Method for quickly constructing knowledge graph based on multi-source heterogeneous data

Abstract

This invention relates to the field of knowledge graph technology, and more particularly to a method for rapidly constructing knowledge graphs based on multi-source heterogeneous data. The method includes: acquiring real-time uploaded data and generating an anchor node set; acquiring several anchor-based basic knowledge graphs from a highway database based on the anchor node set, and generating the confidence score of each anchor-based basic knowledge graph; generating a central graph vector based on the confidence score deviation rate; constructing an associated knowledge graph; simulating the pre-trained knowledge graph or associated knowledge graph according to event types in the highway database to generate event simulation probabilities, and generating event accuracy coefficients by combining the event probabilities and event simulation probabilities; classifying the pre-trained knowledge graph or associated knowledge graph based on the event accuracy coefficients; and adjusting uploaded tags or updating the highway database based on the classification results. This invention effectively improves the accuracy of rapidly constructing knowledge graphs based on multi-source heterogeneous data.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a method for rapidly constructing knowledge graphs based on multi-source heterogeneous data. Background Technology

[0002] The highway system generates massive amounts of heterogeneous data from multiple sources. This data contains crucial information about highway network operation, facility status, and event evolution, serving as a core resource for intelligent decision-making. Knowledge graphs, as a semantic knowledge organization and representation framework, can transform scattered data into triples of entities, attributes, and relationships with clear connections, enabling intelligent analysis and decision-making for highways. However, existing technologies primarily focus on graph construction for single-type data, lacking a unified semantic mapping and association mechanism. This results in semantic silos between data from different sources, poor knowledge reusability across different domains, an imperfect knowledge quality assessment mechanism, and weak reasoning capabilities.

[0003] Chinese Patent Application Publication No. CN118885624A discloses a temporal knowledge graph completion method based on temporal counterfactual enhancement. This invention discloses a method for temporal knowledge graph completion based on temporal counterfactual enhancement, comprising: acquiring a temporal knowledge graph, defining parameters, and generating a factual relation dataset; constructing a time processing tensor mapping the temporal graph structure using a time-varying k-kernel decomposition algorithm and a community membership function; performing counterfactual processing, combining individual processing effect estimation, and constructing a counterfactual relation dataset; merging the counterfactual relation dataset and the factual relation dataset; training a temporal knowledge graph completion model using a composite loss function; inputting the temporal knowledge graph with missing relations into the trained temporal knowledge graph completion model, and outputting the completed temporal knowledge graph. This method, by introducing a temporal counterfactual enhancement strategy and constructing and utilizing counterfactual data, significantly improves the model's robustness and predictive ability under conditions of incomplete and sparse data, achieving more accurate and nuanced temporal knowledge graph completion.

[0004] It is evident that the existing technology has the following problems: the acquired knowledge data has not been effectively classified and evaluated, and the system has not been adjusted according to the professional relevance of the knowledge. Summary of the Invention

[0005] To address this, the present invention provides a method for rapidly constructing knowledge graphs based on multi-source heterogeneous data, thereby solving the problems in existing technologies where the acquired knowledge data is not effectively classified and evaluated, and the construction system is not adjusted according to the professional relevance of the knowledge.

[0006] To achieve the above objectives, this invention provides a method for rapid construction of knowledge graphs based on multi-source heterogeneous data, comprising:

[0007] Extract keywords and upload tags from the uploaded data respectively. Based on the upload tags, call several preset nodes. Combine the basic text model and keywords to generate several initial nodes. Select nodes that are the same as the preset nodes and initial nodes to generate an anchor node set.

[0008] Based on the anchor node set, several anchor basic knowledge graphs are obtained from the highway database. The relationship mapping of each anchor basic knowledge graph is obtained and the confidence weight of the relationship mapping is determined by combining the uploaded tags to generate the confidence of each anchor basic knowledge graph.

[0009] Based on the confidence bias rate, a pre-trained knowledge graph is generated or a multi-granularity analysis is performed on the anchor node set to determine several central nodes and several key links. Based on the several central nodes, several central basic knowledge graphs are obtained to generate a central graph vector.

[0010] Generate association graph vectors corresponding to each key link, and construct an association knowledge graph based on the similarity between the central graph vector and the association graph vectors;

[0011] Based on the uploaded data, the event type and event probability are matched, and the event type is combined with the event type to simulate the operation of the pre-trained knowledge graph or the associated knowledge graph to generate the event simulation probability. The event accuracy coefficient is then generated by combining the event probability.

[0012] Based on the event accuracy coefficient, the pre-trained knowledge graph or the associated knowledge graph is categorized to adjust the uploaded tags or update the highway database.

