Mass network data intelligent mining system based on AI technology

Abstract

The invention relates to the field of data processing, and discloses an AI technology-based massive network data intelligent mining system, which comprises a standardized data set acquisition module, a multi-dimensional feature vector acquisition module, an initial knowledge graph construction module, a target event association path construction module and a final mining result acquisition module, performing standardization processing on the original data set to obtain a standardized data set; extracting entity features and relation features in the standardized data set to obtain multi-dimensional feature vectors of the massive network data; mapping the multi-dimensional feature vector to a topological structure to obtain an initial knowledge graph; performing deep reasoning on the initial knowledge graph by using a graph neural network to obtain a target event association path of the massive network data; optimizing the target event association path based on a preset service rule to obtain a final mining result of the massive network data; according to the invention, the efficiency of intelligent mining of massive network data can be improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data processing, in particular to a massive network data intelligent mining system based on AI technology. BACKGROUND

[0002] In the prior art, when massive network data is intelligently mined, the original data contains various types such as numerical and categorical types, the forms of different types of data are significantly different, and there is a lack of unified standardization processing mechanism, so that a large amount of time is consumed for format adjustment and scale unification before the data enters the mining process, which increases the complexity and time consumption of data preprocessing; meanwhile, the extraction of entities and relationships in the data is not systematic, and it is difficult to form a structured feature representation, so that the subsequent analysis process lacks a reliable basis, and the efficiency of data intelligent mining is reduced as a whole.

[0003] In terms of mining effect, in the prior art, when the data correlation structure is constructed, the node connection density and neighborhood similarity are not effectively combined for optimization, so that a large amount of redundant information exists in the correlation relationship, which affects the identification of the core correlation; in addition, in the process of reasoning the correlation path and optimizing the result, the accuracy of the path effectiveness is not accurately judged, and the conflict path is not reasonably processed, so that the result mined often contains invalid or contradictory contents, which cannot accurately reflect the internal logic between the data, and cannot meet the demand for deep and accurate mining of massive network data, so that the effect of data intelligent mining is not good. SUMMARY

[0004] The application provides a massive network data intelligent mining system based on AI technology, which mainly aims to solve the problem of low efficiency of massive network data intelligent mining.

[0005] To achieve the above purpose, the application provides a massive network data intelligent mining system based on AI technology, characterized in that the system comprises a standardized data set acquisition module, a multi-dimensional feature vector acquisition module, an initial knowledge graph construction module, a target event correlation path construction module and a final mining result acquisition module, wherein: The standardized data set acquisition module is used for standardizing the original data set to obtain a standardized data set of massive network data. The multi-dimensional feature vector acquisition module is used for extracting entity features and relationship features in the standardized data set to obtain a multi-dimensional feature vector of the massive network data. The initial knowledge graph construction module is used for mapping the multi-dimensional feature vector to a topological structure to obtain an initial knowledge graph of the massive network data. The target event association path construction module is configured to perform deep reasoning on the initial knowledge graph by using a graph neural network to obtain a target event association path of the massive network data. The final mining result acquisition module is configured to optimize the target event association path based on a preset business rule to obtain a final mining result of the massive network data.

[0006] In a preferred embodiment, when the normalized data set acquisition module performs standardization processing on the original data set to obtain a normalized data set of the massive network data, the normalized data set acquisition module is specifically configured to: perform feature normalization processing on numerical data in the original data set to obtain feature weight data of the original data set; perform one-hot encoding on categorical data in the original data set to obtain encoded categorical data; aggregate the weighted numerical data and the categorical data into the normalized data set of the massive network data.

[0007] In a preferred embodiment, when the multi-dimensional feature vector acquisition module performs extraction of entity features and relationship features in the normalized data set to obtain a multi-dimensional feature vector of the massive network data, the multi-dimensional feature vector acquisition module is specifically configured to: extract structured entities of a text field in the normalized data set to obtain an entity set of the massive network data; analyze a syntax dependency relationship of entity units in the entity set to obtain a syntax relationship feature of the text field; perform topological reconstruction on the vectorized entity set and the syntax relationship feature to obtain the multi-dimensional feature vector of the massive network data.

[0008] In a preferred embodiment, when the initial knowledge graph construction module performs mapping of the multi-dimensional feature vector to a topological structure to obtain an initial knowledge graph of the massive network data, the initial knowledge graph construction module is specifically configured to: construct an initial graph structure of the massive network data based on the multi-dimensional feature vector; optimize an association relationship of the initial graph structure based on an influence weight of node connection density and neighborhood similarity in the initial graph structure; perform topological structure construction on the optimized initial graph structure to obtain the initial knowledge graph of the massive network data.

[0009] In a preferred implementation, the initial knowledge graph construction module, when performing influence weight based on node connection density and neighborhood similarity in the initial graph structure, and optimizing the association relationship of the initial graph structure, is specifically configured to: calculate the influence weight of the node connection density and the neighborhood similarity based on the initial graph structure to obtain the node influence weight of the initial graph structure, wherein the calculation formula of the node influence weight is as follows: In the formula, is the weight of node , is a damping coefficient is a set of adjacent nodes of node , is the weight of node , is a cosine similarity, is a feature vector of node , is a feature vector of node , is the out-degree of node .

[0010] Optimizing the association relationship of the initial graph structure to obtain an optimized knowledge graph.

[0011] In a preferred implementation, the target event association path construction module, when performing deep reasoning on the initial knowledge graph using a graph neural network to obtain a target event association path of the massive network data, is specifically configured to: Update the node features of the initial knowledge graph; Aggregate the updated node feature vectors to obtain a node state vector of the massive network data; Connect the path association relationship based on the node state vector to obtain a target event association path of the massive network data.

[0012] In a preferred implementation, the initial knowledge graph performs deep reasoning, and when performing deep reasoning on the initial knowledge graph using a graph neural network to obtain a target event association path of the massive network data, is specifically configured to: Calculate the association strength parameter of adjacent node feature vectors, wherein the calculation formula of the association strength parameter is as follows: In the formula, is the association strength parameter of node and is a feature vector of node , is a feature vector of node , a node a weight of an edge between a node a weight of an edge between a node a norm of a vector; perform weighted fusion on the adjacent node feature vectors based on the correlation strength parameters to obtain a weighted neighborhood feature of the mass network data; perform feature fusion on the weighted neighborhood feature and the center node feature vector to obtain an updated node feature vector; perform path mining on the updated node feature vector to obtain a target event correlation path of the mass network data.

