Link prediction method and device based on dynamic measurement and medium
By adopting a dynamic metric method in event risk level link prediction, the event map is constructed and graph neural network calculation is performed, the problem of inaccurate prediction of Hard Sample sample risk level is solved, and the accuracy and efficiency of prediction are improved.
Patent Information
- Application Number
- CN202411626465.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-14
- Publication Date
- 2025-02-18
AI Technical Summary
In the event risk level link prediction, the risk level prediction of the Hard Sample sample is inaccurate, resulting in unreasonable resource allocation and inappropriate response measures.
Using a link prediction method based on dynamic metrics, the event map is constructed, the graph neural network is calculated, the positive and negative sample vectorized data is determined, the mean processing and dynamic metric analysis is performed, the dynamic metric threshold is obtained, and the measurement loss is calculated to update the graph. Neural network weights.
The efficiency of positive and negative samples distinguishing and the accuracy of link prediction are improved, and the problem of inaccurate prediction of Hard Sample sample risk level is solved.
Smart Images
Figure CN119476953A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a link prediction method, device and medium based on dynamic metrics. Background Art
[0002] Event risk level link prediction can predict the risk level (such as four levels of risk: low risk, medium risk, high risk, and crisis risk) of various social events that have occurred using graph link prediction analysis methods and techniques.
[0003] Due to the diversity of descriptions of social events, covering different fields, scenarios and characteristics, the learning problem of Hard Sample (difficult to learn samples) in the event risk level link prediction process is prone to inaccuracies, which in turn has an adverse impact on subsequent decision-making and judgment, and may lead to problems such as unreasonable resource allocation and inappropriate response measures. Summary of the invention
[0004] The embodiments of the present application provide a link prediction method, device and medium based on dynamic metrics, which solve the technical problem of inaccurate prediction of the risk level of Hard Samples (difficult to learn samples).
[0005] In a first aspect, an embodiment of the present application provides a link prediction method based on dynamic metrics, characterized in that the method includes: based on a preset event construction requirement, obtaining link prediction training samples by constructing an event graph analysis; performing graph neural network inference on the link prediction training samples to determine positive and negative sample vectorized data; performing mean processing on the positive and negative sample vectorized data to obtain a sample similarity set mean, and performing dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; performing metric loss inference on the dynamic metric threshold to obtain graph neural network weight update data; based on the graph neural network weight update data, determining the updated graph data through similar node links; and determining the event risk level similarity set through graph neural network analysis based on the updated graph data.
[0006] In one implementation of the present application, based on a preset event construction requirement, link prediction training samples are obtained by constructing an event graph analysis, specifically including: constructing an event risk level for the event construction requirement to obtain an event risk level graph; based on the event risk level graph, determining an event similarity graph through event similarity analysis; according to the event risk level graph and the event similarity graph, obtaining link prediction training samples through positive and negative sample setting.
[0007] In one implementation of the present application, the positive and negative sample vectorized data are averaged to obtain the sample similarity set mean, specifically including: performing node similarity analysis on the positive and negative sample vectorized data to determine the similarity data between the positive and negative sample nodes; based on the similarity data between the positive and negative sample nodes, obtaining a sample similarity set through sample integration; wherein the sample similarity set includes: a positive sample similarity set and a negative sample similarity set; performing similarity averaging processing on the sample similarity set to obtain the sample similarity set mean.
[0008] In one implementation of the present application, a dynamic measurement analysis is performed on the sample similarity set mean to obtain a dynamic measurement threshold, specifically including: performing a positive sample lower limit analysis on the sample similarity set mean to determine the positive sample lower limit threshold; performing a negative sample upper limit analysis on the sample similarity set mean to determine the negative sample upper limit threshold; and obtaining a dynamic measurement threshold based on the positive sample lower limit threshold and the negative sample lower limit threshold.
