Entity dependency mining method, device, medium and product based on time sequence graph

Abstract

The present application discloses a method, device, medium, and product for mining entity dependency relationships based on a time series graph, relating to the technical field of entity relationship mining. The method comprises: obtaining a relationship graph at each moment in a time period to be mined and determining a node feature change rate matrix at the current moment and dividing the time period to be mined into multiple sub-periods; updating the initial node encoding vector of the current node in the target sub-period to obtain a target node encoding vector of the current node in the target sub-period; determining a fused node encoding vector of the current node in the target sub-period based on the target node encoding vector of the current node in the target sub-period and the target node encoding vector in the corresponding adjacent sub-period; and determining the Pearson correlation coefficient between any two nodes in the target sub-period based on the fused node encoding vectors of the two nodes in the target sub-period to determine the dependency relationship between the corresponding entities. The present application improves the mining effect of entity dependency relationships.

Description

Technical Field

[0001] The present application relates to the technical field of entity relationship mining, and in particular to a method, device, medium and product for mining entity dependency relationships based on a time sequence graph. Background Art

[0002] Graph models are widely used in complex information modeling due to their powerful ability to represent entities, attributes, and relationships between entities. Time series graphs, a key extension of graph models, are widely used in large knowledge bases in social networks, transportation networks, and financial markets. The entities in these knowledge bases are not static but evolve over time. This requires the ability to accurately and completely capture and update information to support downstream decision-making and fact-checking, thereby improving data consistency and integrity.

[0003] In dynamic environments, changes in entities and relationships need to be captured and updated in real time to maintain data accuracy and reliability. The process of mining dependencies in time series graphs involves identifying and extracting recurring patterns in time series that can reveal the underlying laws of entity behavior. For example, by analyzing time series graphs of user interactions in social networks, we can identify patterns in the formation and dissolution of user groups, which is crucial for social network analysis and community detection. In bioinformatics, analyzing time series graphs of biological networks can reveal how gene expression changes over time, which is of great value for disease research and drug development.

[0004] In order to effectively mine the dependencies of entities in these time series graphs, a variety of algorithms and techniques have been developed. However, existing dependency mining methods often face the following challenges: (1) Inadaptability of time partitioning: Most time series graph dependency mining methods use fixed time windows to process time series data. This approach makes it difficult to adaptively adjust time partitions according to the dynamic characteristics of different nodes or graphs. (2) Limitations of spatial features: Traditional graph convolution methods usually rely on static graph structures and find it difficult to fully capture the temporal dependencies between nodes. (3) Lack of temporal information modeling: Many existing methods fail to effectively integrate long-term temporal dependency information, thereby limiting the mining effect of temporal graph dependencies.

[0005] In summary, the current entity dependency mining effect is poor and cannot be well realized. Summary of the Invention

[0006] The purpose of this application is to provide a method, device, medium and product for mining entity dependency relationships based on a time sequence diagram to solve the problem of poor mining effect of entity dependency relationships.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides an entity dependency mining method based on a time sequence graph, comprising:

[0009] Obtain a relationship graph at each moment in the period to be mined; the relationship graph includes multiple nodes and multiple edges; one node corresponds to one entity, and one edge corresponds to a direct relationship between two entities; the number of nodes and corresponding entities in the relationship graph at each moment are the same;

[0010] Based on the relationship graph at each moment, a node feature matrix at the corresponding moment is determined; the elements in the node feature matrix are the number of interactions between entities corresponding to the two nodes;

[0011] Starting from the second moment, any moment is determined as the current moment, and the node feature change rate matrix of the current moment is determined based on the node feature matrix of the current moment and the node feature matrix of the previous moment;

[0012] The time period to be mined is divided based on the node feature change rate matrix at each moment to obtain multiple sub-periods, and any sub-period in the time period to be mined is determined as a target sub-period, and any node is determined as a current node;

[0013] Based on the relationship graph of each moment in the target sub-period, determine the initial node encoding vector of the current node in the target sub-period and the target time attenuation function value between the current node and each corresponding neighbor node in the target sub-period;

[0014] Based on the target time decay function value between the current node and each corresponding neighbor node in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated to obtain the target node encoding vector of the current node in the target sub-period;

[0015] Determine a fused node coding vector of the current node in the target sub-period based on the target node coding vector of the current node in the target sub-period and the target node coding vector in the corresponding adjacent sub-period;

[0016] Determine the Pearson correlation coefficient between any two nodes in the target sub-period based on the fusion node code vectors of the two nodes in the target sub-period;

[0017] Based on the Pearson correlation coefficient between the two nodes in the target sub-period, a dependency relationship between entities corresponding to the two nodes in the target sub-period is determined; the dependency relationship is dependent or non-dependent.

