Event knowledge graph improvement method and device based on data driving and medium

Abstract

The invention relates to an event knowledge graph improvement method and device based on data driving and a medium. The method comprises the steps of collecting and preprocessing numerical data; learning event causal relationships by using a Markov boundary discovery algorithm, and constructing an event causal relationship set; screening events directly related to the target event through a condition independence test; quantitatively evaluating the causal strength of the event pair, verifying the triad in the event knowledge graph, and adding the triad with the causal strength lower than a preset threshold value into a to-be-verified set; for event pairs which are insufficient in causality but significant in correlation, events which meet significance conditions are reserved based on correlation test; complementing event pairs with significant causal relationships in the data but missing in the knowledge graph, and generating and bidirectionally verifying a complemented event triple; and the event knowledge graph is updated by integrating the verified and complemented triad. The method effectively improves the accuracy and efficiency of causal inference.

Description

Technical Field

[0001] The present invention relates to the field of data mining and knowledge engineering, and in particular to a method, device and medium for improving event knowledge graphs based on data-driven. Background Art

[0002] With the rapid development of big data and artificial intelligence technologies, event knowledge graphs have become an important infrastructure in the fields of information retrieval, risk prediction, emergency response, etc. Event knowledge graphs express event entities and their temporal and causal relationships in a structured way, helping computers to better understand and reason about complex event logic. However, in the process of building and applying event knowledge graphs, we often face problems such as incorrect event knowledge, missing relationships, and sparse data. These problems directly affect the accuracy, completeness, and availability of event knowledge graphs, thereby limiting their performance in actual application scenarios. Therefore, how to effectively improve the quality of event knowledge graphs has become an important direction of current research.

[0003] Existing methods for improving event knowledge graphs mainly include methods based on statistical features, methods based on machine learning embedding, and methods that rely on manual verification. Statistical feature methods identify incorrect relationships and complete missing events by analyzing the time path, node type, and event frequency characteristics in the event graph. However, such methods have limited effectiveness in sparse data and high-noise scenarios. Embedding methods based on machine learning attempt to map event knowledge graphs to vector spaces and quantitatively evaluate the relationships between event entities through scoring functions, but the performance of such methods degrades under sparse data, and the models often lack interpretability. Although manual verification can better ensure the accuracy of event relationships, it requires the participation of a large number of domain experts, which is not only costly but also difficult to promote on a large scale.

[0004] In the existing technology, for example, patent (CN202411249513.6) verifies the quality of news by calculating the scores of title text and related text; patent disclosure (CN119089995A) proposes a knowledge graph embedding method and system based on multi-relational knowledge enhanced graph convolutional network, which can efficiently construct knowledge graph embedding as a representation of the relationship between evaluation entities; patent disclosure (CN119150966A) proposes a quality assessment method for vertical field knowledge graphs and designs a general idea for verifying the quality of knowledge graphs in specific fields. These methods all belong to the three types of traditional knowledge graph improvement methods mentioned above, and face many challenges such as high operating costs, noisy data, and sparse features.

[0005] To solve the above problems, data-driven event knowledge graph improvement methods have gradually attracted attention. Numerical data, as a quantitative record of real-world events, is widely used in emergency management, medical safety, financial risks and other fields, and contains a large amount of implicit causal relationships and statistical correlation information. Data-driven methods attempt to perform causal analysis and correlation verification through numerical data to detect erroneous event knowledge and complete missing event relationships. This method not only reduces the dependence on the internal structure of the event knowledge graph, but also shows strong adaptability in sparse and high-noise scenarios. However, there are significant differences in the representation of numerical data and event knowledge graphs - numerical data is usually presented in the form of a time series matrix, while the event knowledge graph consists of discrete event-relationship-event triplets. This difference increases the difficulty of expression fusion and joint reasoning between the two.

[0006] Therefore, how to effectively integrate numerical data with event knowledge graphs to achieve automatic verification, error detection and relationship completion of event knowledge has become a key issue that needs to be solved urgently. Summary of the invention

[0007] The present invention provides a data-driven event knowledge graph improvement method, device and medium, the purpose of which is to improve the accuracy, completeness and availability of the event knowledge graph, so as to better support application scenarios such as event reasoning, risk prediction and decision assistance.