[0013] In one preferred implementation, when the final mining result acquisition module performs optimization on the target event correlation path based on a preset business rule to obtain a final mining result of the mass network data, it is specifically configured to: perform rule matching on the target event correlation path to obtain an effective path set of the target event correlation path; perform conflict detection on the target event correlation path to obtain a conflict path set of the target event correlation path; strengthen a core logic chain of the target event correlation path based on the effective path set to obtain a core logic path of the target event correlation path; perform path pruning on the conflict path set, and integrate the path after pruning and the core logic path into the final mining result.

[0014] In one preferred implementation, when the final mining result acquisition module performs rule matching on the target event correlation path to obtain an effective path set that meets a preset business rule, it is specifically configured to: compare an event node sequence of the target event correlation path with an event logic template to obtain a matching degree parameter of the target event correlation path; when the matching degree parameter meets a preset condition, extract a corresponding path segment to generate the effective path set.

[0015] In one preferred implementation, when the final mining result acquisition module performs path pruning on the conflict path set, and integrates the path after pruning and the core logic path into the final mining result, it is specifically configured to: perform structural reconstruction on a conflict topology subgraph; perform node positioning on the conflict topology subgraph to obtain a coordinate identifier of a path divergence point; cut off a connection relationship based on the coordinate identifier to obtain a topology structure of a pruned path; Topological structure of the pruned path is topologically fused with the core logical path to obtain a final mining result of the mass network data.

[0016] Compared with the prior art, the present application has the following beneficial effects: 1. The method eliminates the format differences of different types of data by standardizing the original data, integrates numerical and categorical data to form a unified standardized data set, and provides consistent and high-quality basis for subsequent mining; at the same time, the entity features and relationship features are accurately extracted, which are converted into a multi-dimensional feature vector and an initial knowledge graph is constructed, the association structure between data is clearly presented, and the accuracy and systematicness of mass network data mining are effectively improved, so that the mining process is more in line with the internal logic of data.

[0017] 2. With the help of graph neural network, the initial knowledge graph is deeply inferred, the node features are accurately updated, and the target event association path is mined, the path is optimized combined with the preset business rules, the effective path is screened, the conflict path is processed, and the final result is integrated, which highlights the core logic and retains the effective information, significantly enhances the accuracy of the association path and the integrity of the result, and improves the depth and comprehensiveness of intelligent mining of mass network data. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A system architecture diagram of an intelligent mining system of mass network data based on AI technology is provided for an embodiment of the present application. The implementation of the object of the present application, functional characteristics and advantages will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the object, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments belong to part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise. "Multiple" generally includes at least two.

[0021] Depending on context, the word "if" or "if" can be interpreted to mean "when" or "while" or "in response to determining" or "in response to detecting", as the case can be. Similarly, depending on context, the phrase "if it is determined" or "if [a stated condition or event] is detected" can be interpreted to mean "when it is determined" or "in response to determining" or "when [a stated condition or event] is detected" or "in response to detecting [a stated condition or event] ", as the case can be.

[0022] In addition, the step timing in each of the following method embodiments is only an example, not a strict limitation.

[0023] In fact, the service end device deployed by the AI technology-based massive network data intelligent mining system can be composed of one or more devices. The AI technology-based massive network data intelligent mining system can be implemented as a business instance, a virtual machine, or a hardware device. For example, the AI technology-based massive network data intelligent mining system can be implemented as a business instance deployed on one or more devices in a cloud node. In short, the AI technology-based massive network data intelligent mining system can be understood as a software deployed on a cloud node, which provides an AI technology-based massive network data intelligent mining system for each user end. Alternatively, the AI technology-based massive network data intelligent mining system can also be implemented as a virtual machine deployed on one or more devices in a cloud node. The virtual machine has application software for managing each user end. Alternatively, the AI technology-based massive network data intelligent mining system can also be implemented as a service end composed of a plurality of same or different types of hardware devices, and one or more hardware devices are provided to provide an AI technology-based massive network data intelligent mining system for each user end.

[0024] In terms of implementation, the AI technology-based massive network data intelligent mining system and the user end are mutually adapted. That is, the AI technology-based massive network data intelligent mining system is installed as an application on a cloud service platform, and the user end is a client that establishes a communication connection with the application; or the AI technology-based massive network data intelligent mining system is implemented as a website, and the user end is a webpage; or the AI technology-based massive network data intelligent mining system is implemented as a cloud service platform, and the user end is a small program in an instant messaging application.

[0025] As Figure 1 shown is a system architecture diagram of an AI technology-based massive network data intelligent mining system according to an embodiment of the present application.

[0026] The AI technology-based massive network data intelligent mining system 100 can be set in a cloud server, and in terms of implementation, can be used as one or more service devices, or installed as an application on a cloud (such as a server of a mobile service operator, a server cluster, etc.), or developed as a website. According to the functions implemented, the AI technology-based massive network data intelligent mining system 100 can include a standardized data set acquisition module 101, a multi-dimensional feature vector acquisition module 102, an initial knowledge graph construction module 103, a target event association path construction module 104, and a final mining result acquisition module 105. The modules in the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, and are stored in the memory of the electronic device.

[0027] In the embodiment of the present application, each of the above modules can be independently implemented and called by other modules. Here, calling can be understood as connecting a module to multiple modules of another type and providing corresponding services to the connected multiple modules. In the AI technology-based massive network data intelligent mining system provided in the embodiment of the present application, without modifying the program code, the application range of the AI technology-based massive network data intelligent mining system architecture can be adjusted by adding modules and directly calling them, realizing cluster-type horizontal expansion, so as to achieve the purpose of quickly and flexibly expanding the AI technology-based massive network data intelligent mining system. In actual application, the above modules can be set in the same device or different devices, or in a virtual device, such as a service instance in a cloud server.

