and the path with the fewest relational edges is the minimum path. Specifically, the shortest path algorithm built into the graph database is used to limit the required node types to consecutive combinations of "paper / theory", "patent / prototype", "pilot test / test report", and "production line / order".

[0064] If there is no complete connected path from the original theoretical node to the engineering prototype node, then the technological achievement is deemed to lack complete evidence of a transformation path, and the completeness of the transformation path is thus determined. .

[0065] If at least one complete connected path exists, the transformation path completeness is calculated based on the length of the selected minimum path and the coverage of the required nodes, as shown in the following expression: , in, To assess the completeness of the transformation path of the scientific and technological achievements to be matched, As a regulating factor, Coverage of necessary nodes This represents the length of the minimum path.

[0066] In this application, Values This design ensures that the shorter the path and the more comprehensive the coverage, the higher the score, with the value range controlled within... to between.

[0067] Step S204: Calculate the reputation gain coefficient of the R&D entity: Calculate the reputation gain coefficient of the R&D entity based on the historical technology transfer success rate, the cumulative number of science and technology awards obtained, and the academic status indicators in the relevant sub-technical fields of the R&D entity to which the scientific and technological achievement to be matched belongs.

[0068] In this application, the historical technology transfer success rate of the R&D entity to which the matched scientific and technological achievement belongs is the proportion of the number of scientific and technological achievements under that R&D entity that have passed the "successful technology transfer" standard to the total number of its scientific and technological achievements; the cumulative number of science and technology awards received by the R&D entity to which the matched scientific and technological achievement belongs is the cumulative number of provincial or ministerial-level science and technology awards received by that R&D entity; and the academic status indicator of the R&D entity to which the matched scientific and technological achievement belongs in the relevant sub-technical field is the number of articles published by that R&D entity in its core journals. and citation frequency Perform Min-Max normalization on each part, and then sum the normalization results to obtain the final result.

[0069] In this application, the criteria for determining "successful technology transfer" are as follows: A scientific and technological achievement is considered to have been successfully transferred if and only if, in the dynamic knowledge graph of scientific and technological achievements, the node of the scientific and technological achievement is directly associated with at least one of the following: an industrial application contract entity, a production order entity, a large-scale operation certificate entity, or a network access license or sales license certificate entity, and the associated timestamp of the entity is earlier than the current system time.

[0070] These indicators are weighted using preset weights to calculate the reputation gain coefficient of the R&D entity. : , in, , , The weights are 0.1, 0.05, and 0.05 respectively. A represents the historical technology transfer success rate of the R&D entity to which the matched scientific and technological achievement belongs, and A represents the cumulative number of science and technology awards received by the R&D entity to which the matched scientific and technological achievement belongs. This refers to the number of core journal publications by the research and development entity to which the scientific and technological achievements belong. The citation frequency of core journals to which the scientific and technological achievements belong is matched. This indicates Min-Max normalization.

[0071] In this application, The range of values ​​is .

[0072] Step S205: Generate the initial maturity level of the scientific and technological achievements to be matched: Based on the maturity benchmark value, correlation density, transformation path completeness, and R&D entity reputation gain coefficient of the scientific and technological achievements to be matched, the initial maturity level of the scientific and technological achievements to be matched is generated, as shown in the following expression. : , in, The initial maturity level of the scientific and technological achievement to be matched. This serves as the maturity benchmark for the scientific and technological achievements to be matched.

[0073] Step S206: Generate the final maturity level of the technology to be matched: The initial maturity level of the technology to be matched is truncated to generate the final maturity level of the technology to be matched, specifically: if... Then the value is 9, if The value is 1 if the condition is met, otherwise the original value is retained, thus obtaining the final maturity level of the technological achievement. .

[0074] The steps S201 to S206 above together define the complete calculation logic of the technology maturity evaluation operator, from setting the initial benchmark value, identifying structural features, and leaping the production-side association, to multi-dimensional adjustment of association density, path integrity and credibility, and finally producing a quantitative maturity score.

[0075] Through step S2, the system can accurately quantify the readiness level of each scientific and technological achievement in the knowledge graph, distinguishing technologies at different stages such as basic research, prototype development, engineering verification, and large-scale application. This dynamic maturity scoring mechanism directly provides quantifiable supply-side characteristic values ​​for the subsequent step S3, which involves building a demand-side risk preference model and matching the engineering capabilities of demand-side enterprises with their technological maturity.

[0076] S3. Construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched.

[0077] After quantifying the maturity of the technology supply side in step S2, step S3 shifts to in-depth modeling of the demand side, constructing a demand-side risk preference model. This model aims to transform the vague search intentions of demand-side enterprises into calculable expected ranges of technology maturity. The model construction process is as follows.

[0078] Step S301: Collection of historical behavior data of demand-side enterprises: Through interface connection with external enterprise information platforms, extract the following types of data of demand-side enterprises: historical technology introduction records, R&D investment scale, existing intellectual property database and industry attributes.

[0079] Historical technology import records include the technology description of the imported project, the contract signing date, and the subsequent industrialization status; the scale of R&D investment is measured by the average annual R&D expenditure and the proportion of R&D personnel over the past five years; the intellectual property database consists of valid patents and patent applications under the name of the requesting company; the industry attributes include industry background, registered capital, and business scope.

[0080] Step S302: Extraction of Historical Project Average Maturity Level: Retrospectively analyze each technology project previously introduced or independently developed by the client company. Based on the technology maturity evaluation operator defined in Step S2, obtain its maturity level at the time of introduction or project initiation. Then, calculate the arithmetic mean of the maturity scores of all valid projects at the time of introduction or project initiation to obtain the client company's historical project average maturity level. In this context: if a technology project can successfully obtain its maturity level through the dynamic knowledge graph of scientific and technological achievements, it is defined as a valid project and included in the statistics; if the maturity level cannot be obtained due to data missing, the lack of records of the project or its related entities in the knowledge graph, etc., it is defined as an invalid project and excluded from the statistics.

[0081] If a project is introduced in 2020, the maturity level of the technology achievement in 2020 is calculated based on the technology maturity assessment operator defined in step S2.

[0082] Step S303, Determining the Risk Tolerance Level of the Demand-Side Enterprise: Based on the average maturity level of the demand-side enterprise's historical projects. The risk tolerance level of the demand-side enterprise is determined as follows: like This indicates that the demand-side enterprises have successfully introduced and transformed low-maturity scientific and technological achievements that are in the basic research or prototype development stage multiple times, marking their risk tolerance as high level. like This indicates that the demand-side enterprises mainly purchase mature equipment or process systems that can be put into operation directly, marking their risk tolerance level as low. like Their risk tolerance was marked as medium.