[0009] In one implementation of the present application, a dynamic metric threshold is subjected to a metric loss calculation to obtain graph neural network weight update data, specifically including: based on the dynamic metric threshold, a similarity set is set to zero to obtain an updated similarity set mean; a metric loss calculation is performed on the updated similarity set mean to determine the metric loss data; wherein the metric loss data is represented by the following formula: loss = 1-s p_ave +s n_ave ; where s p_ave is the mean of the positive sample similarity set, s n_ave is the mean of the negative sample similarity set, and loss is the metric loss data. Based on the metric loss data, the graph neural network weight update data is obtained through the gradient descent method.
[0010] In one implementation of the present application, based on the graph neural network weight update data, the updated graph data is determined through similar node links, specifically including: based on the graph neural network weight update data, the graph neural network model is determined; according to the graph neural network model, through the event prediction stage analysis, the event semantic embedding vector is determined; the event semantic embedding vector is connected with graph similar event nodes to determine the updated graph data.
[0011] In one implementation of the present application, based on the updated graph data, an event risk level similarity set is determined through graph neural network analysis, specifically including: inputting the updated graph data into the graph neural network model to obtain a final vector of the event risk level; performing similarity set processing on the final vector of the event risk level, and sorting the similarity sets obtained by the similarity set processing to obtain a similarity set sequence; setting the risk level with the highest similarity in the similarity set sequence as the final result to determine the event risk level similarity set.
[0012] In one implementation of the present application, after determining the event risk level similarity set based on the updated graph data through the input of the graph neural network, the method also includes: determining the risk response strategy through risk estimation based on the event risk level similarity set; configuring monitoring nodes for the risk response strategy to obtain risk response processing data; and determining the risk plan improvement data through management plan analysis based on the risk response processing data.
[0013] In the second aspect, an embodiment of the present application also provides a link prediction device based on dynamic metrics, characterized in that the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: based on a preset event construction requirement, obtain link prediction training samples by constructing an event graph analysis; perform graph neural network inference on the link prediction training samples to determine the positive and negative sample vectorized data; perform mean processing on the positive and negative sample vectorized data to obtain the sample similarity set mean, and perform dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; perform metric loss inference on the dynamic metric threshold to obtain graph neural network weight update data; based on the graph neural network weight update data, determine the updated graph data through similar node links; according to the updated graph data, determine the event risk level similarity set through graph neural network analysis.
[0014] In a third aspect, an embodiment of the present application also provides a non-volatile computer storage medium for link prediction based on dynamic metrics, storing computer executable instructions, characterized in that the computer executable instructions are set to: based on a preset event construction requirement, obtain link prediction training samples by constructing an event graph analysis; perform graph neural network inference on the link prediction training samples to determine the positive and negative sample vectorized data; perform averaging on the positive and negative sample vectorized data to obtain the sample similarity set mean, and perform dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; perform metric loss inference on the dynamic metric threshold to obtain graph neural network weight update data; based on the graph neural network weight update data, determine the updated graph data through similar node links; and determine the event risk level similarity set through graph neural network analysis based on the updated graph data.
[0015] The embodiments of the present application provide a link prediction method, device and medium based on dynamic measurement, which solves the technical problem of inaccurate prediction of the risk level of Hard Sample by analyzing the dynamic measurement threshold and measurement loss of Hard Sample (difficult to learn sample) of event risk level link prediction, and improves the efficiency of distinguishing between positive and negative samples and the accuracy of link prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0017] Figure 1 A flow chart of a link prediction method based on dynamic metrics provided in an embodiment of the present application;
[0018] Figure 2 A schematic diagram of the internal structure of a link prediction device based on dynamic metrics provided in an embodiment of the present application. DETAILED DESCRIPTION
[0019] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.
[0020] The embodiments of the present application provide a link prediction method, device and medium based on dynamic measurement, which solves the technical problem of inaccurate prediction of the risk level of Hard Sample by analyzing the dynamic measurement threshold and measurement loss of Hard Sample (difficult to learn sample) of event risk level link prediction, and improves the efficiency of distinguishing between positive and negative samples and the accuracy of link prediction.