[0028] The following will describe the components and specific work flow of the AI technology-based massive network data intelligent mining system with reference to specific embodiments: The standardized data set acquisition module 101 is configured to standardize the original data set to obtain a standardized data set of massive network data. In the embodiment of the present application, when the standardized data set acquisition module performs standardization on the original data set to obtain a standardized data set of massive network data, it is specifically configured to: perform feature normalization on the numerical data in the original data set to obtain feature weight data of the original data set; perform one-hot encoding on the category data in the original data set to obtain encoded category data; combine the weighted numerical data and the category data to obtain the standardized data set of massive network data.

[0029] Specifically, the numerical data in the original data set is subjected to feature normalization processing to obtain the original data set feature weight data.

[0030] Further, all data represented in numerical form, such as user access duration, click count, and other numerical data, are filtered from the original data set.

[0031] Further, the value range of these data is determined, i.e., the maximum and minimum values of each numerical feature.

[0032] Further, each data value is subtracted by the minimum value of the feature and then divided by the difference between the maximum and minimum values, so that the processed data values are all between 0 and 1. These processed values are the original data set feature weight data.

[0033] Specifically, the categorical data in the original data set is subjected to one-hot encoding to obtain the encoded categorical data.

[0034] Further, all data representing classification information, such as user region and device type, are extracted from the original data set.

[0035] Further, each category in each categorical feature is assigned a unique binary bit. When a sample belongs to the category, the corresponding binary bit is marked as 1, and the binary bits corresponding to other categories are marked as 0.

[0036] Further, each categorical data is converted into an encoding consisting of 0 and 1 in this way. These encodings are the encoded categorical data.

[0037] Specifically, the weighted numerical data and the categorical data are collected to form the normalized data set of the massive network data.

[0038] Further, the original data set feature weight data obtained through feature normalization processing (i.e., the weighted numerical data) and the encoded categorical data obtained through one-hot encoding are placed in the same data table.

[0039] Further, according to the corresponding relationship of the samples in the original data set, the weighted numerical data and the categorical data of each sample are arranged in sequence to form a collection containing the processed data of all samples. This collection is the normalized data set of the massive network data.

[0040] In summary, the numerical data in the original data set is subjected to feature normalization processing to obtain the original data set feature weight data, which can eliminate numerical scale differences, enable direct comparison of different features, and improve the accuracy of data mining.

[0041] In general, the category data in the original data set is one-hot encoded to obtain encoded category data, which can convert the category information into a machine recognizable form, provide adaptive input for intelligent mining, and enhance data utilization efficiency.

[0042] In general, the weighted numerical data and the category data are integrated into the normalized data set of the massive network data, which can integrate different types of data, form a unified analysis basis, and improve the systematicness of data mining.

[0043] The multi-dimensional feature vector acquisition module 102 is configured to extract entity features and relationship features in the normalized data set to obtain a multi-dimensional feature vector of the massive network data. In the embodiment of the present application, when the multi-dimensional feature vector acquisition module extracts entity features and relationship features in the normalized data set to obtain a multi-dimensional feature vector of the massive network data, it is specifically configured to: extract structured entities in the text field of the normalized data set to obtain an entity set of the massive network data; parse the syntactic dependency relationship of entity units in the entity set to obtain syntactic relationship features of the text field; topologically reconstruct the vectorized entity set and the syntactic relationship features to obtain a multi-dimensional feature vector of the massive network data.

[0044] Specifically, the structured entities in the text field of the normalized data set are extracted to obtain an entity set of the massive network data.

[0045] Further, all fields containing text information, such as user comments, product descriptions, and other text fields, are filtered from the normalized data set.

[0046] Further, an entity recognition tool is used to scan these text fields one by one, and according to a pre-set entity type library (including types such as name, product name, time, and place), words or phrases in the text that meet the type library standards are identified.

[0047] Further, these identified words or phrases are marked as structured entities, and all marked structured entities are summarized and sorted.

[0048] Further, a list containing entity names and corresponding types is formed, which is an entity set of the massive network data.

[0049] Specifically, the syntactic dependency relationship of entity units in the entity set is parsed to obtain syntactic relationship features of the text field.

[0050] Further, text segments containing multiple entity units are selected from the entity set, and syntax analysis tools are used to analyze the syntactic structures of these text segments.

[0051] Further, the syntactic relationships between entity units are determined by analysis, such as the subject-object relationship between "User A" and "Commodity B" in the sentence "User A buys Commodity B", and the modifier relationship between "Time C" and "Event D" in the sentence "Event D occurs at Time C".

[0052] Further, these determined syntactic relationships are recorded one by one to form entries containing entity unit names and corresponding syntactic relationships, which together constitute the syntactic relationship features of the text field.

[0053] Specifically, the vectorized entity set and the syntactic relationship features are topologically reconstructed to obtain the multi-dimensional feature vector of the massive network data.

[0054] Further, each structured entity in the entity set is converted into an ordered set of numerical values (each entity corresponds to a unique set of numerical values), forming a vectorized entity set.

[0055] Further, the numerical value sets in the vectorized entity set are arranged in association according to the syntactic relationship features between entity units.

[0056] Further, the numerical value sets corresponding to two entities with syntactic relationships are arranged adjacent to each other, and the numerical value sets of entities without direct relationships are arranged according to the order of appearance in the text.

[0057] Further, an ordered structure containing multiple dimensions is formed, each dimension corresponding to an entity feature or a relationship feature, and this ordered structure is the multi-dimensional feature vector of the massive network data.

[0058] In summary, extracting structured entities from the text field of the normalized data set to obtain the entity set of the massive network data can accurately lock the core information unit, providing a clear analysis object for subsequent mining and improving the targeting of mining.

[0059] In summary, analyzing the syntactic dependency relationships of entity units in the entity set to obtain the syntactic relationship features of the text field can reveal the internal relationships between entities, providing a relationship basis for intelligent analysis and enhancing the depth of mining.

[0060] In summary, topologically reconstructing the vectorized entity set and the syntactic relationship features to obtain the multi-dimensional feature vector of the massive network data can integrate entity and relationship features into a unified vector form, adapt to intelligent algorithm processing, and improve the efficiency of mining.

