Enterprise scientific and technological achievement adaptation method based on big data accurate retrieval and query

Abstract

The invention discloses an enterprise scientific and technological achievement adaptation method based on big data accurate retrieval and query, and relates to the technical field of big data processing and knowledge maps, and the method comprises the steps: constructing an enterprise technical field knowledge map based on a standardized multi-modal feature matrix, receiving incremental technical data in real time by adopting an Apache Flink streaming processing framework, and carrying out the real-time retrieval and query of the enterprise technical field knowledge map. Dynamically updating a graph node relationship, removing expired nodes through a pruning algorithm, and outputting a dynamic knowledge graph with a timestamp; and receiving a query request of a user, extracting a query vector by utilizing the cross-modal embedding model, retrieving a Top-N candidate node list through a dynamic knowledge graph, calculating a comprehensive similarity score in combination with a graph relation weight, and outputting a sorted technical achievement list. Multi-source heterogeneous data is mapped to a unified semantic space through a cross-modal embedding model, and the intelligent level and the actual application effect of scientific and technological achievement adaptation are comprehensively improved in combination with construction and updating of a dynamic knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical fields of big data processing and knowledge graph, and particularly to an enterprise scientific and technological achievement adaptation method based on precise retrieval and query of big data. Background Art

[0002] With the rapid development of big data technology, the management and retrieval of enterprise scientific and technological achievements have gradually become the key links to improve the enterprise's innovation ability and technology transformation efficiency. Traditional technology retrieval methods mainly rely on keyword matching and structured database query. This method performs well when dealing with single data types, but its limitations become increasingly prominent when facing multi-source heterogeneous data.

[0003] When generating adaptation solutions in the prior art, the integration degree of multi-objective optimization algorithms and the design of constraint conditions are not reasonable enough, making it difficult to meet the requirements of multiple optimization objectives simultaneously, resulting in limited practical application effects of the adaptation solutions. There is an urgent need for a technical solution that can effectively integrate multi-modal data, realize the dynamic update of the knowledge graph, and integrate multi-objective optimization algorithms to improve the efficiency and accuracy of enterprise scientific and technological achievement retrieval and adaptation. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an enterprise scientific and technological achievement adaptation method based on precise retrieval and query of big data to solve the problem that the adaptation solution is difficult to meet the requirements of multiple optimization objectives simultaneously.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides an enterprise scientific and technological achievement adaptation method based on precise retrieval and query of big data, which includes: collecting enterprise multi-source heterogeneous data, preprocessing it, and generating a standardized multi-modal enterprise data set; Mapping the standardized multi-modal enterprise data set to a unified semantic space through a cross-modal embedding model to generate a standardized multi-modal feature matrix; Based on the standardized multi-modal feature matrix, constructing an enterprise technology field knowledge graph, using the Apache Flink streaming processing framework to receive incremental technology data in real time, dynamically updating the graph node relationships, and removing expired nodes through a pruning algorithm, and outputting a dynamic knowledge graph with a timestamp; Receiving a user query request, extracting a query vector using the cross-modal embedding model, retrieving a Top-N candidate node list through the dynamic knowledge graph, calculating a comprehensive similarity score in combination with the graph relationship weights, and outputting a sorted list of technical achievements; Using the sorted list of technical achievements, generating an adaptation solution through the integration of multi-objective constraints by a multi-objective optimization algorithm.

[0007] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: collecting multi-source heterogeneous data of enterprises, preprocessing the data, and generating a standardized multimodal enterprise data set, the specific steps are as follows: Collect R & D project data, technical parameter tables, technical drawings and design documents to generate internal enterprise data, and collect industry data, supply chain data, industry reports and market analysis data to generate external enterprise data, so as to obtain multi-source heterogeneous data of enterprises; Use the interpolation method to fill in missing values, remove duplicates from the unique identifier fields, use regular expressions to filter out invalid characters, and use statistical methods to detect and process outliers; Unify the unit and time format of structured data, convert unstructured data into a unified format, and encode categorical data to generate a standardized multimodal enterprise data set.

[0008] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: mapping the standardized multimodal enterprise data set to a unified semantic space through a cross-modal embedding model to generate a standardized multimodal feature matrix, the specific steps are as follows: Based on the historical standardized multimodal enterprise data set, construct a multimodal training data set, use the contrast loss function to minimize the embedding distance of positive sample pairs and maximize the embedding distance of negative sample pairs, and train the CLIP model to learn the semantic alignment of multimodal data; Based on the standardized multimodal enterprise data set, map it to a unified semantic space through the trained CLIP model, and dynamically adjust the weight ratio of different modalities, and combine them into a standardized multimodal feature matrix.

[0009] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: constructing an enterprise technology field knowledge graph based on the standardized multimodal feature matrix, specifically as follows: Based on the multi-source heterogeneous data of enterprises, extract technical entities, enterprise entities and external entities, and based on the explicit associations between the multi-source heterogeneous data of enterprises, define three types of relationships: technology, enterprise and external; Merge the same entity in different multi-source heterogeneous data sources of enterprises into a unique node through the entity matching algorithm, and standardize the same relationship through the relationship mapping rule; Use Neo4j to store entities and relationships as knowledge graph nodes and edges, and use the entity semantic vectors and cross-modal association information in the standardized multimodal feature matrix as node attributes and relationship weights to obtain the enterprise technology field knowledge graph.