[0008] To achieve the above object, the first aspect of the present invention provides a data-driven event knowledge graph improvement method, comprising the following steps: Collecting numerical data related to the event knowledge graph and preprocessing the numerical data; Using the Markov boundary discovery algorithm, we learn the direct causal relationship between events from the preprocessed numerical data and construct a set of event causal relationships; The event pairs in the event causal relationship set are screened by the conditional independence test method, and the events with direct causal relationship with the target event are retained; The causal strength calculation formula is used to quantitatively evaluate the relationship between each pair of screened events, and the causal strength value between each pair of events is calculated; Preliminary screening of event triplets in the event knowledge graph is performed based on the causal strength value, and event triplets whose causal strength values ​​do not reach the preset threshold are added to the set to be verified; For the event triples in the set to be verified, the significance level is calculated based on the correlation test method. If the significance level meets the preset threshold condition, the event triple is retained, otherwise it is deleted from the event knowledge graph; Based on the calculation of causal strength, the event pairs that have significant causal relationships in the preprocessed numerical data but are missing in the event knowledge graph are completed to generate completed event triples; Perform bidirectional causal strength verification on the completed event triples to make the completed relationship bidirectionally valid in the event knowledge graph; Integrate the verified and completed event triples into the original event knowledge graph and update the event knowledge graph.

[0009] Furthermore, collecting numerical data related to the event knowledge graph includes the following steps: Collecting event-related numerical data from different data sources, including sensor records, time series data, log files, statistical databases, or experimental observations, the collected numerical data including event attributes, event occurrence time, event frequency, and correlation between events; The collected event-related numerical data are integrated into a data matrix, in which each column represents an event attribute and each row represents an event record, to obtain numerical data related to the event knowledge graph.

[0010] Furthermore, preprocessing the numerical data includes: standardizing, normalizing and denoising the collected numerical data.

[0011] Furthermore, the standardization and normalization processing is performed through the maximum and minimum value normalization formula, which is: in, is the normalized event attribute value, is the original event attribute value, and The original event attributes The maximum and minimum values ​​of .

[0012] Furthermore, when the conditional independence test satisfies the following conditions, the event pair is considered to have a direct causal relationship: in, For a given set of condition variables Under the conditions, the event and directly causally related events The probability of simultaneous occurrence, For a given set of condition variables Under the conditions, the event occurs The probability of For a given set of condition variables Under the condition of direct causal relationship The probability of occurrence, is the set of conditional variables used in the conditional independence test, For events, It is an event directly causally related events, For events , Direct causal related events The value of a property or variable.

[0013] Furthermore, the causal strength value between the calculated events is quantified using the causal strength function, where: in, For events and directly causally related events The causal strength value between Indicates directly causally related events About the event The degree of influence Indicates things directly causally related events degree of impact.

[0014] Furthermore, the correlation test uses the significance level calculation formula: in, For events and directly causally related events The statistic for the significance test of the correlation between For events A variable value in Direct causal related events A variable value in For events The set of all variable values ​​in , Direct causal related events The set of all variable values ​​in , is the observed value, is the expected value.

[0015] Furthermore, the conditions for event triple completion are: in Indicates an event and directly causally related events There is a direct causal relationship. is the Markov frontier set, is the causal strength threshold, which determines whether to complete This event triple.

[0016] To achieve the above-mentioned purpose, the second aspect of the present invention provides an electronic device, including a processor and a memory, wherein the processor is used to implement the steps of the data-driven event knowledge graph improvement method when executing a computer program stored in the memory.

[0017] To achieve the above-mentioned objectives, the third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the data-driven event knowledge graph improvement method are executed.