[0061] In summary, the initial knowledge graph construction module 103 is configured to map the multi-dimensional feature vector to a topology structure to obtain an initial knowledge graph of the massive network data. In the embodiment of the present application, when the initial knowledge graph construction module is used to map the multi-dimensional feature vector to a topology structure to obtain an initial knowledge graph of the massive network data, it is specifically configured to: construct an initial graph structure of the massive network data based on the multi-dimensional feature vector; optimize the correlation relationship of the initial graph structure based on the influence weight of the node connection density and the neighborhood similarity in the initial graph structure; construct a topology structure based on the optimized initial graph structure to obtain the initial knowledge graph of the massive network data.

[0062] When the initial knowledge graph construction module is used to optimize the correlation relationship of the initial graph structure based on the influence weight of the node connection density and the neighborhood similarity in the initial graph structure, it is specifically configured to: calculate the influence weight of the node connection density and the neighborhood similarity based on the initial graph structure to obtain the node influence weight of the initial graph structure, wherein the calculation formula of the node influence weight is as follows: In the formula, is the weight of the node , is a damping coefficient, is a set of adjacent nodes of the node , is the weight of the node , is a cosine similarity, is a feature vector of the node , is a feature vector of the node , is the out-degree of the node .

[0063] optimize the correlation relationship of the initial graph structure to obtain an optimized knowledge graph.

[0064] Specifically, the initial graph structure of the massive network data is constructed based on the multi-dimensional feature vector.

[0065] Further, each entity feature in the multi-dimensional feature vector is extracted as a node of the initial graph structure, and the relationship feature between the features is extracted as an edge connecting the corresponding nodes.

[0066] Further, each node is labeled with corresponding entity information, and each edge is labeled with a corresponding relationship type, in accordance with the arrangement order of the entities and the relationships in the multi-dimensional feature vector.

[0067] Further, the connections of nodes and edges are sequentially established to form a mesh structure containing all entity nodes and relationship edges, which is the initial graph structure of the massive network data.

[0068] Specifically, the influence weight of the node connection density and the neighborhood similarity based on the initial graph structure is optimized to optimize the association relationship of the initial graph structure.

[0069] Further, the connection number of each node and other nodes in the initial graph structure is counted to determine the node connection density.

[0070] Further, the feature similarity of the adjacent nodes around each node is compared to determine the neighborhood similarity according to the preset influence weight.

[0071] Further, the association strength of each edge is evaluated, the edges with high influence weight are retained, and the redundant edges with low influence weight are deleted.

[0072] Further, the connection tightness of the retained edges is adjusted according to the weight to make the association relationship more consistent with the internal logic of the data.

[0073] Specifically, the optimized initial graph structure is constructed to obtain the initial knowledge graph of the massive network data.

[0074] Further, according to the association relationship of the nodes in the optimized initial graph structure, the relative position of each node is fixed, the nodes with direct relationship are closer, and the nodes with close relationship are clustered.

[0075] Further, the direction and attribute label of each edge are simultaneously determined to form a topological framework with ordered node distribution and clear edge annotation relationship.

[0076] Further, the entity details corresponding to the nodes and the relationship explanations corresponding to the edges are supplemented in the framework.

[0077] Further, the final complete graph containing entities, relationships and hierarchical structure is the initial knowledge graph of the massive network data.

[0078] Specifically, the influence weight of the node connection density and the neighborhood similarity based on the initial graph structure is calculated to obtain the node influence weight of the initial graph structure.

[0079] Further, the actual connection number of each node and other nodes in the initial graph structure is counted, and the number is compared with the maximum connection number that the node can reach to obtain the connection density of each node.

[0080] Further, the feature coincidence degree of each node and its directly adjacent nodes is compared to determine the neighborhood similarity.

[0081] Furthermore, according to a preset ratio, the connection density and neighborhood similarity of each node are combined into a comprehensive value, which is the node influence weight of the initial graph structure.

[0082] Specifically, the association relationships of the initial graph structure are optimized to obtain an optimized knowledge graph.

[0083] Furthermore, based on the influence weight of each node, the importance of each connecting edge in the initial graph structure is evaluated, edges connecting nodes with high influence weights are retained, and redundant edges connecting nodes with low influence weights are deleted.

[0084] Furthermore, for the retained edges, the display method of connection strength is adjusted according to the node influence weight, so that the relationship between nodes in the graph highlights the core nodes and reduces the interference of irrelevant connections.

[0085] Furthermore, the graph formed after such adjustments is the optimized knowledge graph.

[0086] Specifically, the source of the parameters in the formula.

[0087] Furthermore, nodes weight The damping coefficient is a result obtained through calculation. It is a fixed value set by humans to adjust the ratio of the node's own basic weight to the influence of adjacent nodes.

[0088] Furthermore, nodes The set of adjacent nodes From the initial graph structure and nodes All directly connected nodes, nodes weight It is a node The calculation results.

[0089] Furthermore, cosine similarity By comparing nodes eigenvectors and nodes eigenvectors The similarity is obtained from the nodes. eigenvectors and nodes eigenvectors From the node and nodes Quantized representation of the included feature information, nodes out of degree Nodes in the initial graph structure The number of outward connections.

[0090] Specifically, used to compute nodes Influence weight ,in Partial representative nodes Its own base weight reflects the inherent influence of the node itself. Multiplying by the summation part represents adjacent node pairs. Its influence and contribution.

[0091] Furthermore, in the summation part, the nodes weight Reflecting nodes The magnitude of one's own influence, cosine similarity Reflecting nodes With nodes Feature similarity, nodes out of degree Used for balancing nodes The overall formula for allocating influence among multiple adjacent nodes is obtained by combining the node's own basic weight and the influence of its adjacent nodes. The final influence weight.

[0092] Specifically, when node Adjacent nodes weight When the size is larger, under the condition that other factors remain unchanged, the node weight It will get bigger.

[0093] Furthermore, when the node With adjacent nodes cosine similarity The higher the value, meaning the more similar the features of the two nodes, the better. weight The larger it is, the more likely it is to become a node out of degree When the size is larger, the node For nodes The smaller the node's influence contribution, the less effective it will be. weight The corresponding decrease.