[0018] Optionally, the time period to be mined is divided based on the node feature change rate matrix at each moment to obtain multiple sub-periods, including:

[0019] The moment when the node feature change rate matrix exceeds the change rate threshold is determined as the time segmentation point;

[0020] The time period to be mined is divided according to all time division points to obtain multiple sub-time periods.

[0021] Optionally, based on the relationship graph at each time in the target sub-period, determining the initial node encoding vector of the current node in the target sub-period and the target time attenuation function value between the current node and each corresponding neighbor node in the target sub-period includes:

[0022] Determine the initial node encoding vector of the current node in the target sub-period based on the interaction between the entity corresponding to the current node and the entities corresponding to other nodes in the relationship graph at all times in the target sub-period;

[0023] Determine the nodes that have edges with the current node as neighbor nodes of the current node, and determine any moment as the first moment;

[0024] Calculate the initial time decay function value between the current node and each corresponding neighbor node at the first moment based on the central moment, first moment, and time decay hyperparameter of the target sub-period;

[0025] The target time decay function value between the current node and each corresponding neighbor node in the target sub-period is obtained by averaging the initial time decay function values ​​at all times in the target sub-period between the current node and each corresponding neighbor node.

[0026] Optionally, based on the target time decay function values ​​between the current node and each corresponding neighbor node in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated to obtain the target node encoding vector of the current node in the target sub-period, including:

[0027] Using multi-layer graph convolution, according to the target time attenuation function value between the current node and the corresponding neighbor nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated by the preset maximum number of convolution layers to obtain the target node encoding vector of the current node in the target sub-period.

[0028] Optionally, determining the fused node coding vector of the current node in the target sub-period based on the target node coding vector of the current node in the target sub-period and the target node coding vector in the corresponding adjacent sub-period includes:

[0029] Determine any sub-period from the second sub-period to the second-to-last sub-period in the period to be mined as the current sub-period;

[0030] Determine a fused node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period, the target node coding vector in the previous sub-period, and the target node coding vector in the next sub-period;

[0031] Determine a fusion node coding vector of the current node in the first sub-period based on the target node coding vector of the current node in the first sub-period and the target node coding vector in the second sub-period;

[0032] Based on the target node coding vector of the current node in the last sub-period and the target node coding vector in the second-to-last sub-period, the fusion node coding vector of the current node in the last sub-period is determined.

[0033] Optionally, determining the fused node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period, the target node coding vector in the previous sub-period, and the target node coding vector in the next sub-period includes:

[0034] Determine the forward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the previous sub-period;

[0035] Determine a backward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the next sub-period;

[0036] Determine the fusion node coding vector of the current node in the current sub-period according to the forward node coding vector and the backward node coding vector of the current node in the current sub-period.

[0037] Optionally, determining the dependency relationship between entities corresponding to the two nodes in the target sub-period based on the Pearson correlation coefficient between the two nodes in the target sub-period includes:

[0038] Determine whether the Pearson correlation coefficient between two nodes in the target sub-period exceeds a preset threshold;

[0039] If so, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined as dependency;

[0040] If not, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined to be non-dependence.

[0041] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned entity dependency mining methods based on time sequence diagrams.

[0042] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned entity dependency mining methods based on time sequence diagrams.

[0043] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned entity dependency mining methods based on timing diagrams.

[0044] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0045] The present application discloses a method, device, medium and product for mining entity dependency relationships based on a time series graph. By dynamically dividing the mining period according to the node feature change rate matrix, a flexible time partitioning strategy is implemented to cope with the changes in time series data at different time scales, overcome the limitations of the fixed time window method, and better capture the dynamic changes in the time series graph. Based on the target time attenuation function value between the current node and the corresponding neighboring nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated, and the local time series dependency relationship is mined to obtain the target node encoding vector of the current node in the target sub-period, which helps to accurately identify the microscopic dynamic relationship between nodes and edges in the graph. Based on the target node encoding vector of the current node in the target sub-period and the target node encoding vector in the corresponding adjacent sub-period, the fused node encoding vector of the current node in the target sub-period is determined, and the time series dependency is modeled over a long time span, capturing the macroscopic laws of node and edge attributes changing over time. The mining effect of entity dependency relationships is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 A flow chart of a method for mining entity dependency relationships based on a time sequence diagram according to an embodiment of the present application;

[0048] Figure 2 It is a schematic diagram of the relationship diagram structure;

[0049] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0051] The purpose of this application is to provide a method, device, medium and product for entity dependency mining based on a time sequence diagram, aiming to improve the mining effect of entity dependency.