[0083] Step S304: Dynamic adjustment of the expected range of technology maturity of demand-side enterprises: The basic maturity range of demand-side enterprises is determined based on their risk tolerance level. Then, the basic maturity range of demand-side enterprises is revised according to two factors: enterprise size and industry position, to obtain the expected range of technology maturity of demand-side enterprises.

[0084] In this application, the basic maturity range corresponding to the higher level is: The basic maturity range corresponding to the intermediate level is: The basic maturity range corresponding to the lower level is: .

[0085] During the correction process, the system reads the judgment result of whether the requesting company is a large leading enterprise. The judgment is based on whether the R&D center of the requesting company has more than 50 employees and whether the annual R&D budget exceeds 50 million yuan. If the company is judged to be a large leading enterprise, the system automatically adjusts the expected range: if the lower boundary is already at the minimum value of 1, the lower boundary remains unchanged; otherwise, the lower boundary is reduced by 1, while the upper boundary is expanded by 1. For example, a demand-side enterprise with a high risk tolerance level has a basic maturity range of [1,5] and an adjusted expected technology maturity range of [1,6] (the lower boundary is already 1, so it remains unchanged); a demand-side enterprise with a medium risk tolerance level has a basic maturity range of [3,7] and an adjusted expected technology maturity range of [2,8]; a demand-side enterprise with a low risk tolerance level has a basic maturity range of [5,9] and an adjusted expected technology maturity range of [4,10] (since the highest level of technology maturity is 9, its expected technology maturity range is actually truncated to 9, which is [4,9]), thus enabling the search scope to cover earlier potential disruptive technologies.

[0086] For resource-constrained small and medium-sized enterprises (SMEs), the expected technology maturity range is not expanded, and the upper limit can even be lowered by 0.5, allowing them to focus more on high-maturity achievements. Through this step, the differentiated preferences of different enterprises for technology maturity are quantified into a calculable expected maturity range. .

[0087] Step S305, semantic modeling and technology proximity calculation of the demand-side enterprise's technology reserves, specifically includes: Step S305-1: Extract the titles and abstracts of all valid patents from the intellectual property database of the demand-side enterprise. Encode each patent text using a pre-trained BERT model, and take the 768-dimensional hidden state of the [CLS] tag as the semantic vector of the patent. Then, calculate the average of all patent vectors to obtain a vector representing the technology profile of the demand-side enterprise. .

[0088] Step S305-2: For each scientific and technological achievement to be matched, a supply vector is also generated using the same pre-trained BERT model for the descriptive text. .

[0089] Step S305-3: Calculate the initial similarity using a vector space-based cosine similarity algorithm, and then use the ReLU function to set negative values ​​to zero to obtain the technical proximity score. The expression is as follows: , The numerator is the dot product of the two vectors, and the denominator is the product of the magnitudes of the two vectors. The range of values ​​is The closer the score is to 1, the closer the existing technology reserves of the demanding enterprise are to the technology to be matched, and the lower the difficulty of absorbing the technology.

[0090] It's important to note that the BERT model itself outputs a vector (usually a 768-dimensional vector labeled [CLS]). In classification tasks, this vector is then input into a fully connected layer + Softmax, ultimately outputting probabilities. Therefore, the natural language understanding engine and step S305 use the same BERT model; the only difference is the task header: one connects to a fully connected classification layer to output probabilities, while the other directly outputs a vector.

[0091] Preferably, if the number of valid patents of the demanding enterprise is 0 (e.g., a newly established enterprise or an enterprise without patents), then the enterprise's technology profile vector is replaced by the standard technology vector of its industry. This standard vector is calculated from the average value of the patent vectors of all enterprises in the industry.

[0092] Preferably, if the demanding enterprise is a new entrant lacking historical technology import records, and its average maturity level of historical projects is unavailable, the risk tolerance level and expected technology maturity range of this new entrant demanding enterprise are determined based on the risk preference model parameters of similar demanding enterprises. Specifically, a collaborative filtering-based method is used to retrieve the most similar enterprises based on three fields: industry background, registered capital, and business scope, as recorded in the business registration. For each type of demand-side enterprise, the risk preference model parameters of these enterprises are integrated and calculated to obtain the average risk tolerance level and the average expected range of technology maturity. These parameters are then used as the risk tolerance level and expected range of technology maturity for new demand-side enterprises. The data will be updated based on actual data after the new demand-side enterprises subsequently engage in actual technology introduction activities. The risk preference model parameters include the risk tolerance level and the expected range of technology maturity.

[0093] In this application, the specific calculation method for the average risk tolerance level is as follows: First, the three levels of high, medium, and low are mapped to numerical values ​​respectively. , , (The smaller the value, the higher the risk tolerance); then, for Calculate the arithmetic mean of the mapping values ​​of companies with similar demand sides. Finally, the average value is mapped back to the discrete levels based on the following thresholds: if If so, the average risk tolerance level is high; if If so, the average risk tolerance level is medium; if If so, the average risk tolerance level is low.

[0094] In this application, the specific calculation method for the expected range of average technology maturity is as follows: Let the first... The expected range for companies with similar demand is: Calculate the arithmetic mean of all lower bounds. and the arithmetic mean of all upper bounds The expected range of average technology maturity is then: If the calculated lower or upper bound exceeds the valid range of maturity level. If it is, then it will be truncated.

[0095] In this application, similarity calculation adopts a weighted fusion method after independent calculation of each field: for the industry background field, the Jaccard coefficient is used to calculate the set similarity, with a weight of 1. For the registered capital field, a numerical similarity calculation method based on the Gaussian kernel function is used, and the formula is as follows: ,in and These are the registered capital amounts of the two companies. For bandwidth parameters, the default value is the standard deviation of the registered capital. Times, weighted by For the business scope field, the cosine similarity calculation method after TF-IDF vectorization is adopted. First, the business scope text is segmented and stop word removed. Then, the TF-IDF feature vector is calculated. Finally, the text similarity is calculated using the cosine similarity formula, with weights of 1 / 2. The similarity scores of the three fields mentioned above are weighted and summed to obtain the comprehensive similarity score between the two requesting companies, which is used to find the most similar company. A number of similar demand-side companies.

[0096] In the embodiments of this application, The default setting is .

[0097] Through step S3, the system constructs a framework for each demand-side enterprise that includes its risk tolerance level and dynamic technology maturity expectation range. Technology Proximity Score The risk preference model, which includes a risk-preference model, transforms the subjective search intent of demand-side enterprises into quantifiable demand characteristics. This provides accurate demand-side constraints and scoring benchmarks for the weighted fusion matching of four dimensions—technology similarity, time evolution trend, maturity matching, and path activity—using the time-series path reasoning algorithm in step S4.

[0098] S4. Based on the time-series path reasoning algorithm, extract multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. Calculate the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time-axis logical filtering and multi-dimensional weighted scoring. Sort the scientific and technological achievements to be matched according to the comprehensive score and output the final matching recommendation list with an explanation report.