[0094] Furthermore, the damping coefficient The larger the value, the more adjacent nodes are to each other. The higher the percentage of influence contribution, the better the node weight The more pronounced the influence of adjacent nodes, the lower the damping coefficient. The smaller the node The higher the node's own basic weight ratio, the better. weight The greater the influence of its own inherent properties.

[0095] In summary, constructing an initial graph structure for the massive network data based on the multidimensional feature vectors can transform vector forms into graph structures, intuitively presenting data relationships, providing a basic framework for data mining, and improving the display effect of data relationships.

[0096] In summary, optimizing the association relationships of the initial graph structure based on the influence weights of node connection density and neighborhood similarity can strengthen effective associations, weaken redundant associations, make the graph structure more accurate, and improve the reliability of mining.

[0097] In summary, by constructing a topology structure from the optimized initial graph structure, an initial knowledge graph of the massive network data is obtained, which can form a structured knowledge system, providing a high-quality knowledge carrier for intelligent mining and enhancing the systematic nature of the mining.

[0098] In summary, by calculating the influence weights of node connection density and neighborhood similarity based on the initial graph structure, the node influence weights of the initial graph structure are obtained, which can quantify node importance, provide accurate basis for association optimization, and improve the targeting of mining.

[0099] In summary, optimizing the relationships in the initial graph structure yields an optimized knowledge graph, which can streamline invalid associations, strengthen core connections, make the graph more aligned with data logic, and improve mining efficiency.

[0100] The target event association path construction module 104 is used to perform deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association path of the massive network data. In this embodiment of the invention, when the target event association path construction module performs deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association path of the massive network data, it is specifically used for: The node features of the initial knowledge graph are updated; The updated node feature vectors are aggregated to obtain the node state vectors of the massive network data. Based on the node state vector, feature connections are made to the path association relationship to obtain the target event association path of the massive network data.

[0101] When performing deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association path of the massive network data, the specific application is as follows: The correlation strength parameter of the adjacent node feature vector is calculated, wherein the calculation amount of the correlation strength parameter is as follows: In the formula, is the correlation strength parameter of the node and the node, is a nonlinear activation function, is the feature vector of the node is the feature vector of the node is the feature vector of the node is the weight of the edge between the node and the node is the norm of the vector; The adjacent node features are weighted and fused based on the correlation strength parameter to obtain a weighted neighborhood feature of the massive network data. The weighted neighborhood feature and the center node feature vector are fused to obtain an updated node feature vector. The updated node feature vector is path mined to obtain a target event correlation path of the massive network data.

[0102] Specifically, the node features of the initial knowledge graph are updated.

[0103] Further, each node in the initial knowledge graph is traversed, and the feature information of all directly connected nodes of the node is collected.

[0104] Further, the features of these adjacent nodes are compared with the original features of the node itself, and the features that better reflect the role of the node in the graph are retained.

[0105] Further, the repeated or secondary features are removed, and the key information unique to the adjacent nodes is supplemented to the feature description of the node to complete the update of the feature of each node.

[0106] Further, the updated node features more comprehensively reflect the association with the surrounding nodes.

[0107] Specifically, the updated node feature vectors are aggregated to obtain a node state vector of the massive network data.

[0108] Further, for each node, the updated feature vector of the node and the updated feature vectors of all adjacent nodes are placed together, and the nodes are sorted according to the closeness of the connection between the nodes.

[0109] Further, the sorted feature vectors are sequentially combined to form a comprehensive vector containing the node itself and all related node features, and the elements in each vector correspond to different feature dimensions. The comprehensive vector is the node state vector of the massive network data. ​​

[0110] Specifically, the path association relationship is feature-connected based on the node state vector, to obtain a target event association path of the mass network data.

[0111] Further, the node state vector of the starting node corresponding to the target event is compared with the node state vectors of other nodes.

[0112] Further, the adjacent node with the highest feature matching degree is found out, and a connection between the two is established, and then the above comparison and matching process is repeated with the adjacent node as a new starting point.

[0113] Further, the subsequent nodes are sequentially connected until the terminal node related to the target event is reached, and the coherent line formed by the sequentially connected nodes is the target event association path of the mass network data.

[0114] Specifically, the association strength parameter of the adjacent node feature vector is calculated.

[0115] Further, any center node is selected from the initial knowledge graph, and all adjacent nodes directly connected to the center node are found out.

[0116] Further, the feature vector of the center node and the feature vector of each adjacent node are extracted, and the number and matching degree of the same features in these feature vectors are compared.

[0117] Further, according to the principle that the more the same features and the higher the matching degree, the stronger the association, the close degree of association between each adjacent node and the center node is determined.

[0118] Further, a specific numerical value is used to represent this close degree, and this numerical value is the association strength parameter of the adjacent node feature vector.

[0119] Specifically, the adjacent node feature vectors are weighted and fused based on the association strength parameter, to obtain the weighted neighborhood feature of the mass network data.

[0120] Further, the feature vector of each adjacent node is adjusted according to the corresponding association strength parameter.

[0121] Further, the greater the association strength parameter, the greater the proportion of the feature vector of the adjacent node in the fusion process.

[0122] Further, all the adjusted adjacent node feature vectors are then combined together to form a feature set that comprehensively reflects the features of all adjacent nodes and embodies the differences in association strength, and this set is the weighted neighborhood feature of the mass network data.

[0123] Specifically, the weighted neighborhood features are fused with the center node feature vector to obtain an updated node feature vector.

[0124] Further, the adjacent node feature information contained in the weighted neighborhood features is merged with the original feature vector of the center node to retain the core features in the center node feature vector.

[0125] Further, the representative features in the weighted neighborhood features are added while the repeated or secondary features are removed to form a new feature vector that contains both the center node features and the key features of the adjacent nodes, which is the updated node feature vector.

[0126] Specifically, the updated node feature vector is path-mined to obtain the target event association path of the massive network data.

[0127] Further, starting from the updated feature vector of the starting node corresponding to the target event, the updated feature vector of the adjacent node with the highest feature matching degree in the feature vector is found.