[0021] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.
[0022] Figure 1 A flow chart of a link prediction method based on dynamic metrics provided in an embodiment of the present application. Figure 1 As shown, a link prediction method based on dynamic metrics provided by an embodiment of the present application specifically includes the following steps:
[0023] Step 101: Based on the preset event construction requirements, link prediction training samples are obtained by constructing event graph analysis.
[0024] Specifically, it includes: constructing event risk levels for event requirements to obtain an event risk level map; based on the event risk level map, determining the event similarity map through event similarity analysis; and obtaining link prediction training samples through positive and negative sample settings based on the event risk level map and the event similarity map.
[0025] This application is based on the preset construction event requirements. By constructing event graph analysis, link prediction training samples are obtained, and the construction and graph analysis of what needs to be constructed are realized.
[0026] In the examples of the present application, detailed explanation is given by the following Example 1.
[0027] Example 1: First, construct an event-risk level map, which contains specific events and corresponding risk level descriptions. Use the Bert semantic embedding method to vectorize the event and risk level descriptions.
[0028] Then, construct an event-event graph. Based on the event vectorization description, construct an event similarity matrix. According to the obtained similarity matrix, calculate the N most similar events for each event; in general, 5 <N<20。
[0029] Finally, based on the above event-risk level graph and event-event graph, events and risk levels are vectorized to construct graph-oriented link prediction training samples.
[0030] Among them, events-events and event-risk levels with node associations constitute positive samples, and events-events and event-risk levels without node associations constitute negative samples.
[0031] Step 102: Perform graph neural network inference on the link prediction training samples to determine the vectorized data of positive and negative samples.
[0032] This application implements vectorized processing of positive and negative samples by performing graph neural network inference on link prediction training samples to determine vectorized data of positive and negative samples.
[0033] In practical applications, the GraphSAGE graph neural network is used to perform forward inference calculations on training samples to obtain new vectorized representations of positive and negative samples.
[0034] Step 103: perform mean processing on the positive and negative sample vectorized data to obtain a sample similarity set mean, and perform dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold.
[0035] The positive and negative sample vectorized data are averaged to obtain the sample similarity set mean, specifically including: performing node similarity analysis on the positive and negative sample vectorized data to determine the similarity data between the positive and negative sample nodes; based on the similarity data between the positive and negative sample nodes, obtaining a sample similarity set through sample integration; wherein the sample similarity set includes: a positive sample similarity set and a negative sample similarity set; performing similarity averaging on the sample similarity set to obtain the sample similarity set mean.
[0036] A dynamic measurement analysis is performed on the sample similarity set mean to obtain a dynamic measurement threshold, specifically including: performing a positive sample lower limit analysis on the sample similarity set mean to determine the positive sample lower limit threshold; performing a negative sample upper limit analysis on the sample similarity set mean to determine the negative sample upper limit threshold; and obtaining a dynamic measurement threshold based on the positive sample lower limit threshold and the negative sample lower limit threshold.
[0037] The present application performs averaging processing on the vectorized data of positive and negative samples to obtain the sample similarity set mean, and performs dynamic measurement analysis on the sample similarity set mean to obtain a dynamic measurement threshold, thereby realizing dynamic analysis of the measurement threshold and improving the efficiency of distinguishing between positive and negative samples and the accuracy of link prediction.
[0038] In the examples of the present application, detailed explanation is given by the following Example 2.
[0039] Example 2: Based on the new vectorized representation, calculate the similarity between each positive sample and negative sample node, and calculate the average similarity of all positive samples and all negative samples.
[0040] The similarity data calculation method between the positive and negative sample nodes of each positive or negative sample is expressed by the following formula:
[0041] s i =v i1 *v i2 (1)
[0042] The average similarity calculation method for all positive samples or all negative samples is expressed by the following formula:
[0043] s ave =sum(s i ) / len(s i ) (2)
[0044] Among them, i is the i-th sample or node relationship;
[0045] v i1 and v i2 are two normalized vectors of the i-th sample respectively;
[0046] save is the mean of sample similarity set;
[0047] sum is the sum of the positive sample similarity set or the negative sample similarity set;
[0048] len is the number of positive or negative samples.