[0010] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: the incremental technology data stream is accessed in real time through Apache Flink, entities, relationships and timestamps are parsed and extracted, out-of-order events are processed based on event time assignment and watermarks, dynamic update strategies are used to aggregate operations within a time window, and they are batch written into Neo4j through asynchronous I / O. At the same time, expired nodes in the enterprise technology domain knowledge graph are pruned based on node activity markers and business rules, and Cypher deletion operations are executed regularly, and finally a dynamic knowledge graph with timestamps is output.

[0011] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: the user query request is received, the query vector is extracted by using the cross-modal embedding model, and the Top-N candidate node list is retrieved through the dynamic knowledge graph. The specific steps are as follows. An API interface is built using the Flask framework, the JSON format of the query request is defined, the user query request is received, the user query request is parsed by a natural language processing tool, query keywords are extracted, and the parameter legality is verified. The query keywords are mapped to a unified semantic space through the trained CLIP model, and the query vector is output using a text encoder. Based on the dynamic knowledge graph, an index is built for all node vectors using FAISS, the Top-N candidate nodes are retrieved using the search method, and the semantic similarity scores between the query vector and some node vectors are calculated through the approximate nearest neighbor algorithm, and the Top-N candidate node list is output.

[0012] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: based on the Top-N candidate node list, the relationship weights of the Top-N candidate nodes are queried from the knowledge graph, the Top-N candidate nodes and relationship weights are output, the semantic similarity scores are adjusted according to the relationship weights, the comprehensive similarity scores are obtained, sorted in descending order, and the Top-K results are intercepted, and the sorted list of technological achievements is output.

[0013] As a preferred solution of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query of the present invention, wherein: the sorted list of technological achievements is used to generate an adaptation plan by integrating multi-objective constraints through a multi-objective optimization algorithm. The specific steps are as follows: Based on the sorted list of technological achievements, technological maturity constraints, commercial value constraints, implementation cost constraints, and policy compliance constraints are defined, a multi-objective constraint set is output, and the weights are dynamically adjusted through a sliding window mechanism. Based on the multi-objective constraint set and the dynamic weight vector, the improved NSGA-III algorithm is used to select the top M items from the sorted list of technological achievements as the initial population, and a reference point set is uniformly generated in the four-dimensional objective space; The simulated binary crossover and polynomial mutation are used for crossover and mutation operations to generate an optimized population and perform non-dominated sorting, construct a three-dimensional decision space, and generate a set of optimal solutions through feature contribution analysis and visualization of key influencing factors; The set of optimal solutions is optimized through an internal and external double-loop optimization mechanism to generate an adapted solution.

[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: The multi-source heterogeneous data is mapped to a unified semantic space through a cross-modal embedding model to generate a standardized multi-modal feature matrix, significantly improving the accuracy and efficiency of multi-modal data fusion; combining the construction and update of a dynamic knowledge graph realizes the real-time and precise retrieval of enterprise scientific and technological achievements; using the sorted list of technological achievements, an adapted solution is generated through a multi-objective optimization algorithm to integrate multi-objective constraints, solving the problem that traditional methods are difficult to meet the multi-dimensional requirements of technology, business, cost, and policies at the same time, providing efficient, precise, and scientific decision-making support for the management, retrieval, and transformation of enterprise technological achievements, and comprehensively improving the intelligent level and practical application effect of scientific and technological achievement adaptation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query in Embodiment 1.

[0019] Figure 2 It is a flowchart of the preprocessing of enterprise multi-source heterogeneous data in Embodiment 1.

[0020] Figure 3 It is the flowchart for training the cross-modal embedding model and generating the feature matrix in Embodiment 1.

[0021] Figure 4 It is the flowchart for constructing and updating the dynamic knowledge graph in Embodiment 1. Detailed implementation manners

[0022] To make the above objects, features and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the drawings of the specification.

[0023] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0024] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude other embodiments.

[0025] Embodiment 1, referring to Figures 1 to 4 , this embodiment provides an enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query, including the following steps: S1, collect multi-source heterogeneous data of the enterprise, and perform preprocessing to generate a standardized multi-modal enterprise data set.

[0026] Collect R & D project data, technical parameter tables, technical drawings and design documents to generate internal enterprise data, and collect industry data, supply chain data, industry reports and market analysis data to generate external enterprise data, so as to obtain multi-source heterogeneous data of the enterprise.

[0027] Use the interpolation method to fill in the missing values, remove duplicates for the unique identifier fields, filter out invalid characters using regular expressions, and use statistical methods to detect and process outliers.

[0028] It should be noted that identify the positions of the missing values in the multi-source heterogeneous data of the enterprise, select a suitable interpolation method according to the type of the multi-source heterogeneous data of the enterprise (such as linear interpolation, polynomial interpolation or time series interpolation). For the numerical multi-source heterogeneous data of the enterprise, based on the known values of the multi-source heterogeneous data of the enterprise, calculate the missing values through the interpolation formula, and fill the calculated values into the missing positions; Determine the unique identification fields (such as ID, number, etc.) in the enterprise's multi-source heterogeneous data, use deduplication algorithms (such as hash tables or sorting methods) to detect duplicate values, and according to business rules (such as retaining the latest record, retaining the first record, or merging records), retain the first record or a specific record (such as the latest record), and delete subsequent duplicate records; Identify the invalid characters that need to be filtered (such as special symbols, garbled characters, invisible characters, etc.), write regular expressions that match these invalid characters, for example, a regular expression for filtering non-alphanumeric characters. Traverse the text fields in the enterprise's multi-source heterogeneous data, and use regular expressions to match and replace or delete invalid characters; According to the distribution characteristics of the enterprise's multi-source heterogeneous data, select appropriate statistical methods (such as Z-score (standard score), IQR (interquartile range), 3σ principle, etc.) to detect outliers. For example, use the Z-score method to calculate the Z-score value of each data point, and determine outliers based on the Z-score critical value (such as Z-score > 3 or Z-score < -3). Process the outliers, and the methods include deletion, replacement with the mean / median, or correction based on business rules.