[0013] Furthermore, the process of selecting nodes that are the same as the preset nodes and the initial nodes to generate the anchor node set includes,

[0014] The upload tag corresponding to the terminal device that acquires the uploaded data is then invoked, and the preset node corresponding to the upload tag is called.

[0015] Extract keywords from the uploaded data to match the corresponding basic words in the basic text model, and call the initial node corresponding to the basic words;

[0016] The preset node and the initial node are compared, and the same nodes are selected to generate the anchor node set.

[0017] Furthermore, the process of generating the confidence scores for each of the anchored basic knowledge graphs includes,

[0018] The anchoring node set is input into the highway database to generate several anchoring basic knowledge graphs related to the anchoring node set;

[0019] Obtain the set of mapping knowledge proportions in different professional fields for the relationship mapping of each anchored basic knowledge graph;

[0020] The similarity between the mapped knowledge proportion set and the tag knowledge proportion set of the uploaded tag is determined as the tag relevance of the relation mapping;

[0021] The confidence weight and confidence level of each relation mapping are determined based on the association degree of several labels.

[0022] The weighted sum of the confidence scores of each of the aforementioned relation mappings is determined to be the confidence score of the corresponding anchored basic knowledge graph.

[0023] Furthermore, the standard deviation of the confidence scores of several anchored basic knowledge graphs is calculated as the confidence bias rate.

[0024] Furthermore, in response to a confidence bias rate less than or equal to a preset confidence bias threshold, several anchored basic knowledge graphs are constructed into a pre-trained knowledge graph;

[0025] In response to a confidence deviation rate exceeding a preset confidence deviation threshold, multi-granularity analysis is performed on the anchor node set.

[0026] Furthermore, the process of generating the central graph vector includes,

[0027] Multi-granularity analysis is performed on the anchor node set to obtain the centrality value and semantic importance value of each anchor node;

[0028] The weighted sum of the centrality value and the semantic importance value is determined as the central node representation value, and the anchor node is determined as a central node based on the central node threshold.

[0029] Based on several central nodes, several central basic knowledge graphs are obtained from the highway database. These central basic knowledge graphs are then transformed into several central subgraph vectors using a graph neural network and aggregated to generate a central graph vector.

[0030] Furthermore, the process of constructing a knowledge graph of connections includes,

[0031] Based on each anchor node, several links are obtained from the highway database, and the path betweenness and node correlation of each link are calculated.

[0032] The weighted sum of the path betweenness and the node correlation is determined as the critical link characterization value, and the critical link threshold is used to determine whether the link is a critical link.

[0033] Based on several key links, several key link knowledge graphs are obtained from the highway database. The key link knowledge graphs are then transformed into several link subgraph vectors and aggregated using a graph neural network to generate key link vectors.

[0034] Calculate the similarity between the central graph vector and the associated graph vector, and aggregate the central basic knowledge graph corresponding to the central graph vector with a similarity greater than a preset similarity threshold and the key link knowledge graph corresponding to the associated graph vector to form an associated knowledge graph.

[0035] Furthermore, the process of generating an event accuracy coefficient by combining the event probability and the event simulation probability includes,

[0036] Based on the traffic flow and timestamp of the uploaded data, obtain the corresponding event types and probabilities from the highway database;

[0037] The event type is input into the pre-trained knowledge graph or the associated knowledge graph to generate event simulation probabilities through simulation.

[0038] The probability deviation between the simulated probability of the event and the actual event probability is determined as the event accuracy coefficient.

[0039] Furthermore, in response to an event accuracy coefficient being greater than or equal to a preset event accuracy threshold, the pre-trained knowledge graph or associated knowledge graph corresponding to the event accuracy coefficient is determined to be a strongly associated graph.

[0040] If the event accuracy coefficient is less than a preset event accuracy threshold, the pre-trained knowledge graph or associated knowledge graph corresponding to the event accuracy coefficient is determined to be a weakly associated graph.

[0041] Furthermore, the process of adjusting the uploaded tags includes,

[0042] In response to the fact that the knowledge graph corresponding to the accuracy coefficient of the event is a weakly related graph, the weakly related node set and the preset node set in the weakly related graph are obtained and fused to update the uploaded tag;

[0043] The pre-trained knowledge graph corresponding to the event accuracy coefficient is a weakly associated graph. The initial node set and the preset node set are fused to update the uploaded tags.

[0044] The preset node set is a collection of preset nodes called by the upload tag, and the initial node set is a collection of initial nodes generated by combining the basic text model and keywords.