[0128] Further, the connection between the two nodes is established, and the above matching process is repeated with the adjacent node as a new starting point to sequentially connect subsequent nodes until the end node corresponding to the target event is reached.

[0129] Further, the continuous line formed by the sequentially connected nodes is the target event association path of the massive network data.

[0130] Specifically, the source of the parameters in the formula is the association strength parameter of the node and the node . The non-linear activation function is a function artificially set.

[0131] Further, the feature vector of the node is a vector obtained by quantifying the feature information contained in the node , and the feature vector of the node is a vector obtained by quantifying the feature information contained in the node , and the weight of the edge between the node and the node is a value set according to the importance of the connection between the node and the node in the initial graph structure, and the norm of the vector is a value set according to the importance of the connection between the node and the node in the initial graph structure. is a value obtained by summing the squares of the elements in the feature vector and then taking the square root, used to represent the length of the vector.

[0132] Specifically, to calculate the association strength parameter between node and node , first, the dot product of the feature vector of node and the feature vector of node is divided by the product of the norms of the two vectors, to obtain a value reflecting the similarity of the two feature vectors.

[0133] Further, the value is multiplied by the weight of the edge between node and node , to obtain an intermediate result that comprehensively considers feature similarity and connection weight, and finally the intermediate result is processed by a nonlinear activation function to obtain the association strength parameter between node and node , so as to quantify the closeness of the association between the two nodes.

[0134] Specifically, when the feature vector of node is more similar to the feature vector of node , the ratio of the dot product to the norm product is larger, and under other conditions unchanged, the association strength parameter will be larger.

[0135] Further, when the weight of the edge between node and node is larger, the intermediate result is larger, and the association strength parameter is also larger, when the input value of the nonlinear activation function increases, its output value will change according to the function set rule (such as gradually approaching a certain maximum value), so as to make the association strength parameter present a corresponding change trend, overall, the higher the node feature similarity and the larger the edge weight, the larger the association strength parameter , and vice versa.

[0136] Overall, updating the node features of the initial knowledge graph can enrich node information, improve feature representation ability, and provide a better foundation for mining.

[0137] In general, the node state vector of the massive network data is obtained by aggregating the updated node feature vector, the node and associated information can be integrated, the data correlation is enhanced, and the mining depth is improved.

[0138] In general, the target event association path of the massive network data is obtained by performing feature connection on the path association relationship based on the node state vector, the event association can be accurately captured, the explicit clues for mining are provided, and the mining pertinence is improved.

[0139] In general, the correlation strength parameter of the adjacent node feature vector is calculated, the node correlation tightness can be accurately measured, the quantitative basis for reasoning is provided, and the mining accuracy is improved.

[0140] In general, the weighted neighborhood feature of the massive network data is obtained by performing weighted fusion on the adjacent node feature vector based on the correlation strength parameter, the important neighborhood information can be highlighted, the feature representativeness is enhanced, and the mining effectiveness is improved.

[0141] In general, the updated node feature vector is obtained by performing feature fusion on the weighted neighborhood feature and the center node feature vector, the multi-dimensional information can be integrated, the node feature connotation is enriched, and the mining depth is improved.

[0142] In general, the target event association path of the massive network data is obtained by performing path mining on the updated node feature vector, the event association clue can be efficiently locked, the mining direction is clear, and the mining pertinence is improved.

[0143] The final mining result acquisition module 105 is configured to optimize the target event association path based on a preset business rule, and obtain the final mining result of the massive network data.

[0144] In the embodiment of the application, when the final mining result acquisition module performs the operation of optimizing the target event association path based on a preset business rule, and obtaining the final mining result of the massive network data, the final mining result acquisition module is specifically configured to: perform rule matching on the target event association path, and obtain an effective path set of the target event association path; perform conflict detection on the target event association path, and obtain a conflict path set of the target event association path; strengthen the core logic chain of the target event association path based on the effective path set, and obtain a core logic path of the target event association path; perform path pruning on the conflict path set, and integrate the path after path pruning and the core logic path into the final mining result.

[0145] The final mining result acquisition module is specifically configured to: compare the event node sequence of the target event association path with an event logic template to obtain a matching degree parameter of the target event association path; extract a corresponding path segment to generate the effective path set when the matching degree parameter meets a preset condition.

[0146] The final mining result acquisition module is specifically configured to: reconstruct the structure of the conflict topology subgraph; position nodes of the conflict topology subgraph to obtain a coordinate identifier of a path divergence point; cut a connection relationship based on the coordinate identifier to obtain a topology structure of a pruned path; topologically integrate the topology structure of the pruned path with the core logic path to obtain the final mining result of the mass network data.

[0147] Specifically, the target event association path is subjected to rule matching to obtain an effective path set of the target event association path.

[0148] Further, a judgment condition for path effectiveness is extracted from a preset business rule, such as a requirement that a node relationship in the path meet a business logic or a requirement that a path length be within a specified range.

[0149] Further, the target event association path is checked piece by piece to see whether it meets the conditions, and all paths meeting the conditions are selected.

[0150] Further, a list containing path identifiers and corresponding node relationships is formed, and the list is the effective path set of the target event association path.

[0151] Specifically, the target event association path is subjected to conflict detection to obtain a conflict path set of the target event association path.

[0152] Further, a node relationship in the target event association path is compared with a prohibitive clause in a preset business rule.

[0153] Further, whether there is a node attribute contradiction, a relationship direction conflict, or the like is checked, and whether there is mutual exclusion between node associations in different paths is compared.

[0154] Further, all the conflicting paths are sorted separately to form records containing conflict types and path details, which collectively constitute a conflict path set of the target event associated path.

[0155] Specifically, the core logic chain of the target event associated path is reinforced based on the effective path set, to obtain a core logic path of the target event associated path.

[0156] Further, the key nodes and necessary relationships common to all paths in the effective path set are extracted, such as the basic relationship chain of "user-browsing-goods" included in multiple effective paths.

[0157] Further, these common parts are used as the basis of the core logic chain, and the extension relationship with the highest frequency of occurrence in the effective path is supplemented.