[0052] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0053] In an exemplary embodiment, Figure 1 As shown, a method for mining entity dependency relationships based on a time sequence diagram is provided, including:

[0054] Step 1: Obtain the relationship graph at each moment in the period to be mined; the relationship graph includes multiple nodes and multiple edges; one node corresponds to one entity, and one edge corresponds to the direct relationship between two entities; the number of nodes and corresponding entities in the relationship graph at each moment are the same.

[0055] Specifically, the relationship diagram at a certain moment is as follows Figure 2 As shown, Figure 2 In the figure, A, B, C, and D are nodes corresponding to the four entities respectively. The edge between A and B represents the direct relationship between the entities corresponding to A and B, the edge between A and D represents the direct relationship between the entities corresponding to A and D, and the edge between B and C represents the direct relationship between the entities corresponding to B and C.

[0056] In practical applications, the relationship graph of each moment in the period to be mined is determined based on the ICEWS data, a public dataset containing N events.

[0057] First, the event set in the public dataset ICEWS is defined as D = {(t n ,s n ,r n ,d n ,o n )|n=1,2,…,N}, where t n is the time of the nth event, s n is the initiating entity of the nth event, r n is the direct relationship between entities of the nth event (CAMEO code), d n is the target entity of the nth event, on is the text description of the nth event, and N is the total number of events in the dataset.

[0058] Then, by processing the public dataset ICEWS data, the temporal event quadruple ε is extracted, as shown in formula (1), and redundant fields are removed to ensure the structured expression of the data.

[0059] ε={(s n ,r n ,t n ,d n )|n=1,2,…,N}(1)

[0060] Secondly, for the extracted time series event quadruple, it is necessary to deduplicate, repair missing data, and normalize entities to improve data quality and enable better data processing.

[0061] Finally, the cleaned time series event quadruple is expressed as a timestamp-equipped graph structure, i.e., the relationship graph of each moment in the mining period. The relationship graph of each moment in the mining period constitutes the time series graph set G, which is defined as follows: G = (V, E, R, T). Where V is the set of nodes in the time series graph set; E is the set of edges in the time series graph set; R is the set of direct relationships between entities; and T is the set of moments.

[0062] Step 2: Based on the relationship graph at each moment, determine the node feature matrix at the corresponding moment; the elements in the node feature matrix are the number of interactions between the entities corresponding to the two nodes.

[0063] Specifically, when the relationship diagram is Figure 2 When , the node feature matrix Z is expressed as:

[0064]

[0065] Among them, each element in Z represents the number of interactions between the entities corresponding to the two letters in the subscript.

[0066] Step 3: Starting from the second moment, any moment is determined as the current moment, and the node feature change rate matrix of the current moment is determined based on the node feature matrix of the current moment and the node feature matrix of the previous moment.

[0067] Specifically, the calculation formula of the node feature change rate matrix is:

[0068] ΔX(T i )=X(T i )-X(T i-1 )(3)

[0069] Among them, ΔX(T i ) is the i-th moment T iThe node feature change rate matrix; X(T i ) is the i-th moment T i Node feature matrix; X(T i-1 ) is the i-1th moment T i-1 The node feature matrix of .

[0070] Step 4: Divide the mining period based on the node feature change rate matrix at each moment to obtain multiple sub-periods, and determine any sub-period in the mining period as the target sub-period, and determine any node as the current node.

[0071] As an optional implementation, step 4 includes:

[0072] Step 41: Determine the time when the node feature change rate matrix exceeds the change rate threshold as the time segmentation point.

[0073] Step 42: Segment the mining period according to all time segmentation points to obtain multiple sub-periods.

[0074] Specifically, according to all time segmentation points M1, M2, ..., M k , the mining period is divided into multiple sub-periods, each sub-period corresponds to a sub-graph set, so the relationship graphs g1, g2, ..., g of k-1 sub-periods are obtained. k-1 Among them, M1 is the first time segmentation point, that is, the initial moment of the mining period; M2 is the second time segmentation point; M k is the kth time segmentation point, i.e., the end of the mining period; g1 is the relationship diagram of the first sub-period; g2 is the relationship diagram of the second sub-period; g k-1 is the relationship diagram of the k-1th sub-period. The bth sub-period is represented by [M b ,M b+1 ), M b is the bth time division point; M b+1 is the b+1th time segmentation point; the relationship graph set of the bth sub-period is represented by g b .