[0045] Compared with existing technologies, the beneficial effects of this invention are as follows: By extracting keywords from the acquired real-time uploaded data and combining these keywords with a basic text model of the highway professional field, this invention determines the entity nodes in the uploaded data that can be used to construct a knowledge graph in the professional field. Simultaneously, it extracts upload tags from the terminal devices corresponding to the uploaded data, sets upload tags to differentiate the professional field capabilities of the terminal devices, and calls entity nodes in the technical fields that the terminal devices are familiar with and proficient in based on the upload tags. The entity nodes under the two analysis conditions are aggregated to obtain the anchor node set in the uploaded data. This allows for preliminary screening of uploaded data while improving the accuracy of entity node identification, breaking through semantic silos, achieving rapid preliminary screening of multi-source heterogeneous data, improving graph construction efficiency, enhancing the domain adaptability of terminal device data, optimizing the professional attributes of the knowledge graph, and improving the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data.

[0046] Furthermore, this invention inputs the pre-processed anchor node set into a pre-constructed highway database to generate several anchor basic knowledge graphs based on each anchor node. It analyzes the proportion of mapping knowledge in different professional fields for each anchor basic knowledge graph, further analyzes the similarity between the mapping knowledge proportion set and the tag knowledge proportion set of uploaded tags, and determines this as the tag correlation degree of the relationship mapping. It then assigns weights to each relationship mapping based on the tag correlation degree, thereby calculating the confidence level of the anchor basic knowledge graph corresponding to the relationship mapping. Finally, it analyzes the confidence deviation rate of the anchor basic knowledge graph. For anchored basic knowledge graphs with low deviation, pre-trained knowledge graphs are constructed. For anchored basic knowledge graphs with high deviation, multi-granularity analysis is performed to construct a quantitative confidence evaluation system to ensure the reliability of the graph knowledge and accurately quantify the matching degree between relation mapping and terminal data domain attributes. For anchored basic knowledge graphs with low deviation, pre-trained knowledge graphs are directly constructed to achieve rapid reuse of high-quality knowledge. For graphs with high deviation, multi-granularity analysis is carried out to locate the root cause of deviation and accurately correct it layer by layer from entity granularity, relation granularity, and attribute granularity, further improving the accuracy of rapid construction of knowledge graphs based on multi-source heterogeneous data.

[0047] Furthermore, in response to the large deviation of the anchored basic knowledge graph, the present invention performs multi-granularity analysis on the anchored node set to obtain the centrality value and semantic importance value corresponding to each anchored node. The weighted sum of the centrality value and the semantic importance value is determined as the central node representation value. A central node threshold is used to determine the central node. Based on the central node, several central basic knowledge graphs and corresponding central graph vectors are obtained from the highway database. Simultaneously, based on each anchored node, several links are obtained from the highway database. The path betweenness and node correlation of each link are calculated. The weighted sum of the path betweenness and node correlation is determined as the key link representation value. A key link threshold is used to determine the key link. Based on several key links, several links are obtained from the highway database. The system constructs a knowledge graph of key links and key link vectors. Further, it selects central and related graph vectors with high similarity to construct a related knowledge graph, accurately locating the root causes of deviations and improving the targeting and effectiveness of graph correction. Based on quantitative indicators, it selects core knowledge to enhance the knowledge density and professional adaptability of the graph. The process, including multi-granularity analysis, core unit selection, and vector similarity matching, can be automated, eliminating the need for manual verification and significantly reducing the manual cost of deviation correction. Simultaneously, by accurately locating core correction targets, it avoids the time consumption of full reconstruction, significantly shortening the optimization cycle of the deviation graph and ensuring the timeliness of knowledge graph construction. This further improves the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data.

[0048] Furthermore, this invention acquires historical event data from a highway database under different traffic volumes and timestamps, analyzes the types of events that are likely to occur and their probabilities under different traffic volumes and time periods, combines the corresponding event type data with a pre-trained knowledge graph or associated knowledge graph for simulation, and obtains the probability of events occurring in the simulated environment. Based on the deviation between the actual event probability and the simulated event probability, it can analyze whether the pre-trained knowledge graph or associated knowledge graph used has a strong or weak correlation with the actual situation. In response to a strong correlation, the pre-trained knowledge graph or associated knowledge graph is identified as a new knowledge graph and stored in the highway database. In response to a weak correlation, the preset node set of uploaded tags on the terminal device is corrected. This improves the accuracy of preset node selection in the next knowledge graph construction process and further enhances the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method for rapid knowledge graph construction based on multi-source heterogeneous data in this embodiment;

[0050] Figure 2A flowchart for generating the confidence scores of each anchored basic knowledge graph in this embodiment;

[0051] Figure 3 This is the decision logic diagram that constitutes the associated knowledge graph in this embodiment;

[0052] Figure 4 This is a logic diagram for classifying pre-trained knowledge graphs or associated knowledge graphs in this embodiment. Detailed Implementation

[0053] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0054] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0055] Specifically, the upload tag is a feature of a mobile terminal that can upload data to the central control system for analysis. It can characterize the level of knowledge expertise of the mobile terminal in different professional fields and can call several preset nodes in professional fields with a high level of knowledge expertise. The assessment of the level of knowledge expertise is set by those skilled in the art based on the accuracy requirements of the map construction, and will not be elaborated here.