[0158] Further, by highlighting the connection strength of the key nodes and the clear relationship direction, a main line running through the target event is formed, which is the core logic path of the target event associated path.

[0159] Specifically, path pruning is performed on the conflict path set, and the path after pruning is integrated with the core logic path as the final mining result.

[0160] Further, for each path in the conflict path set, the nodes and relationships that seriously conflict with the preset business rules are deleted, and the non-conflicting segments in the path are retained.

[0161] Further, if "user A-purchase-goods B" in a conflict path is in conflict, but "goods B-belong-to-category C" is not in conflict, the segment is retained.

[0162] Further, the effective segments obtained after pruning are sorted according to the degree of association with the core logic path.

[0163] Further, the corresponding nodes of the core logic path are sequentially attached to form a complete structure containing the core main line and effective branches, which is the final mining result of the massive network data.

[0164] Specifically, the event node sequence of the target event associated path is compared with the event logic template to obtain a matching degree parameter of the target event associated path.

[0165] Further, the standard node types and the order relationship between nodes are extracted from the preset event logic template.

[0166] For example, the template specifies that "the user operation node needs to be located before the goods node".

[0167] Further, the event node sequence in the target event association path is extracted again, and the type of each node in the sequence is compared one by one according to the order of node appearance, whether the type of each node in the sequence is consistent with the type of the node at the corresponding position in the template.

[0168] Further, the connection order between the nodes is checked simultaneously to see whether it meets the template requirements.

[0169] Further, according to the number of completely matched nodes and the degree of order compliance, a matching degree parameter is generated to represent the overall matching situation, which directly reflects the degree of fit between the path and the template.

[0170] Specifically, when the matching degree parameter meets the preset condition, the corresponding path segment is extracted to generate the effective path set.

[0171] Further, the preset condition is that the matching degree parameter reaches a set qualified standard, for example, the proportion of the number of completely matched nodes is not less than a set proportion and the order completely meets the template requirements.

[0172] Further, when the matching degree parameter of the target event association path meets the condition, the complete node sequence and the relationship between the nodes of the path are extracted as a path segment.

[0173] Further, all path segments that meet the condition are summarized together to form a set containing all qualified path segments, which is the effective path set.

[0174] Specifically, the conflict topology subgraph is structurally reconstructed.

[0175] Further, all nodes in the conflict topology subgraph corresponding to the conflict path set are classified according to entity types, such as user nodes, event nodes, and attribute nodes.

[0176] Further, the connection relationship between the nodes is labeled according to the conflict type, such as attribute contradiction relationship and logic opposite relationship, and then rearranged according to the order of node appearance in the path.

[0177] Further, the same type of nodes are concentrated and distributed, the conflict relationship is highlighted with special markers, and a reconstructed conflict topology subgraph with clear structure is formed.

[0178] Specifically, the conflict topology subgraph is positioned to obtain the coordinate identification of the path divergence point.

[0179] Further, the reconstructed conflict topology subgraph is traversed to find all nodes that extend two or more conflicting connection relationships from the same node, which are the path divergence points.

[0180] Further, a unique coordinate identifier is assigned to each divergence point, which is composed of the horizontal position sequence number and the vertical level sequence number of the node in the subgraph.

[0181] Further, the coordinate identifiers of all divergence points are summarized to form a list, i.e., the coordinate identifiers of the path divergence points, as the third horizontal position and the second vertical level of the divergence point are identified as "3-2".

[0182] Specifically, the connection relationship is cut off based on the coordinate identifier, and the topology structure of the pruned path is obtained.

[0183] Further, according to the coordinate identifier of the path divergence point, the specific position of each divergence point in the conflict topology subgraph is found.

[0184] Further, all conflict connection relationships extending from the divergence point are cut off, and only the connection relationships that do not conflict with the preset business rules are retained.

[0185] Further, for the independent path segments formed after cutting off, the nodes are reorganized according to the connection order to ensure that the node relationship in the segment is conflict-free, and a mesh structure containing multiple conflict-free segments is formed, which is the topology structure of the pruned path.

[0186] Specifically, the topology structure of the pruned path is topologically fused with the core logical path to obtain the final mining result of the massive network data.

[0187] Further, each conflict-free segment in the topology structure of the pruned path is matched with a node of the same type in the core logical path, such as matching the "commodity node" in the pruned segment with the "commodity node" in the core logical path.

[0188] Further, the matched segment is connected to the corresponding node of the core logical path, so that the segment exists as a branch of the core logical path, while the main line of the core logical path remains unchanged.

[0189] Further, a complete topology structure containing the core main line and the effective branch is formed, which is the final mining result of the massive network data.

[0190] In summary, the target event associated path is matched with the rules to obtain the effective path set of the target event associated path, which can filter the paths that meet the business logic, provide a reliable basis for mining, and improve the effectiveness of the result.

[0191] In summary, the target event associated path is detected for conflict to obtain the conflict path set of the target event associated path, which can identify invalid paths, reduce interference, and improve the mining accuracy.

[0192] Overall, based on the effective path set, the core logic chain of the target event association path is strengthened, and the core logic path of the target event association path is obtained, which can highlight the key association, clarify the main line, and enhance the result order.

[0193] Overall, the path pruning is performed on the conflict path set, and the path after pruning is integrated with the core logic path as the final mining result, which can integrate effective information, form a complete result, and improve the comprehensiveness of mining.

[0194] Overall, the event node sequence of the target event association path is compared with the event logic template to obtain a matching degree parameter of the target event association path, which can quantify the degree of fit between the path and the business logic, provide a basis for screening effective paths, and improve the compliance of mining.

[0195] Overall, when the matching degree parameter meets the preset condition, the corresponding path segment is extracted to generate the effective path set, which can accurately retain the paths that meet the business rules, ensure the practicality of the mining result, and enhance the reliability of mining.

[0196] Overall, the structure of the conflict topology subgraph is restructured, which can sort out the conflict path structure, clearly present the conflict node relationship, provide a clear object for pruning, and improve the processing efficiency.