[0075] Step 5: Based on the relationship diagram of each moment in the target sub-period, determine the initial node encoding vector of the current node in the target sub-period and the target time attenuation function value between the current node and each corresponding neighbor node in the target sub-period.

[0076] As an optional implementation, step 5 includes:

[0077] Step 51: Determine the initial node encoding vector of the current node in the target sub-period based on the interaction between the entity corresponding to the current node and the entities corresponding to other nodes in the relationship graph at all times in the target sub-period.

[0078] Specifically, when the relationship graph set of the b-th sub-period contains three Figure 2 The relationship graph shown in Figure 2 is the same, and the edges between the nodes remain unchanged; only the type of direct relationship represented by the edges changes. In the relationship graph at the first moment in the b-th subperiod, the direct relationship between entities A and B is 01000, the direct relationship between entities A and D is 00100, and the direct relationship between entities B and C is 10000. In the relationship graph at the second moment in the b-th subperiod, the direct relationship between entities A and B is 01000, the direct relationship between entities A and D is 00100, and the direct relationship between entities B and C is 10000. In the relationship graph at the third moment in the b-th subperiod, the direct relationship between entities A and B is 00001, the direct relationship between entities A and D is 00100, and the direct relationship between entities B and C is 00010. At this time, the initial node code vector of node A is (00100, [01000+00001], 0) = (00100, 01001, 0). The initial node code vector of node B is (0, [01000+00001], [10000+00010]) = (0, 01001, 10010). The initial node code vector of node C is (0, 0, [10000+00010]) = (0, 0, 10010). The initial node code vector of node D is (00100, 0, 0).

[0079] Step 52: Determine the nodes that have edges with the current node as neighbor nodes of the current node, and determine any moment as the first moment.

[0080] Step 53: Calculate the initial time decay function value between the current node and each corresponding neighbor node at the first moment according to the central moment of the target sub-period, the first moment and the time decay hyperparameter.

[0081] Specifically, the calculation formula for the initial time decay function value is:

[0082]

[0083] in, For node v c and neighbor node v d The initial time decay function value at the mth moment in the bth sub-period; is the central moment of the b-th sub-period; is the mth moment in the bth sub-period; τ is the time decay hyperparameter, which controls the influence of time.

[0084] Step 54: averaging the initial time decay function values ​​between the current node and the corresponding neighbor nodes at all times in the target sub-period to obtain the target time decay function values ​​between the current node and the corresponding neighbor nodes in the target sub-period.

[0085] Step 6: Based on the target time attenuation function value between the current node and the corresponding neighbor nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated to obtain the target node encoding vector of the current node in the target sub-period.

[0086] As an optional implementation, step 6 includes:

[0087] Step 61: Using multi-layer graph convolution, according to the target time attenuation function value between the current node and the corresponding neighbor nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated by the preset maximum number of convolution layers to obtain the target node encoding vector of the current node in the target sub-period.

[0088] Specifically, the update formula for multi-layer graph convolution is:

[0089]

[0090] in, For node v c The node encoding vector after the l+1th layer of graph convolution in the bth sub-period; σ(·) is the activation function; W (l) is the learnable transformation weight matrix of the l-th layer graph convolution; For node v c The node encoding vector after the l-th layer of graph convolution in the b-th sub-period; N(v c ) is the node v c The set of all neighbor nodes of For node v c and neighbor node v d The target time decay function value in the b-th sub-period; For node v c Neighbor node v d The node encoding vector after the l-th layer of graph convolution in the b-th sub-period.

[0091] Step 7: Based on the target node coding vector of the current node in the target sub-period and the target node coding vector in the corresponding adjacent sub-period, determine the fused node coding vector of the current node in the target sub-period.

[0092] As an optional implementation, step 7 includes:

[0093] Step 71: Determine any sub-period from the second sub-period to the second-to-last sub-period in the waiting mining period as the current sub-period.

[0094] Step 72: Determine the fusion node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period, the target node coding vector in the previous sub-period, and the target node coding vector in the next sub-period.

[0095] As an optional implementation, step 72 includes:

[0096] Step 721: Determine the forward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the previous sub-period.