[0056] Please see Figure 1 The diagram shown is a flowchart of a method for rapidly constructing a knowledge graph based on multi-source heterogeneous data in this embodiment. This embodiment provides a method for rapidly constructing a knowledge graph based on multi-source heterogeneous data, including:

[0057] Step S1: Obtain real-time uploaded data, extract keywords and upload tags from the uploaded data, call several preset nodes based on the upload tags, generate several initial nodes by combining the basic text model and keywords, and select nodes that are the same as the preset nodes and initial nodes to generate an anchor node set.

[0058] Step S2: Based on the anchor node set, obtain several anchor basic knowledge graphs in the highway database, obtain the relationship mapping of each anchor basic knowledge graph, and combine the uploaded tags to determine the confidence weight of the relationship mapping to generate the confidence of each anchor basic knowledge graph.

[0059] Step S3: Based on the confidence bias rate, generate a pre-trained knowledge graph or perform multi-granularity analysis on the anchor node set. Based on the results of the multi-granularity analysis, determine several central nodes and several key links. Based on the several central nodes, obtain several central basic knowledge graphs from the highway database to generate central graph vectors.

[0060] Step S4: Generate the association graph vectors corresponding to each key link, and construct the association knowledge graph based on the similarity between the central graph vector and the association graph vectors;

[0061] Step S5: Based on the traffic flow and timestamp of the uploaded data, match the event type and the corresponding event probability. Combine the event type to simulate the operation of the pre-trained knowledge graph or the associated knowledge graph to generate the event simulation probability. Combine the event probability and the event simulation probability to generate the event accuracy coefficient.

[0062] Step S6: Classify the pre-trained knowledge graph or associated knowledge graph based on the event accuracy coefficient, and adjust the uploaded tags or update the highway database based on the classification results.

[0063] Specifically, there are no restrictions on the source of uploaded data. For example, it can originate from active input outside the system, such as event handling information uploaded by users; it can also originate from video surveillance, which will not be elaborated further here.

[0064] Specifically, in implementation, the uploaded data can be text, images, or videos. After extracting feature text from data of different formats, the data is analyzed. There are no restrictions on the specific method of keyword extraction. For example, feature text analysis can be performed using a pre-trained BERT model, and then keywords can be extracted through an attention mechanism. Those skilled in the art can also use other methods, as long as they are reasonable, which will not be elaborated here.

[0065] Specifically, there is a one-to-one correspondence between the terminal device and the uploaded tag.

[0066] Specifically, this invention extracts keywords from the acquired real-time uploaded data and, combined with the keywords and a basic text model of the highway professional field, determines the entity nodes in the uploaded data that can be used to construct a knowledge graph in that professional field. Simultaneously, it extracts upload tags from the terminal devices corresponding to the uploaded data, sets upload tags to differentiate the professional capabilities of the terminal devices, and calls entity nodes in the technical fields that the terminal devices are familiar with and proficient in based on the upload tags. The entity nodes under these two analysis conditions are aggregated to obtain the anchor node set in the uploaded data. This allows for preliminary screening of uploaded data while improving the accuracy of entity node identification, breaking through semantic silos, achieving rapid preliminary screening of multi-source heterogeneous data, improving graph construction efficiency, enhancing the domain adaptability of terminal device data, optimizing the professional attributes of the knowledge graph, and improving the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data.

[0067] Specifically, in step S1, the process of selecting nodes that are the same as the preset node and the initial node to generate the anchor node set includes,

[0068] Obtain the upload tag corresponding to the terminal device that uploaded the data, and call the preset node corresponding to the upload tag;

[0069] Extract keywords from the uploaded data to match the corresponding base words in the basic text model, and then call the initial node corresponding to the base word;

[0070] Compare the preset nodes and the initial nodes, and select the same nodes to generate the anchor node set.

[0071] Specifically, the basic text model defines the core concepts of highway engineering, subdivides entity categories, defines entity attributes, standardizes the relationships between entities, and defines domain constraints. Based on the data text such as the normative texts of highway engineering, maintenance and operation records, and traffic incident data, it establishes several entity nodes corresponding to the basic words in highway engineering. It can analyze the input text and output the basic words and their corresponding entity nodes.

[0072] Specifically, the preset nodes and the initial nodes are compared. If there are no identical nodes, all preset nodes and the initial nodes are used to generate an anchor node set.