[0018] Beneficial effects of the present invention: Compared with the prior art, the present invention provides an improved method, device and medium for event knowledge graph based on data-driven, which verifies, detects errors and completes missing events in an automated manner by utilizing causal relationships and statistical correlation information in numerical data. Specifically, the method uses the Markov boundary discovery algorithm to extract direct causal relationships between events from numerical data, and at the same time identifies other events directly related to the target event through conditional independence tests and causal strength calculations to eliminate erroneous knowledge and complete missing relationships. In addition, correlation analysis is combined to ensure that event relationships with insufficient causality but significant correlation can be effectively identified and retained. By introducing a deep fusion of causal and correlation information, this method significantly reduces reliance on manual verification, reduces operating costs, and improves the robustness of the system in complex event scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0020] Figure 1 It is a flow chart of a data-driven event knowledge graph improvement method disclosed in an embodiment of the present invention.

[0021] Figure 2 It is a flow chart of the main steps of an event knowledge graph improvement task disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] According to an embodiment of the present invention, it should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the following method, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0024] like Figure 1 , Figure 2 As shown, the present invention provides a data-driven event knowledge graph improvement method, comprising the following steps: Step S100: collect numerical data related to the event knowledge graph and pre-process the numerical data; First, we need to collect numerical data related to the event knowledge graph. These data can come from different fields and data sources, such as sensor records, time series data, log files, statistical databases, or experimental observations. Specifically, the collected numerical data should contain information such as the timestamp of the event, event attributes (such as event type, degree of loss, etc.), the frequency of event occurrence, and the correlation between events. These data are stored in the form of a data matrix (DM), where each column represents an event attribute and each row represents an event record, and the numerical data related to the event knowledge graph is obtained.

[0025] Next, the collected numerical data is preprocessed to ensure that the data quality is suitable for subsequent causal analysis and correlation verification. The preprocessing steps include data standardization, normalization and noise reduction. The purpose of standardization is to eliminate the influence of different dimensions and ensure that each event attribute is compared at the same scale; normalization is to map the value of each event attribute to a unified range (such as the [0,1] interval) so that data with different attributes can be processed under the same standard. Noise reduction is mainly to remove irrelevant interference and outliers in the data, which can be done through methods such as median filtering, smoothing or outlier detection to ensure that the data quality meets the requirements of subsequent causal analysis and correlation verification. In addition, for time series data, it is necessary to divide the time window, and count and aggregate the time series data according to the appropriate time granularity to better reveal the causal relationship between events. In this process, if there are missing values ​​in the data, they should be filled in by reasonable methods to ensure the integrity and validity of the data.

[0026] In this embodiment, for the event data matrix , each event attribute column The maximum-minimum value normalization is performed, and the formula is as follows: in, is the normalized event attribute value, is the original event attribute value, and The original event attributes The maximum and minimum values ​​of .

[0027] Step S200, using the Markov boundary discovery algorithm to learn the direct causal relationship between events from the preprocessed numerical data, and construct an event causal relationship set; Each event in the event data matrix is ​​regarded as a node, and an event causal relationship network is constructed by analyzing the event attributes and their correlation. In this network, the edge represents the causal relationship between events, and the edge weight represents the strength of the causal relationship. By using the Markov boundary discovery algorithm, other event sets directly related to the target event (such as a specific event) can be automatically identified from the data, excluding those events that have no direct causal relationship with the target event. Based on the principle of conditional independence, the algorithm gradually eliminates redundant and indirect event relationships, and finally obtains an event causal relationship set that only contains direct causal relationships, ensuring that the extracted causal structure is highly accurate and representative.

[0028] Step S300: screening event pairs in the event causal relationship set by a conditional independence test method, and retaining events that have a direct causal relationship with the target event; For each pair of events and directly causally related events , the conditional independence test formula is used to determine whether there is a direct causal relationship between the two: in, For a given set of condition variables Under the conditions, the event and directly causally related events The probability of simultaneous occurrence, For a given set of condition variables Under the conditions, the event occurs The probability of For a given set of condition variables Under the condition of direct causal relationship The probability of occurrence, is the set of conditional variables used in the conditional independence test, For events, It is an event directly causally related events, For events , Direct causal related events The value of a property or variable.