[0029] Unify the unit and time format for structured data, convert unstructured data into a unified format, and encode categorical data to generate a standardized multi-modal enterprise dataset.

[0030] It should be noted that identify the unit and time fields in the structured enterprise multi-source heterogeneous data, and clarify the format that needs to be unified (such as unifying the length unit to "meter" and the time format to "year-month-day hour:minute:second"). Use scripts or tools (such as the pandas library in Python) to traverse the enterprise's multi-source heterogeneous data, convert the unit fields (such as converting "foot" to "meter"), and format the time fields (such as adjusting "day / month / year" to "year-month-day"); For unstructured enterprise multi-source heterogeneous data (such as text, images, audio), define the target format (such as converting text to plain text encoded in UTF-8, converting images to JPEG format, and converting audio to MP3 format). Use corresponding tools (such as the Pillow library in Python to process images and the pydub library to process audio) to convert the enterprise's multi-source heterogeneous data to ensure that all unstructured enterprise multi-source heterogeneous data conforms to the unified format standard; Identify categorical fields (such as product categories, technology types) in the enterprise's multi-source heterogeneous data, and determine the encoding method (such as One-Hot encoding, label encoding). Use an encoding tool (such as the sklearn library in Python) to transform the categorical fields. For example, encode "Category A" and "Category B" in "Product Category" as [1, 0] and [0, 1] respectively. After encoding, verify the enterprise's multi-source heterogeneous data to ensure that the numerical representation of the categorical fields is correct; Integrate the processed structured enterprise multi-source heterogeneous data, unstructured enterprise multi-source heterogeneous data, and encoded categorical enterprise multi-source heterogeneous data into a unified enterprise multi-source heterogeneous dataset. Use an enterprise multi-source heterogeneous data integration tool (such as the pandas library in Python) to associate different modalities of enterprise multi-source heterogeneous data by the unique identifier field, ensure the consistency and integrity of the enterprise multi-source heterogeneous data, check the enterprise multi-source heterogeneous dataset, verify whether the standardization process of all fields is completed, and generate the final standardized multi-modal enterprise dataset.

[0031] S2. Map the standardized multi-modal enterprise dataset to a unified semantic space through a cross-modal embedding model to generate a standardized multi-modal feature matrix.

[0032] It should be noted that cross-modal embedding models include CLIP, ViLBERT, LXMERT, UNITER, etc. They aim to map data of different modalities (such as text, images, audio, etc.) to a unified semantic space. The CLIP model is used here because it has been pre-trained on a large-scale dataset of text-image pairs through contrastive learning and can efficiently learn the semantic alignment of multi-modal data.

[0033] Based on the historical standardized multi-modal enterprise dataset, construct a multi-modal training dataset, and use a contrastive loss function to minimize the embedding distance of positive sample pairs and maximize the embedding distance of negative sample pairs to train the CLIP model to learn the semantic alignment of multi-modal data.

[0034] It should be noted that the historical standardized multi-modal enterprise dataset is divided into multi-modal sample pairs (such as text-image, text-table, etc.), and each sample pair is labeled with positive and negative sample relationships (positive sample pairs are semantically related, and negative sample pairs are semantically unrelated). Use the encoder of the CLIP model to extract the embedding vectors of text and images (or other modalities) respectively, and calculate the embedding distance of positive sample pairs (such as cosine similarity) and the embedding distance of negative sample pairs. Optimize the parameters of the CLIP model through a contrastive loss function (such as InfoNCE Loss) to minimize the embedding distance of positive sample pairs and maximize the embedding distance of negative sample pairs. During the training process, use the gradient descent method to iteratively update the weights of the CLIP model, and monitor the performance of the CLIP model through the validation set.

[0035] Based on a standardized multi-modal enterprise dataset, it is mapped to a unified semantic space through a trained CLIP model, and the weight ratios of different modalities are dynamically adjusted and combined into a standardized multi-modal feature matrix.

[0036] It should be noted that based on a standardized multi-modal enterprise dataset, the trained CLIP model is used to extract the embedding vectors of modalities such as text, images, and tables respectively. The embedding vectors of all modalities are mapped to a unified semantic space. According to business requirements or data characteristics, an initial weight is assigned to each modality (e.g., text weight 0.5, image weight 0.3, table weight 0.2), and the weight ratio is optimized through a dynamic adjustment algorithm (such as a feedback mechanism based on modality importance or task performance). The embedding vectors of each modality are weighted and combined according to the adjusted weights to generate a standardized multi-modal feature matrix.

[0037] S3. Based on the standardized multi-modal feature matrix, construct an enterprise technology domain knowledge graph.

[0038] Based on enterprise multi-source heterogeneous data, extract technical entities, enterprise entities, and external entities, and define three types of relationships: technical, enterprise, and external, based on the explicit associations between enterprise multi-source heterogeneous data.