[0099] Finally, after modeling the risk preferences on the demand side in step S3, step S4 proceeds to the cross-dimensional matching stage between supply and demand. Step S4 utilizes a temporal path reasoning algorithm to perform path search and multi-dimensional weighted fusion calculation within the constructed dynamic knowledge graph of scientific and technological achievements. The specific process is as follows.

[0100] Step S401, Candidate Path Extraction and Time Axis Intelligent Filtering: In the dynamic knowledge graph of scientific and technological achievements, extract multiple potential matching paths connecting the nodes of scientific and technological achievements to be matched and the nodes of demanding enterprises. During the search process, discard paths with time inversion and reduce the path activity score of paths with long-term stagnation logic breakpoints.

[0101] To control the explosion of the number of paths, the following search constraints are set: the maximum path length does not exceed 10 steps, i.e., the number of relation edges does not exceed 10; the path must start from the supply-side technology achievement node, pass through one or more intermediate nodes, and finally reach the demand-side enterprise node; the types of intermediate nodes are limited to five categories: patents, papers, institutions, products, and industries. "Passing through in sequence" means that the nodes in the path are connected sequentially, but it is not required to include all five types of intermediate nodes; it can include only some types, and the order of appearance of different types of intermediate nodes must conform to the reasonable logic of technological evolution. For example, a typical legal path could be: technology achievement node → patent node → institution node → product node → industry node → enterprise node; another shorter legal path could be: technology achievement node → paper node → institution node → enterprise node. Examples of disallowed paths are: technology achievement node → enterprise node → patent node (because the enterprise node appears in the middle rather than the endpoint) or technology achievement node → equipment node → enterprise node (because the equipment node does not belong to the limited five types of intermediate nodes).

[0102] It should be noted that, according to the statistics of knowledge graphs in this field, the effective path length is usually between 3 and 8 steps. 10 steps can cover the effective path while controlling the complexity. Therefore, the maximum path length is set to no more than 10 steps in this application.

[0103] The temporal path reasoning algorithm incorporates an intelligent filtering mechanism during the search process, retaining paths with long-term stagnant logical breakpoints while reducing their path activity scores. For time inversion checks, the semantic type of relation edges is differentiated: only relation edge sequences with technological evolution semantics are checked for temporal monotonicity; non-technological evolution relation edges are excluded from time inversion judgment. Technological evolution relation edges include paper citation edges, patent citation edges, technological evolution edges, pilot-scale verification edges, and industrial application edges; the temporal order of these relation edges must strictly increase. Non-technological evolution relation edges include enterprise attribute edges, industry classification edges, R&D entity affiliation edges, and static association edges; the timestamps of these relation edges are not required to increase and are not included in the time inversion check. The specific checking method is as follows: along a potential matching path, the sequence of relation edges with technological evolution semantics is traversed sequentially. If the associated timestamp of a technological evolution relation edge is detected to be earlier than the associated timestamp of its preceding technological evolution relation edge, it is determined to be time inversion, and the path is discarded directly. Non-technical evolution relationship edges are not included in this judgment, and even if their timestamps are reversed, it will not affect the validity of the path.

[0104] If a technological achievement node fails to generate any new patent updates, subsequent citations, or industry-related connections within a continuous time window exceeding three years, the path is deemed to have a logical breakpoint, and a path activity score is defined. The complete path If logical breakpoints exist, then .

[0105] Step S402, Calculation of Technical Field Similarity Score: Based on the vector of the demand-side enterprise technical profile generated in step S3. And the supply vector generated from the descriptive text of each scientific and technological achievement The initial similarity is calculated using a vector space-based cosine similarity algorithm, and then the negative values ​​are set to zero using the ReLU function to obtain the technical field similarity score. The expression is as follows: , The numerator is the dot product of the two vectors, and the denominator is the product of the magnitudes of the two vectors. The range of values ​​is The closer the score is to 1, the higher the semantic fit in the technical field. This pre-trained deep learning model is the same as the BERT model used in step S305, and the training process will not be described again here.

[0106] It should be noted that the technical field similarity score Proximity score with technology They are the same metric, both based on the cosine similarity calculation of BERT vectors. The only difference in name is that they appear in different steps, which does not affect the clarity and completeness of the technical solution.

[0107] Step S403, Time Evolution Trend Consistency Calculation: Based on the duration of the scientific and technological achievement to be matched and its current maturity level, as well as the historical evolution patterns of other scientific and technological achievements in the technical field to which the scientific and technological achievement belongs, calculate the time evolution trend consistency score.

[0108] It should be noted that, since different scientific and technological achievements first appear in different entity types and their starting points are not uniform, this application defines equivalent initial maturity levels and effective R&D duration to address this issue.

[0109] In this application, the equivalent initial maturity level is determined based on the type of entity node in the dynamic knowledge graph of scientific and technological achievements where the technology to be matched first appears. The details are as follows: If the first instance of an entity is an academic paper, theoretical report, or grant proposal, then the scientific and technological achievement is determined to be in the basic research stage, with an equivalent initial maturity level. .

[0110] If the first appearance of an entity is a granted invention patent, utility model patent, laboratory prototype photograph, or prototype model, then the technological achievement is determined to have entered the laboratory verification stage, equivalent to the initial maturity level. .

[0111] If the first instance of an entity is a pilot plant report, a third-party testing report, or a small-batch trial production order, then the technological achievement is determined to be in the engineering verification stage, with an equivalent initial maturity level. .

[0112] If the first instance of an entity is an industrial application contract, a large-scale production order, or a network access license, then the technological achievement is determined to have entered the mass production and deployment stage, equivalent to the initial maturity level. .

[0113] For cases where the entity type cannot be explicitly mapped to any of the above categories for the first time, a conservative estimate is adopted, taking... .

[0114] In this application, the effective R&D duration of the scientific and technological achievements to be matched is required. The calculation formula is as follows: , , In the formula, The time period from the first time the scientific and technological achievement to be matched appeared in the dynamic knowledge graph of scientific and technological achievements to the current time point is the actual duration of existence, in years (rounded to one decimal place). To match the technological achievements, start from the equivalent maturity level. Back to maturity level Estimated time required For all scientific and technological achievements within the sub-field to which the matched scientific and technological achievements belong, categorized by maturity level... Upgrade to level The required average time span.

[0115] This linear retrospective hypothesis applies to the early stages, i.e., maturity levels. to The rate of increase in internal maturity is basically constant, and this rate is determined by the average pace of the domain.

[0116] It should be noted that maturity levels 1 to 3 represent the critical leap from "basic theory (paper)" to "laboratory prototype (patent)," and are core indicators for measuring the "early activity" of a technology. However, the evolution time from level 3 to level 9 is greatly influenced by external factors such as funding, policy, and the market, and lacks a uniform statistical pattern. Therefore, this application calculates the consistency of the time evolution trend based on the average time span required for a large number of technologies to advance from maturity level 1 to maturity level 3.