[0029] If the test results show that the causal relationship between the two events is direct, that is, the conditional independence test is not established, then the event pair is retained; if the causal relationship is not direct, it is eliminated. Through this screening step, those indirect or irrelevant event relationships can be effectively removed, ensuring that the causal relationship that is finally retained has a high degree of relevance and practical significance.

[0030] Step S400: using a causal strength calculation formula to quantitatively evaluate each pair of event relationships screened out, and calculating the causal strength value between each pair of events; By calculating each pair of events and directly causally related events The causal strength value between: in, For events and directly causally related events The causal strength value between Indicates directly causally related events About the event The degree of influence, Indicates things directly causally related events degree of impact.

[0031] This causal strength function can reflect the strength of the causal relationship between events and help quantify the direct causal impact between each pair of events. Through this calculation step, a causal strength value can be assigned to each pair of event relationships, so as to clarify which event pairs have a strong causal relationship and which may be indirect or weak. , event pairs with causal strength below the threshold are eliminated, and only event relationships with significant causal strength are retained, ensuring that the final causal network has higher accuracy and effectiveness.

[0032] Step S500: Preliminarily screen the event triples in the event knowledge graph according to the causal strength values, and add the event triples whose causal strength values ​​do not reach the preset threshold to the set to be verified; Retrieve each event triple in the event knowledge graph , and check whether the event relationship is verified in the causal strength calculation. , if its event Not within the Markov frontier (MB) of the target event, that is: It is considered that the event relationship may be invalid and may need to be deleted from the event knowledge graph and added to the set to be verified.

[0033] It should be noted that an event pair is two events that form a triple (the causal relationship has not yet been confirmed), and an event triple is a semantic unit consisting of two events and their causal relationship edges. Deleting an event triple is equivalent to deleting the causal relationship between events (retaining the event nodes and only removing the relationship edges).

[0034] Step S600: For the event triples in the set to be verified, the significance level is calculated based on the correlation test method. If the significance level meets the preset threshold condition, the event triple is retained, otherwise it is deleted from the event knowledge graph; Identify pairs of events that have insufficient causal strength but are statistically significant. Then, use the correlation test method to calculate the event and directly causally related events The significance level of the correlation between , through statistics To evaluate, the formula is: in, For events and directly causally related events The statistic for the significance test of the correlation between For events A variable value in Direct causal related events A variable value in For events The set of all variable values ​​in , Direct causal related events The set of all variable values ​​in , is the observed value, is the expected value.

[0035] If the significance level meets the preset threshold , then the event pair is retained. Through this step, event pairs that may lack direct causal relationships but still have statistically significant correlations can be verified, thereby ensuring the integrity of the event knowledge graph and avoiding missing potential important event relationships. Under the dual verification of causality and correlation, dynamic optimization of the event knowledge graph is achieved.

[0036] Step S700: based on the calculation of causal strength, complete the event pairs that have significant causal relationships in the preprocessed numerical data but are missing in the event knowledge graph, and generate completed event triples; Using the Markov frontier discovery algorithm, we can identify pairs of events that have significant causal relationships in the data matrix. , but these event pairs have not yet appeared in the existing event knowledge graph. For each event pair that meets the completion conditions , if the event has a significant causal relationship, that is: And causal strength , then the event pair is taken as a new triple ( ,lead to, ) to the event knowledge graph. This process can effectively fill in the key event relationships that are missed due to data scarcity or noise, thereby improving the completeness and accuracy of the event knowledge graph. In the conditions for event triple completion, Indicates an event and directly causally related events There is a direct causal relationship. is the Markov frontier set, is the causal strength threshold, which determines whether to complete This triplet.