[0039] It should be noted that based on enterprise multi-source heterogeneous data, named entity recognition (NER) tools (such as SpaCy or Stanford NLP) are used to extract technical entities (such as patents, technical terms), enterprise entities (such as company names, departments), and external entities (such as industry standards, policies and regulations) from text data. Explicit associations (such as "Company A developed Technology B" or "Policy C affects Technology D") are identified from enterprise multi-source heterogeneous data through a relation extraction algorithm (such as a rule-based method or a deep learning model), and three types of relationships are defined based on these associations: technical relationships (such as technical dependence, technical citation), enterprise relationships (such as enterprise cooperation, department subordination), and external relationships (such as policy impact, industry standard citation); It should also be noted that SpaCy is a high-performance and easy-to-use open-source NLP library suitable for industrial applications; Stanford NLP is an NLP toolkit developed by Stanford University that provides high-quality models.

[0040] Through an entity matching algorithm, the same entity in different enterprise multi-source heterogeneous data sources is merged into a unique node, and the same relationship is standardized through a relationship mapping rule.

[0041] It should be noted that an entity matching algorithm (such as a rule-based method, a machine learning model, or a pre-trained entity linking tool) is used to match entities in multi-source heterogeneous data sources of different enterprises. By comparing the attributes of entities (such as names, descriptions, IDs) and context information (such as associated entities, events), it is determined whether they are the same entity, and the matched entities are merged into a unique node (such as merging "Company A" and "A Corp" into "Company A"). Relationship mapping rules are defined (such as unifying "development" and "research and development" as "development"), all relationship data is traversed, and the same relationship is standardized according to the relationship mapping rules (such as mapping "research and development" to "development"), and the relationship names in the relationship data are updated.

[0042] Use Neo4j to store entities and relationships as nodes and edges in a knowledge graph, and use the entity semantic vectors and cross-modal association information in the standardized multi-modal feature matrix as node attributes and relationship weights to obtain the enterprise technology domain knowledge graph.

[0043] It should be noted that data structures for nodes and edges are created in Neo4j. The extracted entities are used as nodes (such as "Company A", "Technology B"), and the defined relationships are used as edges (such as "development"). Cypher statements (such as CREATE (create) and MERGE9 (merge)) are used to batch import entities and relationships into Neo4j to ensure the uniqueness of nodes and edges. The entity semantic vectors in the standardized multi-modal feature matrix are used as node attributes, and the cross-modal association information is used as relationship weights. The integrity and consistency of the knowledge graph are verified to ensure that nodes, edges, and their attributes are correctly stored, and the final knowledge graph is output. It should also be noted that verifying the integrity and consistency of the knowledge graph specifically means: checking whether the number of nodes and edges meets expectations, ensuring that all entities and relationships have been correctly imported, verifying the integrity of node attributes (such as checking whether attributes exist) and the correctness of relationship weights (such as checking whether the weight values are reasonable) through Cypher queries, and using graph traversal algorithms (such as breadth-first search or depth-first search) to check the connectivity of the knowledge graph to ensure that there are no isolated nodes or broken relationships.

[0044] S4. The Apache Flink streaming processing framework is used to receive incremental technology data in real time, dynamically update the graph node relationships, and remove expired nodes through a pruning algorithm, and output a dynamic knowledge graph with timestamps.

[0045] Real-time access to incremental technology data streams through Apache Flink, parse and extract entities, relationships, and timestamps, and handle out-of-order events based on event time assignment and watermarks. Adopt a dynamic update strategy to aggregate operations within time windows, and batch write to Neo4j through asynchronous I / O. At the same time, prune expired nodes in the enterprise technology domain knowledge graph based on node activity markers and business rules, and periodically execute Cypher deletion operations to finally output a dynamic knowledge graph with timestamps.

[0046] The details are as follows: Real-time subscribe to incremental data streams (such as new patents, technology parameter updates, policy changes, etc.) through the Kafka connector of Apache Flink, configure Kafka topics and consumer groups, and output incremental technology data streams; Use the DataStream API of Flink (Data Stream API) to parse the incremental technology data stream one by one, verify the integrity of fields (such as patent ID and technical terms cannot be empty), extract key fields (entity ID, event type, timestamp, association relationship), filter out invalid data with missing necessary fields or incorrect formats, and output a structured incremental data stream; Assign timestamps based on the event time field in the structured incremental data stream, generate watermarks, allow fixed delays (such as 5 minutes) to handle out-of-order events, and ensure the correct processing of the structured incremental data stream within the specified time window by configuring watermark strategies (such as BoundedOutOfOrderness), and output an incremental data stream with timestamps and watermarks; BoundedOutOfOrderness (bounded out-of-order) is a watermark generation strategy that allows data to arrive out of order within a specified time range; Dynamically generate Cypher operation instructions according to the event types in the incremental data stream with timestamps and watermarks: if the incremental data stream contains new entities (such as new patent IDs), generate CREATE operations to create nodes and initialize attributes; if the incremental data stream contains attribute changes (such as technical term extensions), generate SET operations to update node attributes; if the incremental data stream contains new associations (such as patents citing new standards), generate MERGE operations to create or update relationships and record association weights, and output a Cypher operation instruction stream (a set of add, delete, and modify operations); Aggregate multiple update operations of the same entity in the Cypher operation instruction stream based on a sliding window (such as a 10-minute window with a 5-minute sliding step), group and merge similar operations by entity ID (such as multiple attribute updates merged into a single SET operation), reduce the write frequency, and output an aggregated batch Cypher operation instruction stream; Invoke the Neo4j driver through Flink's asynchronous I / O interface, execute Cypher operations in batches, optimize the batch writing performance using the UNWIND statement, configure the asynchronous request timeout and maximum concurrency, and use Neo4j's batch transaction interface to submit operations, outputting the nodes and relationships of the enterprise technology domain knowledge graph with real-time updates (including the latest attributes and timestamps). Add a last_updated field to each node in the nodes and relationships of the enterprise technology domain knowledge graph with real-time updates (including the latest attributes and timestamps) to record the last update timestamp, define business expiration rules (e.g., a technology node is marked as expired if it has not been updated for more than 5 years), automatically write the current timestamp when the entity is updated, periodically scan the nodes and add an expiration label (such as Expired), and output the set of nodes marked as expired. Expired is a label or status indicating that a certain node or data has become invalid or is no longer in use. Based on a time decay function (such as , where is the node activity score and is the time difference, representing the interval between the current time and the last activity time of the node), calculate the node activity scores in the set of nodes marked as expired. Nodes with scores lower than the elimination threshold are considered expired. At the same time, directly delete specific types of nodes (such as abolished industry standards, invalid policies), execute Cypher statements to match the expired nodes, and batch delete the nodes and associated relationships, outputting the enterprise technology domain knowledge graph after removing the expired nodes. Trigger the pruning task regularly (such as at 0:00 every day) through Flink's ProcessFunction, avoid executing during the peak period of the online service, register a timer to call Neo4j's Cypher interface to execute the deletion operation, record the pruning log, and output the dynamic knowledge graph with timestamps. ProcessFunction is a low-level API provided for implementing custom stream processing logic.