[0097] Specifically, node v y The calculation formula of the forward node encoding vector in the bth sub-period is:

[0098]

[0099] in, For node v y The forward node encoding vector in the b-th sub-period; LSTM(·) is a long short-term memory network; For node v y The target node encoding vector in the b-1th sub-period, L is the preset maximum number of convolution layers; For node v y The target node encoding vector in the bth sub-period.

[0100] Step 722: Determine the backward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the next sub-period.

[0101] Specifically, node v y The calculation formula of the backward node encoding vector in the bth sub-period is:

[0102]

[0103] in, For node v y Backward node encoding vector at the bth sub-period; For node v y The target node encoding vector in the b+1th sub-period.

[0104] Step 723: Determine the fused node coding vector of the current node in the current sub-period based on the forward node coding vector and the backward node coding vector of the current node in the current sub-period.

[0105] Specifically, node v y The calculation formula of the fusion node coding vector in the bth sub-period is:

[0106]

[0107] Among them, y,b For node v y The fusion node encoding vector in the b-th sub-period; concat(·) is the concatenation operation.

[0108] Step 73: Determine the fusion node coding vector of the current node in the first sub-period based on the target node coding vector of the current node in the first sub-period and the target node coding vector in the second sub-period.

[0109] Specifically, node v y The calculation formula of the fusion node coding vector in the first sub-period is:

[0110]

[0111] Among them, y,1 For node v y The fusion node encoding vector in the first sub-period; For node v y The target node encoding vector in the first sub-period; For node v y The target node encoding vector in the second sub-period.

[0112] Step 74: Determine the fusion node coding vector of the current node in the last sub-period based on the target node coding vector of the current node in the last sub-period and the target node coding vector in the second to last sub-period.

[0113] Specifically, node v y The calculation formula of the fusion node coding vector in the last sub-period is:

[0114]

[0115] Among them, y,k-1 For node v y The fusion node encoding vector in the last sub-period (i.e., the k-1th sub-period); For node v y The target node encoding vector in the second-to-last sub-period (i.e., the k-2th sub-period); For node v y The target node encoding vector in the last sub-period (i.e., the k-1th sub-period).

[0116] Step 8: Based on the fused node code vectors of any two nodes in the target sub-period, determine the Pearson correlation coefficient between the two nodes in the target sub-period.

[0117] Specifically, the calculation formula of the Pearson correlation coefficient is:

[0118]

[0119] in, For node v x and v y The Pearson correlation coefficient between the two in the bth sub-period; con(·) is the covariance; o x,b For node v x The fusion node encoding vector in the b-th sub-period; ζ(·) is the standard deviation.

[0120] Step 9: Based on the Pearson correlation coefficient between the two nodes in the target sub-period, determine the dependency relationship between the entities corresponding to the two nodes in the target sub-period; the dependency relationship is dependent or non-dependent.

[0121] As an optional implementation, step 9 includes:

[0122] Step 91: Determine whether the Pearson correlation coefficient between two nodes in the target sub-period exceeds a preset threshold.

[0123] Step 92: If yes, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined as dependency.

[0124] Step 93: If not, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined to be non-dependent.

[0125] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement an entity dependency mining method based on a time sequence graph.

[0126] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for mining entity dependency relationships based on a time sequence graph is implemented.

[0127] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements an entity dependency mining method based on a time sequence graph when executed by a processor.

[0128] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for mining entity dependencies based on a timing diagram is implemented.

[0129] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0130] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0131] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0132] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0133] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for mining entity dependency relationships based on time series graphs, characterized in that: The entity dependency mining method based on the time sequence diagram includes: Obtain a relationship graph at each moment in the period to be mined; the relationship graph includes multiple nodes and multiple edges; one node corresponds to one entity, and one edge corresponds to a direct relationship between two entities; the number of nodes and corresponding entities in the relationship graph at each moment are the same; Based on the relationship graph at each moment, a node feature matrix at the corresponding moment is determined; the elements in the node feature matrix are the number of interactions between entities corresponding to the two nodes; Starting from the second moment, any moment is determined as the current moment, and the node feature change rate matrix of the current moment is determined based on the node feature matrix of the current moment and the node feature matrix of the previous moment; The time period to be mined is divided based on the node feature change rate matrix at each moment to obtain multiple sub-periods, and any sub-period in the time period to be mined is determined as a target sub-period, and any node is determined as a current node; Based on the relationship graph of each moment in the target sub-period, determine the initial node encoding vector of the current node in the target sub-period and the target time attenuation function value between the current node and each corresponding neighbor node in the target sub-period; Based on the target time decay function value between the current node and each corresponding neighbor node in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated to obtain the target node encoding vector of the current node in the target sub-period; Determine a fused node coding vector of the current node in the target sub-period based on the target node coding vector of the current node in the target sub-period and the target node coding vector in the corresponding adjacent sub-period; Determine the Pearson correlation coefficient between any two nodes in the target sub-period based on the fusion node code vectors of the two nodes in the target sub-period; Based on the Pearson correlation coefficient between the two nodes in the target sub-period, a dependency relationship between entities corresponding to the two nodes in the target sub-period is determined; the dependency relationship is dependent or non-dependent.