[0117] Extract samples of all scientific and technological achievements within this sub-field that have undergone a leap from Level 1 to Level 3. For each sample, calculate the time taken to traverse this interval, in years, rounded to one decimal place. Then, calculate the arithmetic mean of all samples. (Unit: year, rounded to one decimal place). This reflects the mainstream pace of technological evolution in this field from basic research (Level 1) to laboratory prototypes (Level 3).

[0118] In this application, the historical evolution patterns of other scientific and technological achievements within the technical field to which the matched scientific and technological achievement belongs are characterized using a nonlinear regression model based on an S-curve. The specific steps are as follows: The first step is to determine the statistical granularity.

[0119] The selection of sub-fields follows the following hierarchical principles: based on the classification numbers in the International Patent Classification (IPC) or the Chinese Patent Classification (CPC), the optimal statistical granularity is determined by a progressive approach, from fine to coarse. The specific steps are: obtain one or more IPC classification numbers associated with the scientific and technological achievement to be matched, and using the main classification number as the reference, locate the finest available classification granularity, usually a subgroup, preceded by the classification number. Bits or more complete bits, such as G06N3 / 0464.

[0120] If the technological achievement is not associated with the group level, then start from the finest granularity of the actual association. At the current granularity, count the number of technological achievements with complete or partial evolutionary trajectories; this sample needs to contain at least continuous... For each time point record at a different maturity level, if the sample size is greater than or equal to the preset sample size (in this application, the preset sample size is set to 30), then that granularity is used as the statistical granularity. If the sample size at the current granularity is less than the preset sample size, then the granularity is sequentially traced back one level to a coarser granularity: from subgroup to major group (the first 7 digits of the classification number, such as G06N3 / 04), and then to minor category (the first digit of the classification number). (e.g., G06N3), then to the major category (before the classification number). (e.g., G06N).

[0121] Repeat the previous step until the sample size reaches the preset sample size or the major category level. If the sample size is still less than the preset sample size after reaching the major category, further expand to the technical field to which the major category belongs, such as according to the industry classification of the IPC classification table, or directly use the evolution data of all scientific and technological achievements in the global database as a reference. At the same time, add a note to the matching interpretation report of the scientific and technological achievement, indicating that "the sample size of the sub-field to which this technology belongs is limited, and the statistical significance of the consistency score of the time evolution trend is low".

[0122] For example, if the main classification number of a scientific and technological achievement is G06N3 / 0464 (neural network convolutional structure), first count the number of samples under the G06N3 / 0464 group. If it reaches 30, then the G06N3 / 0464 group is selected. If it is insufficient, then the G06N3 / 04 group is selected. If it is still insufficient, then the G06N3 subclass is selected. If it is still insufficient, then the G06N group is selected. If the number of samples in the G06N group is still less than 30, then the scope is expanded to the technical field to which the group belongs (such as the industry classification according to the IPC classification table), or the average evolution rhythm of all scientific and technological achievements in the global database is directly used as a reference.

[0123] Through the above-mentioned exploration mechanism that progresses from fine to coarse and explores step by step, it is ensured that, with a sufficient number of samples, the mainstream evolution rhythm of the domain granularity calculation is selected as closely as possible to the technical characteristics of the scientific and technological achievements to be matched, thereby improving the accuracy and interpretability of the consistency score of the time evolution trend.

[0124] The second step is to construct an S-shaped curve evolution model.

[0125] At a defined statistical granularity, maturity level records of all scientific and technological achievements at different time points are collected. It should be noted that this fitting process aims to characterize the typical evolution of maturity levels of scientific and technological achievements in this technical field as a function of effective R&D duration, assuming average levels in dimensions such as association density, transformation path completeness, and R&D entity reputation gain. For individual scientific and technological achievements, the deviation between their actual maturity level and the S-curve predicted value (i.e., expected maturity level) will be evaluated through a relative deviation scoring mechanism in subsequent steps. For each scientific and technological achievement, the sequence of points corresponding to "effective R&D duration – maturity level" from its first appearance to the current time point is used as a sample trajectory. Specifically, in the sample trajectory, when each scientific and technological achievement first appears (i.e., when the effective R&D duration is 0), its maturity level is the equivalent initial maturity level corresponding to the entity node type where the scientific and technological achievement first appears in the dynamic knowledge graph of scientific and technological achievements; when the effective R&D duration > 0, its maturity level is the final maturity level calculated by the technology maturity evaluation operator defined in step S2.

[0126] in, The calculation formula has been given in step S403. The maturity level is the final maturity level calculated by the technology maturity evaluation operator defined in step S2.

[0127] For example, suppose that when a scientific and technological achievement first appears in the dynamic knowledge graph of scientific and technological achievements, the type of the first entity node is an academic paper. Then its equivalent initial maturity level .because It can be known that Then its effective research and development duration Therefore, the effective research and development duration of this technological achievement is... =Actual duration of existence If the maturity level of the technological achievement is level 4 (calculated using the technology maturity assessment operator) in the first year after its first appearance (when the actual duration of existence is 1 year, the effective R&D duration is 1 year); if the maturity level of the technological achievement is level 6 (calculated using the technology maturity assessment operator) in the third year after its first appearance (when the actual duration of existence is 3 years, the effective R&D duration is 3 years); and if the maturity level of the technological achievement is level 8 (calculated using the technology maturity assessment operator) in the fifth year after its first appearance (when the actual duration of existence is 5 years, the effective R&D duration is 5 years), then the sample trajectory of the technological achievement is {(0,1),(1,4),(3,6),(5,8)}.

[0128] For example, suppose that when a technological achievement first appears in the dynamic knowledge graph of technological achievements, the type of the first entity node is an invention patent. Then its equivalent initial maturity level .because It can be known that Then its effective research and development duration Therefore, the effective research and development duration of this technological achievement is... ≠ Actual duration of existence If the maturity level of the technological achievement is level 4 (calculated using the technology maturity assessment operator) in the first year after its first appearance (when the actual duration is 1 year, the effective R&D duration = 1 + 2 = 3 years); if the maturity level of the technological achievement is level 6 (calculated using the technology maturity assessment operator) in the third year after its first appearance (when the actual duration is 3 years, the effective R&D duration = 3 + 2 = 5 years); and if the maturity level of the technological achievement is level 8 (calculated using the technology maturity assessment operator) in the fifth year after its first appearance (when the actual duration is 5 years, the effective R&D duration = 5 + 2 = 7 years), then the sample trajectory of the technological achievement is {(2,3),(3,4),(5,6),(7,8)}.