[0047] S5. Receive the user query request, use the cross-modal embedding model to extract the query vector, and retrieve the Top-N candidate node list through the dynamic knowledge graph.

[0048] Build an API interface using the Flask framework, define the JSON format of the query request, receive the user query request, parse the user query request through natural language processing tools, extract the query keywords, and verify the parameter legitimacy.

[0049] It should be noted that the Flask framework is used to create an API interface, define the JSON format of the query request, and set up routes to receive user query requests. The user query requests are parsed by natural language processing tools (such as spaCy or NLTK) to extract query keywords (such as "Company A", "technology"), and the legality of the parameters is verified (such as checking whether the query field is empty and whether the limit is an integer within a reasonable range). query is a field used to pass the query content input by the user (such as search keywords, filtering conditions, etc.). limit is a field used to specify the maximum number of returned results.

[0050] The query keywords are mapped to a unified semantic space through the trained CLIP model, and the query vector is output using a text encoder.

[0051] It should be noted that the extracted query keywords (such as "technologies of Company A") are input into the text encoder of the trained CLIP model (such as the text encoder of CLIP), and the query keywords are converted into text embedding vectors. The query keywords are tokenized and normalized through the preprocessing module of the CLIP model (such as converting to lowercase and removing stop words) to ensure that the input format meets the requirements of the CLIP model. The forward propagation method of the text encoder is called to calculate the semantic vector of the query keywords, and the query vector (such as an array of floating-point numbers with a fixed length) is output.

[0052] Based on the dynamic knowledge graph, FAISS is used to build an index for all node vectors, the search method is used to retrieve the Top-N candidate nodes, and the semantic similarity scores between the query vector and some node vectors are calculated through the approximate nearest neighbor algorithm, and the Top-N candidate node list is output.

[0053] It should be noted that the semantic vectors of all nodes in the dynamic knowledge graph (such as the embedding vectors generated from the CLIP model) are loaded into FAISS, and vector indexes are built using index structures such as IndexFlatL2 or IndexIVFFlat. The search method of FAISS is called, and the query vector is used as the input, and the retrieval parameters are set (such as indicating retrieving the Top-N nodes), the semantic similarity scores between the query vector and the node vectors are calculated through the approximate nearest neighbor algorithm (ANN) (such as Euclidean distance or cosine similarity), the candidate nodes are sorted in descending order (from high to low) according to the similarity scores, and the Top-N candidate node list (such as a JSON format containing node IDs and similarity scores) is output; The expression for calculating the semantic similarity score between the query vector and the node vector is: ; where, The cosine value range between the query vector and part of the node vectors is [-1, 1]. is the query vector, is the node vector, is the query vector and the node vector is the included angle between them; FAISS is an efficient similarity search library developed by Facebook AI Research, specifically used for quickly finding the most similar vectors in large-scale vector data; search is a method for finding the vector most similar to the query vector in the constructed vector index; IndexFlatL2 (flat index based on L2 distance) is an index structure in FAISS that calculates the similarity between vectors using the L2 distance (Euclidean distance); IndexIVFFlat (flat index based on inverted index) is an index structure in FAISS that combines the inverted file index and the flat index to accelerate the search for large-scale vector data.

[0054] S6, combine the graph relationship weights to calculate the comprehensive similarity score, and output the sorted list of technical achievements.

[0055] Based on the Top-N candidate node list, query the relationship weights of the Top-N candidate nodes from the knowledge graph, output the Top-N candidate nodes and their relationship weights, adjust the semantic similarity score according to the relationship weights to obtain the comprehensive similarity score, sort it in descending order, intercept the Top-K results, and output the sorted list of technical achievements.

[0056] It should be noted that based on the Top-N candidate node list, traverse each candidate node, use Cypher query to retrieve the relationship weights of each candidate node from the knowledge graph, combine the retrieved relationship weights with the semantic similarity score, adjust the semantic similarity score according to the relationship weights to obtain the comprehensive similarity score, and the expression is: ; where, is the comprehensive similarity score, is the semantic similarity score between the query vector and part of the node vectors, is the relationship weight coefficient, is the relationship weight of the Top-N candidate nodes; Store the comprehensive similarity scores of candidate nodes in a list, and use a sorting algorithm (such as Python's sorted() function) to sort them in descending order of scores. According to the set value, intercept the first nodes, and format the information of the Top-K nodes (such as node ID, comprehensive score, technical description) into a list of technological achievements.