[0073] Please see Figure 2 As shown, this is a flowchart of generating the confidence scores of each anchored basic knowledge graph in this embodiment. In step S2, the process of generating the confidence scores of each anchored basic knowledge graph includes...

[0074] Step S201: Input the anchor node set into the highway database to generate several anchoring basic knowledge graphs related to the anchor node set;

[0075] Step S202: Obtain the set of mapping knowledge proportions of the relationship mapping of each anchored basic knowledge graph in different professional fields;

[0076] Step S203: Determine the similarity between the mapped knowledge proportion set and the tag knowledge proportion set of the uploaded tags as the tag association degree of the relation mapping;

[0077] Step S204: Determine the confidence weight and confidence of each relation mapping based on the association degree of several labels;

[0078] Step S205: Determine the weighted sum of the confidence scores of each relation mapping as the confidence score of the corresponding anchored basic knowledge graph.

[0079] Specifically, entity nodes in the relation mappings of each anchored basic knowledge graph are classified into corresponding professional knowledge domains. The ratio of the number of entity nodes in each professional knowledge domain to the total number of entity nodes in that relation mapping is defined as the mapping knowledge proportion set. The similarity between the mapping knowledge proportion set and the tag knowledge proportion set of the uploaded tag is calculated as the tag relevance. and tag knowledge proportion set The label relevance MP; where j = 1, 2, ..., m; label relevance: .

[0080] Specifically, the label correlation degree of the relation mapping is determined as the confidence degree of the relation mapping, and the confidence weights are allocated according to the ratio between the label correlation degrees of each relation mapping, with the sum of the confidence weights of each relation mapping being 1.

[0081] Specifically, the weighted sum of the confidence levels of each relation mapping in the anchored basic knowledge graph is determined as the confidence level of the anchored basic knowledge graph.

[0082] Specifically, in step S3, the standard deviation of the confidence level of several anchored basic knowledge graphs is calculated as the confidence level deviation rate.

[0083] In one specific embodiment, in response to a confidence bias rate being less than or equal to a preset confidence bias threshold, several anchored basic knowledge graphs are constructed into a pre-trained knowledge graph.

[0084] In response to a confidence deviation rate exceeding a preset confidence deviation threshold, multi-granularity analysis is performed on the anchor node set.

[0085] Specifically, the confidence deviation threshold is the product of the confidence deviation threshold reference value and the confidence deviation factor. The confidence deviation threshold reference value is the average confidence deviation threshold at the same timestamp in historical data. The confidence deviation factor can be set by those skilled in the art according to the accuracy requirements of constructing the knowledge graph. The higher the accuracy requirement, the smaller the value should be. The value range can be [0.85, 0.95], preferably 0.9.

[0086] Specifically, this invention inputs a pre-processed set of anchor nodes into a pre-constructed highway database to generate several anchor-based basic knowledge graphs. It analyzes the proportion of mapping knowledge in different professional fields for each anchor-based basic knowledge graph, further analyzes the similarity between this proportion and the proportion of tag knowledge in the uploaded tags, and determines this as the tag correlation degree of the relationship mapping. Then, it assigns weights to each relationship mapping based on the tag correlation degree, thereby calculating the confidence level of the anchor-based basic knowledge graph corresponding to the relationship mapping. Finally, it analyzes the confidence level deviation rate of the anchor-based basic knowledge graph. For anchored basic knowledge graphs with low deviation, pre-trained knowledge graphs are constructed. For anchored basic knowledge graphs with high deviation, multi-granularity analysis is performed to construct a quantitative confidence evaluation system to ensure the reliability of the graph knowledge and accurately quantify the matching degree between relation mapping and terminal data domain attributes. For anchored basic knowledge graphs with low deviation, pre-trained knowledge graphs are directly constructed to achieve rapid reuse of high-quality knowledge. For graphs with high deviation, multi-granularity analysis is carried out to locate the root cause of deviation and accurately correct it layer by layer from entity granularity, relation granularity, and attribute granularity, further improving the accuracy of rapid construction of knowledge graphs based on multi-source heterogeneous data.

[0087] Specifically, in step S3, the process of generating the central map vector includes,

[0088] Multi-granularity analysis is performed on the anchor node set to obtain the centrality value and semantic importance value of each anchor node;

[0089] The weighted sum of the centrality value and the semantic importance value is determined as the representation value of the central node, and the anchor node is determined as a central node based on the central node threshold.

[0090] If the value of the central node is greater than or equal to the threshold of the central node, then the anchor node is determined to be the central node.

[0091] If the value of the central node is less than the threshold of the central node, then the anchor node is determined not to be a central node.