[0129] Furthermore, to ensure the rationality of the evolution law of technological maturity, a non-decreasing constraint needs to be applied to the maturity sequence of each scientific and technological achievement when constructing the sample trajectory. The specific rule is as follows: if the maturity level of a scientific and technological achievement calculated at the current time point is greater than or equal to its maturity level at the previous time point, then the maturity level calculated at the current time point is directly used as the maturity level of the scientific and technological achievement at the current time point; if the maturity level of a scientific and technological achievement calculated at the current time point is less than its maturity level at the previous time point, then the maturity level of the scientific and technological achievement at the current time point follows its maturity level at the previous time point, without any decline.

[0130] It should be noted that if a technological achievement shows a situation where "its maturity level calculated at the current time point is less than its maturity level at the previous time point", it indicates that the evaluation result at that time point has an abnormal decline. The system will add a data quality prompt to the matching interpretation report, such as: "The maturity evaluation of this technological achievement at the m-th time point has data fluctuations and has been corrected according to the non-decreasing principle", to ensure the accuracy of subsequent S-curve fitting and maintain data traceability.

[0131] For example, if the sample trajectory of a certain scientific and technological achievement is {(0,1),(1,4),(3,} 3Since the maturity level (3) of its third time node is less than the maturity level (4) of its second time node, the maturity level of its third time node follows the maturity level of its second time node. At this time, the sample trajectory of the scientific and technological achievement is {(0,1),(1,4),(3,}. 4 ),(5,6)}.

[0132] By compiling all corresponding points (i.e., effective R&D duration and maturity level) from the sample trajectories of all scientific and technological achievements, a dataset reflecting the evolution of technological maturity in this field over time is formed. Based on this dataset, a Logistic growth curve is fitted, with the model form as follows: , In the formula, This indicates that the effective research and development duration is At that time, the predicted maturity level of scientific and technological achievements; For effective research and development duration, the unit is years; To approximate the upper limit, it is fixed at . ; For growth rate parameters, This indicates the location of the inflection point of the curve.

[0133] In this application, a nonlinear least squares method is used to fit all sample points in the domain to estimate the parameters. and During the fitting process, obvious outliers are removed, such as those whose maturity level has remained stagnant for an extended period. Noise data from over a year. The fitted curve represents the typical evolution of technological maturity in this field over time.

[0134] As an alternative to S-curve fitting, when the sample size is insufficient to support nonlinear fitting, a quantile-based nonparametric method is used: the effective R&D duration is grouped by percentile, the median of the technological maturity level within each group is calculated, and then interpolation is used to obtain the median expected maturity level corresponding to any R&D duration. It is used as the predicted maturity level corresponding to any R&D duration.

[0135] The third step is to calculate the consistency score of the time evolution trend of the scientific and technological achievements to be matched.

[0136] Obtain the effective R&D duration of the scientific and technological achievements to be matched. And the current maturity level (the final maturity level of the technological achievement obtained in step S2). Using the S-shaped curve or quantile median curve fitted within the field, calculate the effective R&D duration for the same period. The expected maturity level below .

[0137] When using S-curves When using the quantile method, .

[0138] Subsequently, the relative deviation score between the current maturity of the technological achievement and the typical maturity of the field (i.e., the expected maturity) is calculated, which is the consistency score of the time evolution trend. This deviation reflects the degree to which the actual maturity of an individual technological achievement deviates from the average evolution pace of the field under the combined effects of factors such as correlation density, the completeness of the transformation path, and the reputation gain of the R&D entity. To avoid absolute deviation being affected by the maturity level range... to The fixed denominator influence is mitigated by an adaptive scoring mechanism based on the actual deviation distribution within the domain, as shown in the following equation: , In the formula, Scoring for consistency of the time evolution trend; The final maturity level of the technology achievement to be matched, i.e., the current maturity level; This is the standard deviation of the deviation between the current maturity level of all samples in the domain and the expected maturity level obtained based on the S-curve. This formula makes it possible for the deviation to exceed... When the standard deviation is doubled, the score drops to The deviation is Time score .

[0139] Preferably, if the sample size in the domain is insufficient to perform calculations... Then, the fixed denominator 8 (the maximum possible deviation of maturity levels, i.e., the range from level 1 to 9) is used as the normalization factor, that is:

[0140] For example, suppose the parameters of the Logistic curve fitted to a specific sub-sector are... , For example, if a scientific and technological achievement first appears in a paper, and In that year, its equivalent initial maturity level is 1. , When the actual duration of existence When the effective R&D duration is 5 years, it is considered to be the effective R&D duration. =5 years, substituting into the S-curve, we get... .

[0141] If the current maturity level of the technology is 4.5, the standard deviation of the domain bias is... for The deviation is , If the current maturity level of the technological achievement is... The deviation is 2.2, which exceeds twice the standard deviation of the neighborhood. (i.e., 1.6), at this time =0. Therefore, this mechanism can effectively identify technological achievements that significantly deviate from the domain or the normal pace of evolution.

[0142] Step S404, Technology Maturity Matching Calculation: Based on the final maturity level of the technology to be matched and the technology maturity expectation range of the demanding enterprise, calculate the technology maturity matching score, as shown in the following expression: , In the formula, The score is the technology maturity matching score of the scientific and technological achievement to be matched; the denominator 8 is the maximum possible deviation value of the maturity level (the maturity level range is 1 to 9, and the range is 8).

[0143] The score is 1 when the final maturity level of the technology to be matched falls within the expected range; if it deviates from the range, the larger the deviation, the lower the score, and the score is zero when the deviation reaches the maximum possible value of 8.

[0144] Step S405, Multi-dimensional weighted comprehensive scoring: The similarity score between the technical fields of the scientific and technological achievements to be matched is calculated. Consistency score of time evolution trend Technology maturity matching score The overall score is obtained by weighting and combining the path activity score with the path activity score. .

[0145] Among them, the technical field similarity score, the time evolution trend consistency score, and the technology maturity matching score are all global scores for the scientific and technological achievements themselves, and are not related to specific paths; while the path activity score is calculated independently for each candidate path.

[0146] For a given technological achievement node and a demand-side enterprise node, multiple potential matching paths may exist. The maximum path activity score among all paths is taken as the path activity score for that technological achievement-enterprise pair, i.e.: , in This represents the set of all valid paths connecting the technological achievement with the enterprise. For path The activity score is used to determine the technological achievement. This design ensures that as long as a valid path with high activity exists, the technological achievement is considered to have a reliable technological connection with the enterprise.

[0147] Based on this, the overall score The calculation formula is: , Wherein, the weight coefficients satisfy and Path activity score The rule for determining the value is: if the path If the sequence is complete, without time inversion or logical breakpoints, then... If there is a stagnation period of more than 3 years but no time reversal occurs, then .