[0057] S7. Utilize the sorted list of technological achievements and generate an adaptation plan by integrating multi-objective constraints through a multi-objective optimization algorithm.

[0058] Define technological maturity constraints, commercial value constraints, implementation cost constraints, and policy compliance constraints based on the sorted list of technological achievements, output a set of multi-objective constraints, and dynamically adjust the weights through a sliding window mechanism.

[0059] It should be noted that extract the maturity-related indicators of each technology (such as technology development stage, number of patents, degree of technology verification) from the sorted list of technological achievements, and quantify these indicators using a scoring standard (such as 1-10 points). For example: the score for a technology in the experimental stage is 2, 5 for the pilot stage, and 8 for the commercialization stage. Store the maturity score of each technology as a technological maturity constraint; Extract the commercial value indicators of each technology (such as expected revenue, market share, return on investment) from the sorted list of technological achievements, and calculate these indicators using a quantification method (such as amount or percentage). For example: the score for an expected revenue of less than 1 million yuan is 1, 3 for 1 million - 5 million yuan, and 5 for more than 5 million yuan. Store the commercial value score of each technology as a commercial value constraint; Extract the implementation cost indicators of each technology (such as R & D budget, equipment investment, labor cost) from the sorted list of technological achievements, and calculate these indicators using a quantification method (such as amount or cost score). For example: the score for an implementation cost of less than 500,000 yuan is 5, 3 for 500,000 - 1 million yuan, and 1 for more than 1 million yuan. Store the implementation cost score of each technology as an implementation cost constraint; Extract the policy compliance indicators of each technology (such as regulatory compliance, industry standards, environmental protection requirements) from the sorted list of technological achievements, and evaluate these indicators using a quantification method (such as compliance score or boolean value). For example: the score for full compliance is 5, 3 for partial compliance, and 1 for non-compliance. Store the policy compliance score of each technology as a policy compliance constraint; Integrate the technology maturity constraint, commercial value constraint, implementation cost constraint, and policy compliance constraint into a multi-objective constraint set (such as in JSON format) to ensure that the constraint values of each technological achievement are complete and can be used by subsequent optimization algorithms. Dynamically adjust the weights through a sliding window mechanism (such as a fixed-time window or an event-triggered window): within the window, calculate the weight changes of each constraint (such as an increase in commercial value weight and a decrease in implementation cost weight) based on real-time data (such as market feedback, technology updates), update the weight vector, and output the adjusted dynamic weight vector and multi-objective constraint set.

[0060] Based on the multi-objective constraint set and the dynamic weight vector, use an improved NSGA-III algorithm to select the top M items from the sorted list of technological achievements as the initial population, and uniformly generate a reference point set in the 4D objective space.

[0061] It should be noted that based on the multi-objective constraint set and the dynamic weight vector, first select the top items as the initial population (such as ), ensure that the technology maturity, commercial value, implementation cost, and policy compliance constraint values of each technological achievement are clear. In the four-dimensional objective space (technology maturity, commercial value, implementation cost, policy compliance), use a uniform distribution method (such as the Das-Dennis method) to generate a reference point set to ensure that the reference points are uniformly distributed in the objective space. Map the individuals of the initial population to the objective space and calculate the distance between each individual and the reference points; Through the non-dominated sorting and reference point association mechanism of the improved NSGA-III algorithm, perform non-dominated sorting on the individuals in the population, classify the individuals according to the Pareto front rank, ensure that individuals with higher front ranks are preferentially selected, standardize and reduce the dimension of the individual characteristics (such as technical indicators), map the individual characteristics to coordinate points in the objective space (such as [0.7, 0.3]), calculate the Euclidean distance between this coordinate point in the objective space and all predefined reference points, and assign this individual to the reference point with the closest distance (such as the distance to reference point B is 0.28, which is closer than 0.36 to reference point A, so choose B). Select individuals according to the crowding degree of the reference points (such as the number of associated individuals), preferentially select the individuals associated with the reference points with lower crowding degrees as the optimal individuals to ensure the diversity of the population, add the selected optimal individuals to the next generation population, and select the optimal individuals from them to enter the next generation population; It should also be noted that the improved NSGA-III algorithm, based on the standard NSGA-III, significantly improves the diversity, convergence, and computational efficiency of the algorithm in high-dimensional multi-objective optimization problems by optimizing reference point generation (such as using a dynamic reference point generation method to adaptively adjust the reference point position according to the population distribution), crossover and mutation strategies (such as combining simulated binary crossover SBX and polynomial mutation PM, or designing customized operators), constraint handling mechanisms (such as introducing an adaptive penalty function to dynamically adjust the constraint priority), population update strategies (such as adding an elite retention mechanism and local search techniques to accelerate convergence), and parallel computing (such as using GPU acceleration or a distributed computing framework to improve the running efficiency). The feasibility and practicality of the algorithm are proven through theoretical analysis and experimental verification.

[0062] Perform crossover and mutation operations using simulated binary crossover and polynomial mutation to generate an optimized population and perform non-dominated sorting, construct a three-dimensional decision space, and generate a set of preferred solutions through feature contribution analysis and visualization of key influencing factors.