[0092] Based on several central nodes, several central basic knowledge graphs are obtained from the highway database. These central basic knowledge graphs are then transformed into several central subgraph vectors using a graph neural network and aggregated to generate a central graph vector.

[0093] Specifically, centrality is the ratio of the number of links directly connected to a node to the total number of links, while semantic importance is the ratio of the number of times a node appears to the total number of nodes.

[0094] Specifically, the sum of the weight coefficients of the centrality value and the semantic importance value is 1, the weight coefficient of the centrality value is 0.5, and the weight coefficient of the semantic importance value is 0.5.

[0095] Specifically, the center node threshold is the product of the center node threshold reference value and the center node factor. The center node threshold reference value is the average value of the center node thresholds at the same timestamp in historical data. The center node factor can be set by those skilled in the art according to the accuracy requirements of constructing the knowledge graph. The higher the accuracy requirement, the larger the value should be. The value range can be [0.95, 1.1], preferably 1.05.

[0096] Specifically, several central basic knowledge graphs associated with several central nodes are selected from the highway database. These central basic knowledge graphs are transformed into several central subgraph vectors using a graph neural network. For each node, the vectors of its neighboring nodes are aggregated. Feature transformation and dimensionality reduction are performed through convolutional layers. Pooling is then performed on all node vectors in the last layer to obtain the central subgraph vector. Utilizing the message passing mechanism of the graph neural network, the information of neighboring nodes is aggregated layer by layer, which can accurately capture the local structural dependencies and node interaction features of the subgraph vectors. Several central subgraph vectors are aggregated to generate a central graph vector, which will not be elaborated further here.

[0097] Please see Figure 3 As shown, this is the logic diagram for determining the formation of the associated knowledge graph in this embodiment. In step S4, the process of forming the associated knowledge graph includes:

[0098] Based on each anchor node, several links are obtained from the highway database, and the path betweenness and node correlation of each link are calculated.

[0099] The weighted sum of path betweenness and node correlation is used to determine the critical link characterization value, and the critical link threshold is used to determine whether a link is a critical link.

[0100] If the critical link characteristic value is greater than or equal to the critical link threshold, the link is determined to be a critical link.

[0101] If the critical link characteristic value is less than the critical link threshold, the link is determined not to be a critical link.

[0102] Based on several key links, knowledge graphs of several key links are obtained from the highway database. The knowledge graphs of several key links are transformed into several link subgraph vectors through graph neural networks and then aggregated to generate key link vectors.

[0103] Specifically, the path betweenness is the proportion of the shortest path through the link to the total number of shortest paths in all links, and the node correlation is the average of the representation values ​​of the central nodes at both ends of the link.

[0104] Specifically, the sum of the weight coefficients of path betweenness and node association is 1, the weight coefficient of path betweenness is 0.5, and the weight coefficient of node association is 0.5.

[0105] Specifically, the critical link threshold is the product of the critical link threshold reference value and the critical link factor. The critical link threshold reference value is the average critical link threshold at the same timestamp in historical data. The critical link factor can be set by those skilled in the art according to the accuracy requirements of constructing the knowledge graph. The higher the accuracy requirement, the larger the value should be. The value range can be [0.95, 1.2], preferably 1.1.

[0106] Calculate the similarity between the central graph vector and the associated graph vector, and aggregate the central basic knowledge graph corresponding to the central graph vector with a similarity greater than a preset similarity threshold and the key link knowledge graph corresponding to the associated graph vector to form an associated knowledge graph.

[0107] Specifically, the similarity between the central graph vector and the associated graph vector is calculated. and associated spectrum vectors The similarity is NP; where j = 1, 2, ..., m;

[0108] Similarity: .

[0109] Specifically, the purpose of setting a preset similarity threshold is to characterize the consistency between the central graph vector and the associated graph vector. The preset similarity threshold is selected within the range [0.85, 0.95].

[0110] Specifically, this invention responds to multi-granularity analysis of anchored basic knowledge graphs with large deviations. It performs multi-granularity analysis on the anchored node set to obtain the centrality and semantic importance values ​​corresponding to each anchored node. The weighted sum of the centrality and semantic importance values ​​is determined as the central node representation value. A central node threshold is then used to determine the central node. Based on the central node, several central basic knowledge graphs and corresponding central graph vectors are obtained from a highway database. Simultaneously, based on each anchored node, several links are obtained from the highway database. The path betweenness and node correlation of each link are calculated. The weighted sum of the path betweenness and node correlation is determined as the critical link representation value. A critical link threshold is then used to determine the critical link. Based on these critical links, several critical links are obtained from the highway database. The knowledge graph and key link vectors are further selected by constructing a related knowledge graph by choosing central graph vectors and related graph vectors with high similarity. This accurately locates the root cause of deviations, improving the targeting and effectiveness of graph correction. Core knowledge is selected based on quantitative indicators to enhance the knowledge density and professional adaptability of the graph. The process of multi-granularity analysis, core unit selection, and vector similarity matching can all be automated, eliminating the need for manual verification and significantly reducing the manual cost of deviation correction. At the same time, by accurately locating the core correction objects, the time consumption caused by full reconstruction is avoided, significantly shortening the optimization cycle of the deviation graph and ensuring the timeliness of knowledge graph construction. This further improves the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data.