[0049] The mean s of the positive sample similarity set calculated based on the above steps p_ave and the mean of the similarity set of negative samples s n_ave .
[0050] Calculate the current dynamic measurement threshold t p and t n , the calculation method is expressed by the following formula:
[0051] t p =max(s p_ave ,s n_ave ) (3)
[0052] t n =min(s p_ave ,s n_ave ) (4)
[0053] Among them, t p is the lower threshold of positive samples (i.e., the similarity of all positive samples should be greater than this threshold);
[0054] t n is the upper threshold of negative samples (i.e., the similarity of all negative samples should be less than this threshold).
[0055] Step 104: Calculate the metric loss of the dynamic metric threshold to obtain graph neural network weight update data.
[0056] Specifically, it includes: based on the dynamic measurement threshold, by setting the similarity set to zero, obtaining the updated similarity set mean; performing measurement loss calculation on the updated similarity set mean to determine the measurement loss data; wherein the measurement loss data is represented by the following formula: loss = 1-s p_ave +s n_ave ; where s p_ave is the mean of the positive sample similarity set, s n_ave is the mean of the negative sample similarity set, and loss is the metric loss data. Based on the metric loss data, the graph neural network weight update data is obtained through the gradient descent method.
[0057] This application calculates the metric loss of the dynamic metric threshold to obtain the graph neural network weight update data, realizes the dynamic change calculation of the graph neural network weight, improves the efficiency of distinguishing positive and negative samples and the accuracy of link prediction.
[0058] In the examples of the present application, this is explained in detail through the following Example 3.
[0059] Example 3: Set all similarities in the positive sample similarity set greater than t p Set the value of zero and set all the similarities in the negative sample similarity set that are less than t n The value of is set to zero.
[0060] Recalculate the mean s of the positive sample similarity set respectively p_ave , the mean of the negative sample similarity set s n_ave .
[0061] Calculate the metric loss loss of the forward reasoning calculation and update the weights of the GraphSΔGE graph neural network according to the conventional gradient descent method.
[0062] Among them, the metric loss loss calculation method is expressed by the following formula:
[0063] loss=1-s p_ave +s n_ave (5)
[0064] Through the above method, both the measurement loss data and the dynamic measurement threshold can be updated in the link prediction according to the period situation. Through this dynamic measurement processing method, the final risk prediction can be improved by about 5% compared with the existing risk prediction accuracy.
[0065] Step 105: Update the data based on the graph neural network weights and determine the updated graph data through similar node links.
[0066] Specifically, it includes: updating data based on the graph neural network weights to determine the graph neural network model; determining the event semantic embedding vector through event prediction phase analysis based on the graph neural network model; and connecting similar event nodes on the graph of the event semantic embedding vector to determine the updated graph data.
[0067] In the examples of the present application, a detailed explanation is given by the following Example 4.
[0068] Example 4: Based on the above steps, the GraphSAGE graph neural network model is trained. In the prediction stage, for the event to be processed, its Bert semantic embedding vector is first obtained. According to the same method of event-event graph construction, the N event nodes in the graph that are most similar to the event are linked to embed the event graph and obtain the new graph G. new .
[0069] Step 106: Based on the updated graph data, determine the event risk level similarity set through graph neural network analysis.
[0070] Specifically, it includes: inputting the updated graph data into the graph neural network model to obtain the final vector of event risk level; performing similarity set processing on the final vector of event risk level, and sorting the similarity sets obtained by the similarity set processing to obtain a similarity set sequence; setting the risk level with the highest similarity in the similarity set sequence as the final result to determine the event risk level similarity set.