[0063] It should be noted that perform simulated binary crossover (SBX) operations on the individuals in the initial population, randomly select parent individuals, generate offspring individuals according to the crossover probability and distribution index, perform polynomial mutation (PM) operations on the offspring individuals, randomly adjust the individual gene values according to the mutation probability and distribution index to generate an optimized population, perform non-dominated sorting on the optimized population, classify the individuals according to the Pareto front rank, and visualize them in a three-dimensional decision space (such as technology maturity, commercial value, implementation cost), extract the characteristic values (such as technology maturity, commercial value, implementation cost) of each technological achievement, calculate the contribution of the characteristic values of each technological achievement to the comprehensive score (such as through weights or scoring ratios), use visualization tools (such as Matplotlib, Seaborn) to draw a feature contribution diagram, screen out the technological achievements with significant feature contributions and meeting the constraint conditions, sort them according to the comprehensive score, and generate a set of preferred solutions; Matplotlib is a plotting library in Python for creating various static, dynamic, and interactive charts; Seaborn is an advanced Python data visualization library based on Matplotlib, focusing on the drawing of statistical charts.

[0064] Optimize the set of preferred solutions through an internal and external double-loop optimization mechanism to generate adapted solutions.

[0065] It should be noted that in the outer loop, based on the set of preferred solutions, a global optimization goal (such as maximizing business value or minimizing implementation cost) is set, and the solution parameters (such as technology maturity and implementation budget) are adjusted through a global search algorithm (such as genetic algorithm or particle swarm optimization) to generate a preliminary optimized solution. Then, through the inner loop, local optimization is performed on each preliminary optimized solution, and a local search algorithm (such as gradient descent or simulated annealing) is used to fine-tune the technology implementation decision variables (such as technology implementation details and resource allocation), and the feasibility of the solution under the constraint conditions (such as policy compliance and technical feasibility) is verified. The results of the inner and outer loop optimizations are integrated to generate an adapted solution; It should also be noted that the inner and outer double-loop optimization mechanism: Inner loop mechanism: Definition: The inner loop mechanism is a high-frequency and real-time optimization process that continuously receives dynamic data streams (such as market changes and technology status updates) and quickly adjusts model parameters or local variables. Its core is to achieve rapid iteration through incremental learning or online learning. For example, the business value prediction model is updated every 30 minutes.

[0066] Application fields: In the prior art, the inner loop mechanism is widely applied in real-time recommendation systems (such as dynamic pricing in e-commerce platforms), industrial Internet of Things (such as device status monitoring and predictive maintenance), and financial high-frequency trading (such as real-time prediction of stock prices); Outer loop mechanism: Definition: The outer loop mechanism is a low-frequency and global optimization process that focuses on long-term strategy adjustment and system-level parameter optimization. For example, the multi-objective weight allocation strategy is periodically updated through reinforcement learning (such as the PPO algorithm), usually in batch processing on a daily / weekly basis; Application fields: The outer loop mechanism has mature applications in fields such as automated control systems (such as robot path planning), resource scheduling (such as cloud computing task allocation), and long-term risk modeling (such as insurance actuarial); The combination method of the inner and outer double-loop optimization mechanism of the present invention is as follows: Collaboration between data and strategy: Inner loop: Real-time processing of the timeliness decay of technological achievements; Outer loop: Based on the real-time data accumulated by the inner loop, through streaming feedback (user adoption data) to drive reinforcement learning (PPO algorithm) to dynamically adjust the multi-objective weights; Technological fusion and innovation: Combining the streaming computing framework (Apache Flink) of the inner loop with the offline reinforcement learning engine of the outer loop, and realizing double-cycle data synchronization through a shared state storage (such as Redis or HBase); The "fast decision-making" of the inner loop focuses on high-frequency data updates, while the "slow strategy" of the outer loop focuses on low-frequency strategy optimization, forming a complementarity. The two work together through sharing state information and feedback mechanisms to solve the problem that it is difficult to balance real-time performance and globality in a single loop.

[0067] This embodiment also provides a computer device applicable to the situation of the enterprise scientific and technological achievement adaptation method based on big data accurate retrieval and query, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the enterprise scientific and technological achievement adaptation method based on big data accurate retrieval and query as proposed in the above embodiment.

[0068] This computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0069] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the enterprise scientific and technological achievement adaptation method based on big data accurate retrieval and query as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read Only Memory (EPROM for short), Programmable Red-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0070] In summary, the present invention: maps multi-source heterogeneous data to a unified semantic space through a cross-modal embedding model, generates a standardized multi-modal feature matrix, and significantly improves the accuracy and efficiency of multi-modal data fusion; realizes real-time and accurate retrieval of enterprise scientific and technological achievements by combining the construction and update of a dynamic knowledge graph; uses the sorted list of technological achievements, and integrates multi-objective constraints through a multi-objective optimization algorithm to generate an adaptation plan, solving the problem that traditional methods are difficult to meet the multi-dimensional requirements of technology, business, cost, and policies simultaneously, providing efficient, accurate, and scientific decision-making support for the management, retrieval, and transformation of enterprise technological achievements, and comprehensively improving the intelligent level and practical application effect of scientific and technological achievement adaptation.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query, characterized in that: Including: Collect multi-source heterogeneous data of enterprises, preprocess it, and generate a standardized multi-modal enterprise dataset; Map the standardized multi-modal enterprise dataset to a unified semantic space through a cross-modal embedding model to generate a standardized multi-modal feature matrix; Based on the standardized multi-modal feature matrix, construct a knowledge graph of the enterprise's technical field. Use the Apache Flink streaming processing framework to receive incremental technical data in real time, dynamically update the graph node relationships, and remove expired nodes through a pruning algorithm, and output a dynamic knowledge graph with timestamps; Receive a user query request, use the cross-modal embedding model to extract a query vector, retrieve a list of Top-N candidate nodes through the dynamic knowledge graph, calculate a comprehensive similarity score by combining the graph relationship weights, and output a sorted list of technical achievements; Use the sorted list of technical achievements to generate an adaptation plan by integrating multi-objective constraints through a multi-objective optimization algorithm.