[0111] Specifically, in step S5, the process of generating the event accuracy coefficient by combining the event probability and the event simulation probability includes,

[0112] Based on the traffic flow and timestamp of the uploaded data, obtain the corresponding event types and probabilities from the highway database;

[0113] Input the event type into a pre-trained knowledge graph or an associated knowledge graph to generate simulated event probabilities through simulation.

[0114] The probability deviation between the simulated event probability and the actual event probability is determined as the event accuracy coefficient.

[0115] Specifically, the event accuracy coefficient = (event simulation probability - event probability) / event probability.

[0116] Please see Figure 4 As shown, it is the logic diagram for classifying the pre-trained knowledge graph or the associated knowledge graph in this embodiment. In step S6, in response to the event accuracy coefficient being greater than or equal to the preset event accuracy threshold, the pre-trained knowledge graph or the associated knowledge graph corresponding to the event accuracy coefficient is determined to be a strongly associated graph.

[0117] If the accuracy coefficient of an event is less than a preset event accuracy threshold, the pre-trained knowledge graph or associated knowledge graph corresponding to the event accuracy coefficient is determined to be a weakly associated graph.

[0118] Specifically, the purpose of setting the event accuracy threshold is to characterize the consistency between the pre-trained knowledge graph or associated knowledge graph and the actual historical situation. The event accuracy threshold is selected within the range [0.9, 0.95].

[0119] Specifically, in step S6, the process of adjusting the upload tag includes,

[0120] The knowledge graph corresponding to the accuracy coefficient of an event is a weakly related graph. We obtain the weakly related node set in the weakly related graph and the preset node set and merge them to update the uploaded tags.

[0121] The pre-trained knowledge graph corresponding to the accuracy coefficient of the event is a weakly associated graph. The initial node set and the preset node set are merged to update the uploaded tags.

[0122] The pre-trained knowledge graph corresponding to the accuracy coefficient of the event is a strongly associated graph. The pre-trained knowledge graph or associated knowledge graph is identified as a new knowledge graph and stored in the highway database.

[0123] The preset node set is a collection of preset nodes called by the upload tag, while the initial node set is a collection of initial nodes generated by combining the basic text model and keywords.

[0124] Specifically, this invention acquires historical event data from a highway database under different traffic volumes and timestamps, analyzes the types of events that are likely to occur and their probabilities under different traffic volumes and time periods, combines the corresponding event type data with a pre-trained knowledge graph or associated knowledge graph for simulation, and obtains the probability of events occurring in the simulated environment. Based on the deviation between the actual event probability and the simulated event probability, it can analyze whether the pre-trained knowledge graph or associated knowledge graph used has a strong or weak correlation with the actual situation. In response to a strong correlation, the pre-trained knowledge graph or associated knowledge graph is identified as a new knowledge graph and stored in the highway database. In response to a weak correlation, the preset node set of uploaded tags on the terminal device is corrected. This improves the accuracy of preset node selection in the next knowledge graph construction process and further enhances the accuracy of rapid knowledge graph construction based on multi-source heterogeneous data.