[0071] After determining the event risk level similarity set based on the updated graph data through the input of the graph neural network, the method also includes: determining the risk response strategy through risk estimation based on the event risk level similarity set; configuring monitoring nodes for the risk response strategy to obtain risk response processing data; and determining the risk plan improvement data through management plan analysis based on the risk response processing data.
[0072] In the examples of the present application, this is explained in detail through the following Example 5.
[0073] Example 5: Transform the graph G new As the input of GraphSAGE graph neural network, the final vector representation of event and risk level set is obtained. According to the similarity calculation method mentioned above, the similarity set of event and risk level set is calculated, and the similarity set is sorted from high to low, and the risk level with the highest similarity is taken as the final result.
[0074] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, the embodiment of this application also provides a link prediction device based on dynamic metrics, and its structure is as follows: Figure 2 shown.
[0075] Figure 2 A schematic diagram of the internal structure of a link prediction device based on dynamic metrics provided in an embodiment of the present application. Figure 2 As shown, the device includes:
[0076] at least one processor 201;
[0077] and, a memory 202 communicatively connected to the at least one processor;
[0078] The memory 202 stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor 201 to enable at least one processor 201 to:
[0079] Based on the preset event construction requirements, link prediction training samples are obtained by constructing event graph analysis; graph neural network inference is performed on the link prediction training samples to determine the positive and negative sample vectorized data; the positive and negative sample vectorized data are averaged to obtain the sample similarity set mean, and the sample similarity set mean is dynamically measured and analyzed to obtain the dynamic measurement threshold; the dynamic measurement threshold is measured and loss is calculated to obtain the graph neural network weight update data; based on the graph neural network weight update data, the updated graph data is determined by similar node links; based on the updated graph data, the event risk level similarity set is determined through graph neural network analysis.
[0080] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for link prediction based on dynamic metrics stores computer executable instructions, wherein the computer executable instructions are set to:
[0081] Based on the preset event construction requirements, link prediction training samples are obtained by constructing event graph analysis; graph neural network inference is performed on the link prediction training samples to determine the positive and negative sample vectorized data; the positive and negative sample vectorized data are averaged to obtain the sample similarity set mean, and the sample similarity set mean is dynamically measured and analyzed to obtain the dynamic measurement threshold; the dynamic measurement threshold is measured and loss is calculated to obtain the graph neural network weight update data; based on the graph neural network weight update data, the updated graph data is determined by similar node links; based on the updated graph data, the event risk level similarity set is determined through graph neural network analysis.
[0082] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the IoT device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.
[0083] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.
[0084] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0085] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0086] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0088] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0089] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0090] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0091] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0092] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A link prediction method based on dynamic metrics, characterized in that: The method comprises: Based on the preset construction event requirements, link prediction training samples are obtained through construction event graph analysis; Performing graph neural network inference on the link prediction training samples to determine positive and negative sample vectorized data; Performing mean processing on the positive and negative sample vectorized data to obtain a sample similarity set mean, and performing dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; Performing metric loss calculation on the dynamic metric threshold to obtain graph neural network weight update data; Based on the graph neural network weight update data, determine to update the graph data through similar node links; Based on the updated graph data, a set of event risk level similarities is determined through graph neural network analysis.
2. A link prediction method based on dynamic metrics according to claim 1, characterized in that: Based on the preset construction event requirements, link prediction training samples are obtained through construction event graph analysis, including: Constructing event risk levels for the event requirements to obtain an event risk level map; Based on the event risk level map, determining an event similarity map through event similarity analysis; According to the event risk level map and the event similarity map, the link prediction training samples are obtained by setting positive and negative samples.
3. The link prediction method based on dynamic metrics according to claim 1, characterized in that: The positive and negative sample vectorized data are averaged to obtain the sample similarity set mean, specifically including: Performing node similarity analysis on the positive and negative sample vectorized data to determine similarity data between the positive and negative sample nodes; Based on the similarity data between the positive and negative sample nodes, a sample similarity set is obtained through sample integration; wherein the sample similarity set includes: a positive sample similarity set and a negative sample similarity set; A similarity averaging process is performed on the sample similarity set to obtain a mean value of the sample similarity set.