2. The enterprise science and technology achievement adaptation method based on big data precise retrieval query according to claim 1, characterized in that: The steps of collecting multi-source heterogeneous data of enterprises, preprocessing it, and generating a standardized multi-modal enterprise dataset are as follows: Collect R & D project data, technical parameter tables, technical drawings and design documents to generate internal enterprise data, and collect industry data, supply chain data, industry reports and market analysis data to generate external enterprise data, and obtain multi-source heterogeneous data of enterprises; Use the interpolation method to fill in missing values, deduplicate the unique identifier fields, filter invalid characters using regular expressions, and use statistical methods to detect and process outliers; Unify the unit and time format of structured data, convert unstructured data to a unified format, and encode categorical data to generate a standardized multi-modal enterprise dataset.

3. The enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query according to claim 1, characterized in that: The steps of mapping the standardized multi-modal enterprise dataset to a unified semantic space through a cross-modal embedding model to generate a standardized multi-modal feature matrix are as follows: Based on the historical standardized multi-modal enterprise dataset, construct a multi-modal training dataset, use the contrast loss function to minimize the embedding distance of positive sample pairs and maximize the embedding distance of negative sample pairs, and train the CLIP model to learn the semantic alignment of multi-modal data; Based on the standardized multi-modal enterprise dataset, map it to a unified semantic space through the trained CLIP model, and dynamically adjust the weight ratio of different modalities to combine into a standardized multi-modal feature matrix.

4. The enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query according to claim 1, characterized in that: The construction of the knowledge graph of the enterprise's technical field based on the standardized multi-modal feature matrix is as follows: Based on the multi-source heterogeneous data of the enterprise, extract technical entities, enterprise entities, and external entities, and define three types of relationships: technology, enterprise, and external based on the explicit associations between the multi-source heterogeneous data of the enterprise; Merge the same entity in different multi-source heterogeneous data sources of the enterprise into a unique node through an entity matching algorithm, and standardize the same relationship through a relationship mapping rule; Use Neo4j to store entities and relationships as knowledge graph nodes and edges, and use the entity semantic vectors and cross-modal association information in the standardized multi-modal feature matrix as node attributes and relationship weights to obtain the knowledge graph of the enterprise's technical field.

5. The enterprise science and technology achievement adaptation method based on big data precise retrieval query according to claim 1, wherein: Real-time access to incremental technology data streams through Apache Flink, parse and extract entities, relationships, and timestamps, and process out-of-order events based on event time allocation and watermarks. Adopt a dynamic update strategy to aggregate operations within a time window, and batch write to Neo4j through asynchronous I / O. At the same time, prune expired nodes in the enterprise technology domain knowledge graph based on node activity markers and business rules, and periodically execute Cypher deletion operations to finally output a dynamic knowledge graph with timestamps.

6. The enterprise science and technology achievement adaptation method based on big data precise retrieval and query according to claim 5, wherein: Receiving a user query request, using the cross-modal embedding model to extract a query vector, and retrieving a Top-N candidate node list through a dynamic knowledge graph. The specific steps are as follows: Use the Flask framework to build an API interface, define the JSON format of the query request, receive the user query request, parse the user query request through natural language processing tools, extract query keywords, and verify the legality of the parameters; Map the query keywords to a unified semantic space through the trained CLIP model, and use the text encoder to output a query vector; Based on the dynamic knowledge graph, use FAISS to build an index for all node vectors, adopt the search method to retrieve Top-N candidate nodes, and calculate the semantic similarity scores between the query vector and some node vectors through the approximate nearest neighbor algorithm, and output the Top-N candidate node list.

7. The enterprise scientific and technological achievement adaptation method based on big data precise retrieval and query according to claim 6, characterized in that: Based on the Top-N candidate node list, query the relationship weights of the Top-N candidate nodes from the dynamic knowledge graph, output the Top-N candidate nodes and relationship weights, adjust the semantic similarity scores according to the relationship weights to obtain the comprehensive similarity scores, sort them in descending order, intercept the Top-K results, and output the sorted list of technological achievements.

8. The enterprise science and technology achievement adaptation method based on big data precise retrieval query according to claim 1, wherein: Using the sorted list of technological achievements, generate an adaptation plan through a multi-objective optimization algorithm to integrate multi-objective constraints. The specific steps are as follows: Define technological maturity constraints, commercial value constraints, implementation cost constraints, and policy compliance constraints based on the sorted list of technological achievements, output a multi-objective constraint set, and dynamically adjust the weights through a sliding window mechanism; Based on the multi-objective constraint set and the dynamic weight vector, use the improved NSGA-III algorithm to select the top M items from the sorted list of technological achievements as the initial population, and uniformly generate a reference point set in the four-dimensional objective space; Adopt simulated binary crossover and polynomial mutation for crossover and mutation operations, generate an optimized population and perform non-dominated sorting, construct a three-dimensional decision space, and generate a set of preferred solutions through feature contribution analysis and visualization of key influencing factors; Optimize the set of preferred solutions through an internal and external double-loop optimization mechanism to generate an adaptation plan.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the enterprise scientific and technological achievement adaptation method based on big data accurate retrieval query according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the enterprise scientific and technological achievement adaptation method based on big data accurate retrieval query according to any one of claims 1 to 8.

Images