[0197] Overall, the node positioning is performed on the conflict topology subgraph to obtain the coordinate identifier of the path divergence point, which can accurately lock the conflict source, provide a clear target for cutting off redundant connections, and improve the pruning accuracy.

[0198] Overall, the connection relationship is cut off based on the coordinate identifier to obtain the topology structure of the pruned path, which can remove invalid conflict connections, simplify the path, and enhance the effectiveness of the path.

[0199] Overall, the topology structure of the pruned path is topologically integrated with the core logic path to obtain the final mining result of the massive network data, which can integrate effective information, form a complete result, and improve the comprehensiveness and logic of mining.

[0200] For those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0201] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Artificial intelligence is the use of digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0202] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application.

Claims

1. A massive network data intelligent mining system based on AI technology, characterized in that, The system includes a standardized dataset acquisition module, a multi-dimensional feature vector acquisition module, an initial knowledge graph construction module, a target event association path construction module, and a final mining result acquisition module, wherein: The standardized dataset acquisition module is used to standardize the original dataset to obtain a standardized dataset of massive network data. The multidimensional feature vector acquisition module is used to extract entity features and relation features from the normalized dataset to obtain multidimensional feature vectors of the massive network data. The initial knowledge graph construction module is used to map the multidimensional feature vectors to the topological structure to obtain the initial knowledge graph of the massive network data; The target event association path construction module is used to perform deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association path of the massive network data. The final mining result acquisition module is used to optimize the target event association path based on preset business rules to obtain the final mining result of the massive network data.

2. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, The normalized dataset acquisition module, when performing normalization processing on the original dataset to obtain a normalized dataset of massive network data, is specifically used for: The numerical data in the original dataset are subjected to feature normalization to obtain the feature weight data of the original dataset. One-hot encoding is performed on the categorical data in the original dataset to obtain the encoded categorical data; The feature weight data and the category data are combined to form a normalized dataset of the massive network data.

3. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, When the multidimensional feature vector acquisition module extracts entity features and relation features from the normalized dataset to obtain the multidimensional feature vector of the massive network data, it is specifically used for: Extract the structured entities from the text fields of the normalized dataset to obtain the entity set of the massive network data; The syntactic dependencies of entity units in the entity set are analyzed to obtain the syntactic relationship features of the text field; The vectorized entity set and the syntactic relation features are topologically reconstructed to obtain the multidimensional feature vector of the massive network data.

4. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, When the initial knowledge graph construction module maps the multidimensional feature vectors to the topological structure to obtain the initial knowledge graph of the massive network data, it is specifically used for: The initial graph structure of the massive network data is constructed based on the multidimensional feature vectors. Based on the influence weights of node connection density and neighborhood similarity in the initial graph structure, the association relationships of the initial graph structure are optimized; The optimized initial graph structure is used to construct a topology structure, resulting in the initial knowledge graph of the massive network data.

5. The AI-based intelligent mining system for massive network data as described in claim 4, characterized in that, When the initial knowledge graph construction module optimizes the association relationships of the initial graph structure based on the influence weights of node connection density and neighborhood similarity in the initial graph structure, it specifically performs the following steps: calculating the influence weights of node connection density and neighborhood similarity based on the initial graph structure to obtain the node influence weights of the initial graph structure. The calculation formula for the node influence weights is as follows: In the formula, For nodes The weight, The damping coefficient is... For nodes The set of adjacent nodes, For nodes The weight, For cosine similarity, For nodes eigenvectors, For nodes eigenvectors, For nodes The degree of departure; The association relationships of the initial graph structure are optimized to obtain the optimized knowledge graph.

6. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, When the target event association path construction module performs deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association paths of the massive network data, it is specifically used for: The node features of the initial knowledge graph are updated; The updated node feature vectors are aggregated to obtain the node state vectors of the massive network data. Based on the node state vector, feature connections are made to the path association relationship to obtain the target event association path of the massive network data.

7. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, When performing deep reasoning on the initial knowledge graph using a graph neural network to obtain the target event association path of the massive network data, the specific application is as follows: Calculate the association strength parameter of the feature vectors of adjacent nodes, wherein the formula for calculating the association strength parameter is as follows: In the formula, Let be the association strength parameter between nodes and . It is a non-linear activation function. For nodes eigenvectors, For nodes eigenvectors, For nodes With nodes The weight of the edges between them. Let be the norm of the vector; The neighboring node feature vectors are weighted and fused based on the association strength parameter to obtain the weighted neighborhood features of the massive network data; The weighted neighborhood features are fused with the central node feature vector to obtain the updated node feature vector. Path mining is performed on the updated node feature vectors to obtain the target event association paths of the massive network data.

8. The AI-based intelligent mining system for massive network data as described in claim 1, characterized in that, When the final mining result acquisition module performs optimization of the target event association path based on preset business rules to obtain the final mining result of the massive network data, it is specifically used for: Perform rule matching on the target event association path to obtain the valid path set of the target event association path; Conflict detection is performed on the target event associated paths to obtain a set of conflicting paths for the target event associated paths; Strengthen the core logical chain of the target event association path based on the set of effective paths to obtain the core logical path of the target event association path. Path pruning is performed on the set of conflicting paths, and the pruned paths are integrated with the core logical paths to form the final mining result.

9. The AI-based intelligent mining system for massive network data as described in claim 8, characterized in that, When the final mining result acquisition module performs rule matching on the target event association path to obtain a set of valid paths that conform to preset business rules, it is specifically used for: The event node sequence of the target event association path is compared with the event logic template to obtain the matching degree parameter of the target event association path; When the matching degree parameter meets the preset conditions, the corresponding path segment is extracted to generate the set of valid paths.

10. The AI-based intelligent mining system for massive network data as described in claim 8, characterized in that, The final mining result acquisition module performs path pruning on the set of conflicting paths and integrates the pruned paths with the core logic paths to form the final mining result. Specifically, it is used for: Restructuring the conflicting topological subgraph; Nodes in the conflict topology subgraph are located to obtain the coordinates of path divergence points; The connection relationships are cut off based on the coordinate identifiers to obtain the topology of the pruned path; The topology of the pruned path is fused with the core logical path to obtain the final mining result of the massive network data.

Images