[0037] Step S800: Perform bidirectional causal strength verification on the completed event triples to make the completion relationship bidirectionally valid in the event knowledge graph; For each pair of completed event triples ( ,lead to, ), first calculate the event and directly causally related events Causal strength and , respectively evaluate the causal strength between the two. If both causal strength values ​​are greater than the preset causal strength threshold , then the event relationship is considered to be valid in both directions, confirming that the causal relationship of the event triple is stable and has reliable bidirectional support. If the causal strength value is insufficient, the causal relationship of the event triple is considered to be invalid and needs to be further adjusted or deleted. This step ensures that the completed event relationship is not only valid from a unidirectional causal perspective, but also has a solid foundation in bidirectional causal support, improving the reliability and consistency of the event knowledge graph.

[0038] Step S900: Integrate the verified and completed event triples into the original event knowledge graph to update the event knowledge graph.

[0039] All valid triples that have been causally verified and completed Add to the event knowledge graph to ensure that the causal relationship and statistical correlation of each event triple are fully verified and supplemented. Then, the updated event knowledge graph is stored for subsequent query, analysis and application. The updated knowledge graph can be applied to multiple practical scenarios, such as event prediction, risk analysis, fault diagnosis, etc., to provide more accurate and comprehensive decision support. In addition, during the storage process, the event knowledge graph can also be interfaced with other systems to achieve data sharing and integration, providing strong support for event response, risk assessment and other tasks of other systems.

[0040] In this embodiment, the causal strength threshold is optimized for different event knowledge graphs and data scenarios. and correlation significance threshold , and conduct performance evaluation to ensure that the improvement effects achieve the expected goals.

[0041] Among the above methods, the present invention has a number of significant advantages and positive effects. First, by combining causal analysis with correlation verification, it is possible to efficiently identify erroneous event knowledge and complete missing event relationships, thereby significantly improving the accuracy and completeness of the event knowledge graph. Secondly, compared with the traditional method that relies on manual verification by experts, the present invention reduces the cost and time overhead of manual intervention through a data-driven automated verification and completion process. In addition, the present invention innovatively deeply integrates numerical data with the event knowledge graph, and uses the causal and correlation information of the data to achieve two-way support and optimization of data and knowledge. This method shows strong robustness when processing sparse knowledge and high-noise data, and can adapt to the needs of event knowledge graph improvement in different scenarios. Finally, the improved event knowledge graph has broad application prospects and can be used in multiple fields such as event prediction, risk analysis, and fault diagnosis, providing reliable support for decision-making in complex event scenarios.

[0042] To further illustrate the above solution, a specific implementation example of an event knowledge graph improvement method is given below (for the convenience of explanation, example parameters are substituted for example, but the specific implementation is not limited to the specific parameters of this example): 1. Data Collection and Preprocessing Step 1.1: Event data collection Numerical data related to events are collected from a city emergency management system. The data includes attributes such as accident type, occurrence time, location, loss extent, response time, etc. Data sources include sensor logs, historical accident record databases, and real-time monitoring systems. These data are integrated into a data matrix, with each column representing an event attribute and each row representing an event record.

[0043] Step 1.2: Event data preprocessing Standardize and normalize the data matrix. For example, for the event attribute matrix , each attribute column Perform maximum-minimum normalization: in, and The original event attributes In addition, the time series data is aggregated according to the 24-hour time window and missing values ​​are filled to ensure that the data quality meets the needs of subsequent analysis.

[0044] 2. Markov Frontier Discovery Step 2.1: Learning the causal structure of events Using the Markov boundary discovery algorithm, we can extract the direct causal relationship between different events. For example, we can identify the direct causal relationship between “rainstorm event” and “road collapse”.

[0045] Step 2.2: Conditional independence test For each pair of event relationships, a conditional independence test is performed to determine whether there is a direct causal relationship between the two: If the two satisfy the conditional independence test, it is considered that there is no direct causal relationship.

[0046] Step 2.3: Calculation of event causal strength The causal relationship between "rainstorm event" and "road collapse" is quantified and its causal strength is calculated: When the causal strength exceeds the preset threshold , and believed that there was a significant causal relationship between the two.