[0148] Step S406, Dynamic Configuration of Weight Coefficients: The weight coefficients for each dimension are not fixed values, but are dynamically allocated based on the search intent set by the client company in the interactive interface. The system presets three typical weight combinations: Group 1 forward-looking layout pattern: , , , Emphasizing the weighting of time-evolutionary trends to match early-stage technologies with potential disruptive potential.

[0149] Group 2 Instant Production Mode: , , , Emphasizing maturity matching weights to quickly deploy mature technologies.

[0150] Group 3 Balanced Matching Pattern: , , , The weights of each dimension are relatively balanced.

[0151] If the requesting company does not explicitly select a search intent, the system defaults to using the weight coefficients of the third group of balanced matching modes. In addition, the system also supports industry experts defining custom weights based on the analytic hierarchy process (AHP). By inviting experts to conduct pairwise comparisons of the importance of four dimensions—technology fit, consistency of time evolution trends, technology maturity matching, and path activity—a fourth-order judgment matrix is ​​constructed. The system calculates the largest eigenvalue and the corresponding eigenvector of this judgment matrix. After passing the consistency test if the consistency ratio (CR) is less than 0.1, the components of the eigenvector are normalized as follows: to The value of .

[0152] Step S407: Ranking and Interpretation Report Generation of Matching Results: All candidate scientific and technological achievements are ranked according to their comprehensive scores. Sort the results in descending order and select the top-ranked items as the matching results output. The output information includes the achievement name, research and development entity, technical indicators, final maturity level, and comprehensive score.

[0153] Simultaneously, the system generates a matching explanation report based on a knowledge graph. This report utilizes the graph's path backtracking function, following the optimal path extracted in step S401, to detail the logical alignment between the achievement and the company's needs in terms of maturity evolution, key technology nodes, and technological background. For example, the report might include text such as: "This technology is recommended because it is highly correlated with your company's existing patent library at a certain key process node, and its maturity aligns with your company's historical patent adoption preferences."

[0154] In step S4, the system organically integrates the dynamic knowledge graph of scientific and technological achievements and the results of technology maturity assessment from the supply side with the risk preference model from the demand side, completing a weighted fusion matching across four dimensions: technology similarity, time evolution trend, maturity matching, and path activity. The matching results not only provide a ranked list but also include an interpretability report based on the knowledge graph path. This matching result directly provides quantifiable reference for demand-side enterprises' technology introduction decisions. Subsequent embodiment 2 will further refine the interactive filtering function to enhance the user experience.

[0155] To ensure the scientific nature of the weight allocation, this embodiment also introduces the Analytic Hierarchy Process (AHP) as an alternative method for determining weight coefficients. Specifically, a hierarchical structure is first constructed, with the target layer being a comprehensive matching score, and the criteria layer including four evaluation dimensions: technical field fit, consistency of time evolution trends, matching degree of technological maturity, and path activity. Industry experts are invited to use Saaty's 1-9 scale to compare the importance of these four dimensions pairwise.

[0156] In the 1-9 scale, 1 indicates that both dimensions are equally important, 3 indicates that the former is slightly more important than the latter, 5 indicates that the former is significantly more important than the latter, 7 indicates that the former is strongly more important than the latter, and 9 indicates that the former is extremely more important than the latter. 2, 4, 6, and 8 are the median values ​​for adjacent judgments, and the reciprocal indicates the degree to which the latter is more important than the former. Experts need to complete the comparison of all six pairwise combinations and construct a 4th-order judgment matrix. , where matrix elements Indicates the first The dimension relative to the first The importance ratio of each dimension and The values ​​range from 1 to 4.

[0157] The system calculates the judgment matrix. Maximum eigenvalue and the corresponding eigenvectors. The calculation method involves processing the judgment matrix... The process of column vector normalization and row sum calculation is iterated repeatedly until the eigenvectors converge. After obtaining the eigenvectors, the system performs a consistency check, first calculating a consistency index. The calculation formula is: , in, To determine the order of a matrix, here Next, find the average random consistency index of the 4th order matrix. Calculate the consistency ratio :

[0158] when When the consistency check is passed, the components of the feature vector are normalized by dividing each component by the sum of all components. The transformed values ​​are then used as... to The value of . If If so, the expert is prompted to readjust the judgment matrix by performing pairwise comparisons.

[0159] In the final matching result output stage, the system assigns a comprehensive score to all candidate scientific and technological achievements. The results are sorted in descending order, and the top-ranked items are selected as the matching output. The output information includes the achievement name, research and development entity, technical indicators, final maturity level, and overall score. Simultaneously, the system generates a matching explanation report based on a knowledge graph.

[0160] The report utilizes the path backtracking function of the graph to detail the logical alignment between the achievement and the company's needs in terms of maturity evolution, key technology nodes, and technological background, following the optimal path extracted in step S401. For example, the report might include text such as: "This technology is recommended because it is highly correlated with your company's existing patent portfolio at a certain key process node, and its maturity level aligns with your company's historical import preferences."

[0161] This embodiment also includes a technology maturity evolution prediction mechanism. The system analyzes the historical evolution trajectory of similar technologies in the graph and uses a long short-term memory neural network model to predict the expected maturity time of achievements in the middle of the research and development stage.

[0162] The training data for this Long Short-Term Memory (LSTM) neural network model is constructed as follows: All scientific and technological achievement samples with complete evolutionary trajectories are extracted from the dynamic knowledge graph of scientific and technological achievements constructed in step S1. For each scientific and technological achievement with a complete evolutionary trajectory, multiple training samples are extracted using a sliding window approach: a sequence of maturity scores at consecutively predetermined time nodes of a set L equal intervals is used as input, and the maturity score at the next time node following it is used as the prediction target (e.g., L is 9). The window slides at a step size of one time node until the entire trajectory is covered.

[0163] The time interval is set to 6 months. If a technology has not yet emerged at a certain time point, the maturity score for that position is filled with 1. The total number of training samples is no less than 200, divided into training and validation sets in an 8:2 ratio.

[0164] The model structure employs a stacked architecture of two layers of Long Short-Term Memory (LSTM) networks. The input layer receives a time series of length 9, consisting of maturity scores for the target technology at nine consecutive equally spaced time points (each score is a scalar value, i.e., the input dimension is 1). For example, if the maturity scores for the past nine time points are [2,2,3,3,4,5,5,6,6], then the input sequence is these nine values. The first LSM layer contains 64 hidden units and outputs the complete sequence; the second LSM layer contains 32 hidden units and outputs only the hidden state of the last time step. This 32-dimensional hidden vector is fed into a fully connected layer, mapping to a 1-dimensional output, which is the predicted value of the maturity score for the tenth future time point. Long Short-Term Memory (LSTM) networks utilize their internal forgetting gate, input gate, and output gate mechanisms to demonstrate the non-linear evolution of learning maturity over time.