2. The entity dependency mining method based on time sequence graph according to claim 1 is characterized in that: The time period to be mined is divided based on the node feature change rate matrix at each moment to obtain multiple sub-periods, including: The moment when the node feature change rate matrix exceeds the change rate threshold is determined as the time segmentation point; The time period to be mined is divided according to all time division points to obtain multiple sub-time periods.

3. The entity dependency mining method based on time sequence diagram according to claim 1 is characterized in that: Based on the relationship graph of each time in the target sub-period, the initial node encoding vector of the current node in the target sub-period and the target time attenuation function value between the current node and each corresponding neighbor node in the target sub-period are determined, including: Determine the initial node encoding vector of the current node in the target sub-period based on the interaction between the entity corresponding to the current node and the entities corresponding to other nodes in the relationship graph at all times in the target sub-period; Determine the nodes that have edges with the current node as neighbor nodes of the current node, and determine any moment as the first moment; Calculate the initial time decay function value between the current node and each corresponding neighbor node at the first moment based on the central moment, first moment, and time decay hyperparameter of the target sub-period; The target time decay function value between the current node and each corresponding neighbor node in the target sub-period is obtained by averaging the initial time decay function values ​​at all times in the target sub-period between the current node and each corresponding neighbor node.

4. The entity dependency mining method based on time sequence graph according to claim 1 is characterized in that: Based on the target time decay function values ​​between the current node and the corresponding neighbor nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated to obtain the target node encoding vector of the current node in the target sub-period, including: Using multi-layer graph convolution, according to the target time attenuation function value between the current node and the corresponding neighbor nodes in the target sub-period, the initial node encoding vector of the current node in the target sub-period is updated by the preset maximum number of convolution layers to obtain the target node encoding vector of the current node in the target sub-period.

5. The entity dependency mining method based on time sequence graph according to claim 1 is characterized in that: Determining a fused node coding vector of the current node in the target sub-period based on the target node coding vector of the current node in the target sub-period and the target node coding vector in the corresponding adjacent sub-period includes: Determine any sub-period from the second sub-period to the second-to-last sub-period in the period to be mined as the current sub-period; Determine a fused node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period, the target node coding vector in the previous sub-period, and the target node coding vector in the next sub-period; Determine a fusion node coding vector of the current node in the first sub-period based on the target node coding vector of the current node in the first sub-period and the target node coding vector in the second sub-period; Based on the target node coding vector of the current node in the last sub-period and the target node coding vector in the second-to-last sub-period, the fusion node coding vector of the current node in the last sub-period is determined.

6. The entity dependency mining method based on time sequence diagram according to claim 5 is characterized in that: Determining a fused node coding vector of the current node in the current sub-period based on a target node coding vector of the current node in the current sub-period, a target node coding vector in the previous sub-period, and a target node coding vector in the next sub-period includes: Determine the forward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the previous sub-period; Determine a backward node coding vector of the current node in the current sub-period based on the target node coding vector of the current node in the current sub-period and the target node coding vector in the next sub-period; Determine the fusion node coding vector of the current node in the current sub-period according to the forward node coding vector and the backward node coding vector of the current node in the current sub-period.

7. The entity dependency mining method based on time sequence graph according to claim 1, characterized in that: Based on the Pearson correlation coefficient between the two nodes in the target sub-period, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined, including: Determine whether the Pearson correlation coefficient between two nodes in the target sub-period exceeds a preset threshold; If so, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined as dependency; If not, the dependency relationship between the entities corresponding to the two nodes in the target sub-period is determined to be non-dependence.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the entity dependency mining method based on a time sequence diagram as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the entity dependency mining method based on the timing graph described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the entity dependency mining method based on the timing graph described in any one of claims 1 to 7 is implemented.

Images