Claims

1. A method for rapid construction of knowledge graphs based on multi-source heterogeneous data, characterized in that, include: Extract keywords and upload tags from the uploaded data respectively. Based on the upload tags, call several preset nodes. Combine the basic text model and keywords to generate several initial nodes. Select nodes that are the same as the preset nodes and initial nodes to generate an anchor node set. Based on the anchor node set, several anchor basic knowledge graphs are obtained from the highway database. The relationship mapping of each anchor basic knowledge graph is obtained and the confidence weight of the relationship mapping is determined by combining the uploaded tags to generate the confidence of each anchor basic knowledge graph. Based on the confidence bias rate, a pre-trained knowledge graph is generated or a multi-granularity analysis is performed on the anchor node set to determine several central nodes and several key links. Based on the several central nodes, several central basic knowledge graphs are obtained to generate a central graph vector. Generate association graph vectors corresponding to each key link, and construct an association knowledge graph based on the similarity between the central graph vector and the association graph vectors; Based on the uploaded data, the event type and event probability are matched, and the event type is combined with the event type to simulate the operation of the pre-trained knowledge graph or the associated knowledge graph to generate the event simulation probability. The event accuracy coefficient is then generated by combining the event probability. Based on the event accuracy coefficient, the pre-trained knowledge graph or the associated knowledge graph is classified to adjust the uploaded tags or update the highway database; In response to a confidence bias rate less than or equal to a preset confidence bias threshold, several anchored basic knowledge graphs are constructed into a pre-trained knowledge graph. In response to a confidence deviation rate exceeding a preset confidence deviation threshold, multi-granularity analysis is performed on the anchor node set. The process of generating the central map vector includes, Multi-granularity analysis is performed on the anchor node set to obtain the centrality value and semantic importance value of each anchor node; The weighted sum of the centrality value and the semantic importance value is determined as the central node representation value, and the anchor node is determined as a central node based on the central node threshold. Based on several central nodes, several central basic knowledge graphs are obtained from the highway database. The several central basic knowledge graphs are transformed into several central subgraph vectors through graph neural networks and then aggregated to generate central graph vectors. The process of constructing a knowledge graph includes, Based on each anchor node, several links are obtained from the highway database, and the path betweenness and node correlation of each link are calculated. The weighted sum of the path betweenness and the node correlation is determined as the critical link characterization value, and the critical link threshold is used to determine whether the link is a critical link. Based on several key links, several key link knowledge graphs are obtained from the highway database. The key link knowledge graphs are then transformed into several link subgraph vectors and aggregated using a graph neural network to generate key link vectors. Calculate the similarity between the central graph vector and the associated graph vector, and aggregate the central basic knowledge graph corresponding to the central graph vector with a similarity greater than a preset similarity threshold and the key link knowledge graph corresponding to the associated graph vector to form an associated knowledge graph.

2. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 1, characterized in that, The process of generating an anchor node set by selecting nodes that are the same as the preset node and the initial node includes, The upload tag corresponding to the terminal device that acquires the uploaded data is then invoked, and the preset node corresponding to the upload tag is called. Extract keywords from the uploaded data to match the corresponding basic words in the basic text model, and call the initial node corresponding to the basic words; The preset node and the initial node are compared, and the same nodes are selected to generate the anchor node set.

3. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 2, characterized in that, The process of generating the confidence scores for each of the anchored basic knowledge graphs includes, The anchoring node set is input into the highway database to generate several anchoring basic knowledge graphs related to the anchoring node set; Obtain the set of mapping knowledge proportions in different professional fields for the relationship mapping of each anchored basic knowledge graph; The similarity between the mapped knowledge proportion set and the tag knowledge proportion set of the uploaded tag is determined as the tag relevance of the relation mapping; The confidence weight and confidence level of each relation mapping are determined based on the association degree of several labels. The weighted sum of the confidence scores of each of the aforementioned relation mappings is determined to be the confidence score of the corresponding anchored basic knowledge graph; The mapping knowledge proportion set is the ratio of the number of entity nodes included in each professional knowledge domain to the total number of entity nodes in the relation mapping.

4. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 3, characterized in that, The standard deviation of the confidence scores for several anchored basic knowledge graphs is calculated as the confidence bias rate.

5. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 4, characterized in that, The process of generating an event accuracy coefficient by combining the event probability and the event simulation probability includes, Based on the traffic flow and timestamp of the uploaded data, obtain the corresponding event types and probabilities from the highway database; The event type is input into the pre-trained knowledge graph or the associated knowledge graph to generate event simulation probabilities through simulation. The probability deviation between the simulated probability of the event and the actual event probability is determined as the event accuracy coefficient.

6. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 5, characterized in that, If the event accuracy coefficient is greater than or equal to a preset event accuracy threshold, then the pre-trained knowledge graph or associated knowledge graph corresponding to the event accuracy coefficient is determined to be a strongly associated graph. If the event accuracy coefficient is less than a preset event accuracy threshold, the pre-trained knowledge graph or associated knowledge graph corresponding to the event accuracy coefficient is determined to be a weakly associated graph.

7. The method for rapid construction of knowledge graphs based on multi-source heterogeneous data according to claim 6, characterized in that, The process of adjusting upload tags includes, In response to the fact that the knowledge graph corresponding to the accuracy coefficient of the event is a weakly related graph, the weakly related node set and the preset node set in the weakly related graph are obtained and fused to update the uploaded tag; The pre-trained knowledge graph corresponding to the event accuracy coefficient is a weakly associated graph. The initial node set and the preset node set are fused to update the uploaded tags. The preset node set is a collection of preset nodes called by the upload tag, and the initial node set is a collection of initial nodes generated by combining the basic text model and keywords.

Images