[0165] During training, the loss function is defined as mean squared error, and the calculation formula is: , in, This refers to the number of sequence samples in a training batch during the training of a long short-term memory neural network model. Indicates the first The true future maturity score of each sample The model represents the first The predicted maturity score for each sample. The training objective is to minimize this loss function. This means minimizing the deviation between the predicted and actual values. The Adam optimizer is used, with an initial learning rate of [value missing]. During training, the root mean square error on the validation set is monitored. Training is stopped when the root mean square error on the validation set no longer decreases for five consecutive training rounds, and a usable prediction model is obtained.

[0166] In the application phase, any target technological achievement in the mid-stage of R&D that has not yet reached Level 9 will have its maturity score sequence at the past 9 time points extracted and input into the model. The model outputs its expected maturity score 6 months from now. The predicted values ​​output by the model are truncated: if the predicted value is greater than 9, it is set to 9; if the predicted value is less than 1, it is set to 1; otherwise, the original value is used to ensure that the predicted maturity score always falls within the range of [1, 9]. The system further sets multiple future time steps to estimate the time required for the technology to reach a higher maturity level through rolling prediction, and integrates the prediction results into the final matching report. This provides forward-looking technology introduction decision support for demanding enterprises.

[0167] It should be noted that, when those skilled in the art apply the technology maturity evolution prediction mechanism provided by this invention, the time interval of the time nodes can be set according to actual needs, and the length of the input layer can be adjusted according to the actual training effect and domain experience. It should be noted that when adjusting the length of the input layer, the length of the input layer of this technology maturity evolution prediction mechanism needs to be adjusted adaptively.

[0168] On the other hand, the knowledge graph-based precise matching system for the supply and demand of scientific and technological achievements disclosed in the embodiments of this application includes: A knowledge graph-based system for precise matching of supply and demand of scientific and technological achievements, characterized in that a knowledge graph-based method for precise matching of supply and demand of scientific and technological achievements includes: The knowledge graph construction module is used to construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time. The achievement evaluation module uses a technology maturity evaluation operator to quantitatively assess the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generates the final maturity level of the scientific and technological achievements to be matched. The demand modeling module is used to construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched. The matching and recommendation module, based on the temporal path reasoning algorithm, extracts multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. It calculates the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time axis logical filtering and multi-dimensional weighted scoring. Based on the comprehensive score, it sorts the scientific and technological achievements to be matched and outputs the final matching recommendation list with an explanation report.

[0169] In this application, the knowledge graph-based technology achievement supply and demand precision matching system also includes a maturity evolution prediction module, which is used to train a long short-term memory neural network model based on the historical evolution trajectory of technology achievements and predict the expected time required for the target technology achievement to reach a higher maturity level.

[0170] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0171] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs, characterized in that, Includes the following steps: S1. Construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time. S2. Define a technology maturity evaluation operator, and based on this evaluation operator, quantitatively evaluate the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generate the final maturity level of the scientific and technological achievements to be matched. The technology maturity evaluation operator identifies the types of entity nodes adjacent to the scientific and technological achievements in the dynamic knowledge graph of scientific and technological achievements, triggers a preset step-by-step score improvement mechanism, and generates the final maturity level of the scientific and technological achievements to be matched. S3. Construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched. S4. Based on the time-series path reasoning algorithm, extract multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. Calculate the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time-axis logical filtering and multi-dimensional weighted scoring. Sort the scientific and technological achievements to be matched according to the comprehensive score and output the final matching recommendation list with an explanation report.

2. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 1, characterized in that, Step S1 specifically includes: Raw data is collected from multi-source heterogeneous databases, and the collected raw unstructured text data is preprocessed. Using a pre-defined named entity recognition model, entity nodes are identified from pre-processed text, and logical relationships between entity nodes are identified through a combination of rule and pattern matching to establish various relationship edges. For each identified entity, a unique entity node is created in the knowledge graph storage layer, and the identified relationships are instantiated as relationship edges; By analyzing the original documents, key time nodes representing the effective time of each relation edge are extracted, and based on the key time nodes, the relation edges are weighted by the initial association strength and the time decay factor. All entity nodes and relation edges carrying time attributes and weights are uniformly stored and associated to form a dynamic knowledge graph of scientific and technological achievements with a time dimension.

3. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 1, characterized in that, The steps for quantitatively assessing the readiness of matching scientific and technological achievements in the dynamic knowledge graph of scientific and technological achievements based on technology maturity assessment operators, and generating the final maturity level of the matching scientific and technological achievements, specifically include: S201. Calculate the maturity benchmark value of the scientific and technological achievement to be matched, specifically: S201-1. Set the initial maturity benchmark value for the nodes of scientific and technological achievements to be matched; S201-2. If it is identified that the node of the scientific and technological achievement to be matched is associated with an authorized invention patent and the patent text contains structural keywords, then the initial maturity benchmark value of the node of the scientific and technological achievement to be matched is added to the first improvement increment to obtain the first maturity benchmark value; otherwise, step S201-3 is not executed, and the initial maturity benchmark value of step S201-1 is directly used as the maturity benchmark value of the scientific and technological achievement to be matched and the process jumps to step S202. S201-3. If a production-end entity is found to be associated with the node of the scientific and technological achievement to be matched, the first maturity benchmark value of the node to be matched is added to the second improvement increment to obtain the second maturity benchmark value, and the second maturity benchmark value is used as the maturity benchmark value of the scientific and technological achievement to be matched; otherwise, the first maturity benchmark value in step S201-2 is used as the maturity benchmark value of the scientific and technological achievement to be matched, wherein the production-end entity includes the pilot plant entity, the testing report entity issued by the third-party authoritative institution, the national or industry-level network access license entity, and the industrial application contract or production order entity; S202. Calculate the association density of the scientific and technological achievements to be matched: Calculate the in-degree and out-degree of the nodes of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, normalize them respectively, and then sum them by weight to obtain the association density. S203. Calculate the completeness of the transformation path of the scientific and technological achievement to be matched: Search in the dynamic knowledge graph of scientific and technological achievements for the minimum complete connected path that starts from the original theoretical node, passes through the experimental verification node, and finally reaches the engineering prototype node. Calculate the completeness of the transformation path based on the length of the minimum complete connected path and the coverage of the necessary nodes. S204. Calculate the reputation gain coefficient of the R&D entity: Calculate the reputation gain coefficient of the R&D entity based on the historical technology transfer success rate, the cumulative number of science and technology awards obtained, and the academic status indicators in the relevant sub-technical fields of the R&D entity to which the scientific and technological achievements to be matched belong. S205. Generate initial maturity level: Add the maturity benchmark value with the association density and the completeness of the transformation path, and multiply it by the reputation gain coefficient of the R&D entity to generate the initial maturity level of the scientific and technological achievement to be matched. S206. Generate final maturity level: The initial maturity level of the scientific and technological achievement to be matched is truncated to obtain the final maturity level of the scientific and technological achievement to be matched. Specifically: if the initial maturity level of the scientific and technological achievement to be matched is >9, then the value is 9; if the initial maturity level of the scientific and technological achievement to be matched is <1, then the value is 1; otherwise, the original value is taken, and the final maturity level of the scientific and technological achievement is finally obtained.

4. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 1, characterized in that, The specific steps in constructing a demand-side risk preference model include: The interface extracts the historical technology introduction records, R&D investment scale, existing intellectual property database and industry attributes of the demanding enterprise from the external enterprise information platform. The industry attributes include industry background, registered capital and business scope. Based on historical technology import records, each technology project previously imported or independently developed by the demanding enterprise is traced back. Using the technology maturity assessment operator defined in step S2, the maturity level at the import or project initiation time point is obtained. Then, the arithmetic mean of the maturity scores of all valid projects at the import or project initiation time point is calculated to obtain the historical average maturity level of the demanding enterprise's projects. Specifically: if a technology project can successfully obtain its maturity level through the dynamic knowledge graph of scientific and technological achievements, it is defined as a valid project and included in the statistics; if the maturity level cannot be obtained due to data loss, the absence of the project or its related entities in the knowledge graph, etc., it is defined as an invalid project and excluded from the statistics. The risk tolerance level of the demand-side enterprise is determined based on the average maturity level of its historical projects. The basic maturity range of demand-side enterprises is determined based on their risk tolerance level. Then, the basic maturity range of demand-side enterprises is revised based on two factors: enterprise size and industry position, to obtain the expected range of technology maturity of demand-side enterprises. The technology proximity score is calculated based on the vector of the technology profile of the demand-side enterprises and the supply vector of each technology achievement to be matched.

5. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 1, characterized in that, Step S4 specifically includes: In the dynamic knowledge graph of scientific and technological achievements, multiple potential matching paths connecting the nodes of scientific and technological achievements to be matched with the nodes of demanding enterprises are extracted. During the search process, paths with time inversion are discarded, and the path activity score of paths with long-term stagnation logic breakpoints is reduced. Based on the vector of the technology profile of the demand-side enterprises and the supply vector of each technology achievement to be matched, the similarity score of the technology field is calculated. The consistency score of time evolution trend is calculated based on the duration of the scientific and technological achievement to be matched, its current maturity level, and the historical evolution pattern of other scientific and technological achievements in the technical field to which the scientific and technological achievement belongs. The technology maturity matching score is calculated based on the final maturity level of the technology achievement to be matched and the expected range of technology maturity of the demanding enterprise. The similarity scores in the technical field, the consistency scores in the time evolution trend, the matching scores in the technology maturity, and the activity scores in the path of the scientific and technological achievements to be matched are weighted and integrated to obtain a comprehensive score. The scientific and technological achievements to be matched are ranked according to the comprehensive score, and then the final matching recommendation list with an explanation report is output.

6. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 5, characterized in that, The technical field similarity score is calculated as follows: the text of all valid patents is extracted from the intellectual property database of the demand-side enterprises, and a vector representing the technical profile of the demand-side enterprises is generated using a pre-trained BERT model; the descriptive text of each scientific and technological achievement to be matched is used to generate a supply vector through the same pre-trained BERT model; the initial similarity is calculated using a cosine similarity algorithm based on vector space, and then the negative value is set to zero by the ReLU function to obtain the technical field similarity score.

7. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 5, characterized in that, The weighting coefficients in the multi-dimensional weighted fusion calculation are dynamically configured according to the search intent set by the demand-side enterprise in the interactive interface, including at least three preset weight combinations: forward-looking layout mode, instant output mode, and balanced matching mode.

8. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 4, characterized in that, If the demand-side enterprise is a new entrant lacking historical technology import records, and its average maturity level for historical projects is unavailable, the risk tolerance level and expected technology maturity range of this new demand-side enterprise are determined based on the risk preference model parameters of similar demand-side enterprises. Specifically, a collaborative filtering-based method is used to retrieve the most similar enterprises based on three fields: industry background, registered capital, and business scope, as recorded in the enterprise's business registration. For each type of demand-side enterprise, the risk preference model parameters of these enterprises are integrated and calculated to obtain the average risk tolerance level and the average expected range of technology maturity. These parameters are then used as the risk tolerance level and expected range of technology maturity for new demand-side enterprises. The data will be updated based on actual data after the new demand-side enterprises subsequently engage in actual technology introduction activities. The risk preference model parameters include the risk tolerance level and the expected range of technology maturity.

9. The method for precise matching of supply and demand of scientific and technological achievements based on knowledge graphs according to claim 1, characterized in that, It also includes a technology maturity evolution prediction mechanism: extracting samples of scientific and technological achievements with complete evolution trajectories from the constructed dynamic knowledge graph of scientific and technological achievements, and training a long short-term memory neural network model; inputting the maturity score sequence of the target scientific and technological achievement to be predicted and in the middle of the research and development stage at several time nodes into the trained long short-term memory neural network model, and estimating the expected time required for the technology to reach a higher maturity level in a rolling prediction manner.

10. A knowledge graph-based system for precise matching of supply and demand of scientific and technological achievements, characterized in that, A method for accurately matching the supply and demand of scientific and technological achievements based on knowledge graphs, as described in any one of claims 1-9, includes: The knowledge graph construction module is used to construct a dynamic knowledge graph of scientific and technological achievements that includes a time dimension. The entity nodes and relation edges of the dynamic knowledge graph of scientific and technological achievements are extracted from multi-source heterogeneous data. The weight of the relation edge is determined by the initial association strength and the time decay factor, so that the current association weight decreases over time. The achievement evaluation module uses a technology maturity evaluation operator to quantitatively assess the readiness of the scientific and technological achievements to be matched in the dynamic knowledge graph of scientific and technological achievements, and generates the final maturity level of the scientific and technological achievements to be matched. The demand modeling module is used to construct a demand-side risk preference model, which includes the risk tolerance level of demand-side enterprises, the expected range of technology maturity, and the technology proximity score. The technology proximity score is obtained based on the vector of the technology profile of demand-side enterprises and the supply vector of the scientific and technological achievements to be matched. The matching and recommendation module, based on the time-series path reasoning algorithm, extracts multiple potential matching paths connecting the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises in the dynamic knowledge graph of scientific and technological achievements. It calculates the comprehensive score of the nodes of the scientific and technological achievements to be matched and the nodes of the demanding enterprises through time-axis logical filtering and multi-dimensional weighted scoring. Based on the comprehensive score, it sorts the scientific and technological achievements to be matched and outputs the final matching recommendation list with an explanation report. The maturity evolution prediction module is used to train a long short-term memory neural network model based on the historical evolution trajectory of scientific and technological achievements, and to predict the expected time required for the target scientific and technological achievements to reach a higher maturity level.

Images