4. A link prediction method based on dynamic metrics according to claim 1, characterized in that: Performing dynamic measurement analysis on the sample similarity set mean to obtain a dynamic measurement threshold, specifically including: Performing a positive sample lower limit analysis on the sample similarity set mean to determine a positive sample lower limit threshold; Performing negative sample upper limit analysis on the sample similarity set mean to determine a negative sample upper limit threshold; The dynamic metric threshold is obtained based on the positive sample lower limit threshold and the negative sample lower limit threshold.
5. The link prediction method based on dynamic metrics according to claim 1, characterized in that: The dynamic metric threshold is subjected to metric loss calculation to obtain graph neural network weight update data, specifically including: Based on the dynamic measurement threshold, by performing a similarity set zeroing process, an updated similarity set mean is obtained; The updated similarity set mean is subjected to metric loss calculation to determine metric loss data; wherein the metric loss data is represented by the following formula: loss=1-s p_ave +s n_ave where s p_ave is the mean of the positive sample similarity set, s n_ave is the mean of the negative sample similarity set, and loss is the measurement loss data; According to the metric loss data, the graph neural network weight update data is obtained through the gradient descent method.
6. A link prediction method based on dynamic metrics according to claim 1, characterized in that: Based on the graph neural network weight update data, the graph data is updated by linking similar nodes, specifically including: Based on the graph neural network weight update data, a graph neural network model is determined; according to the graph neural network model, an event semantic embedding vector is determined through event prediction phase analysis; The event semantic embedding vector is connected with graph-similar event nodes to determine the updated graph data.
7. A link prediction method based on dynamic metrics according to claim 6, characterized in that: According to the updated graph data, a set of event risk level similarities is determined through graph neural network analysis, specifically including: Inputting the updated graph data into the graph neural network model to obtain a final vector of event risk level; Performing similarity set processing on the final vector of the event risk level, and sorting the similarity sets obtained by the similarity set processing to obtain a similarity set sequence; The risk level with the highest similarity in the similarity set sequence is set as the final result to determine the event risk level similarity set.
8. A link prediction method based on dynamic metrics according to claim 1, characterized in that: After determining the event risk level similarity set according to the updated graph data through the input of the graph neural network, the method further includes: Based on the risk level similarity set of the events, a risk response strategy is determined through risk estimation; Performing monitoring node configuration on the risk response strategy to obtain risk response processing data; Based on the risk response processing data, the risk plan improvement data is determined through management plan analysis.
9. A link prediction device based on dynamic metrics, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Based on the preset construction event requirements, link prediction training samples are obtained through construction event graph analysis; Performing graph neural network inference on the link prediction training samples to determine positive and negative sample vectorized data; Performing mean processing on the positive and negative sample vectorized data to obtain a sample similarity set mean, and performing dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; Performing metric loss calculation on the dynamic metric threshold to obtain graph neural network weight update data; Based on the graph neural network weight update data, determine to update the graph data through similar node links; Based on the updated graph data, a set of event risk level similarities is determined through graph neural network analysis.
10. A non-volatile computer storage medium for link prediction based on dynamic metrics, storing computer executable instructions, characterized in that: The computer executable instructions are configured to: Based on the preset construction event requirements, link prediction training samples are obtained through construction event graph analysis; Performing graph neural network inference on the link prediction training samples to determine positive and negative sample vectorized data; Performing mean processing on the positive and negative sample vectorized data to obtain a sample similarity set mean, and performing dynamic metric analysis on the sample similarity set mean to obtain a dynamic metric threshold; Performing metric loss calculation on the dynamic metric threshold to obtain graph neural network weight update data; Based on the graph neural network weight update data, determine to update the graph data through similar node links; Based on the updated graph data, a set of event risk level similarities is determined through graph neural network analysis.