[0047] 3. Event Knowledge Error Detection Step 3.1: Error event knowledge verification There is a triple (rainstorm, cause, power outage) in the existing event knowledge graph, but data analysis shows that there is no causal relationship between the two. Then add the triple to the set to be verified: Step 3.2: Additional verification of event correlation For some pairs of events with insufficient causality but significant correlation, such as "heavy rain" and "traffic congestion", a correlation significance test is performed:

[0048] If the significance level If the requirements are met, the triplet is retained.

[0049] 4. Completion of missing event knowledge Step 4.1: Event causal knowledge completion Through Markov frontier discovery, a missing event relationship is identified: (Heavy rain caused water to accumulate on the road) According to the causal strength calculation, the relationship satisfies the completion condition:

[0050] Therefore, the triple is added to the event knowledge graph.

[0051] Step 4.2: Bidirectional event causality verification Perform bidirectional verification on the newly added triples to ensure that the causal relationship between events is stable and has bidirectional support.

[0052] 5. Parameter Optimization Step 5.1: Causal Threshold Optimization Select the optimal parameters from the empirical values ​​and finally set the causal strength threshold and correlation significance threshold , and obtain the best improvement effect in this scenario.

[0053] Step 5.2: Performance Evaluation The improved event knowledge graph is evaluated using indicators such as precision, recall, and F1-Score. The results are as follows: Accuracy: 90.5%, Recall: 87.3%, F1-Score: 88.8%.

[0054] 6. Event Knowledge Graph Output Step 6.1: Event knowledge graph update The verified and completed event triples are merged into the original event knowledge graph to generate a new event knowledge graph.

[0055] Step 6.2: Event Knowledge Graph Storage The improved event knowledge graph is stored in the event management database for other systems to call and query.

[0056] Step 6.3: Event Knowledge Graph Application In the urban risk monitoring platform, the improved event knowledge graph is used for real-time event prediction, fault diagnosis and emergency response.

[0057] According to another aspect of an embodiment of the present application, there is further provided an electronic device, including a processor and a memory, wherein the processor is configured to implement the steps of the method when executing a computer program stored in the memory.

[0058] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0059] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0060] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0061] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0062] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A data-driven event knowledge graph improvement method, characterized in that: The steps include: Collecting numerical data related to the event knowledge graph and preprocessing the numerical data; Using the Markov boundary discovery algorithm, we learn the direct causal relationship between events from the preprocessed numerical data and construct a set of event causal relationships; The event pairs in the event causal relationship set are screened by the conditional independence test method, and the events with direct causal relationship with the target event are retained; The causal strength calculation formula is used to quantitatively evaluate the relationship between each pair of screened events, and the causal strength value between each pair of events is calculated; Preliminary screening of event triplets in the event knowledge graph is performed based on the causal strength value, and event triplets whose causal strength values ​​do not reach the preset threshold are added to the set to be verified; For the event triples in the set to be verified, the significance level is calculated based on the correlation test method. If the significance level meets the preset threshold condition, the event triple is retained, otherwise it is deleted from the event knowledge graph; Based on the calculation of causal strength, the event pairs that have significant causal relationships in the preprocessed numerical data but are missing in the event knowledge graph are completed to generate completed event triples; Perform bidirectional causal strength verification on the completed event triples to make the completed relationship bidirectionally valid in the event knowledge graph; Integrate the verified and completed event triples into the original event knowledge graph and update the event knowledge graph.

2. The data-driven event knowledge graph improvement method according to claim 1, characterized in that: Collecting numerical data related to the event knowledge graph includes the following steps: Collecting event-related numerical data from different data sources, including sensor records, time series data, log files, statistical databases, or experimental observations, the collected numerical data including event attributes, event occurrence time, event frequency, and correlation between events; The collected event-related numerical data are integrated into a data matrix, in which each column represents an event attribute and each row represents an event record, to obtain numerical data related to the event knowledge graph.

3. The data-driven event knowledge graph improvement method according to claim 2, characterized in that: The preprocessing of the numerical data includes: standardizing, normalizing and denoising the collected numerical data.

4. The data-driven event knowledge graph improvement method according to claim 3 is characterized in that: The normalization process is performed using the maximum and minimum value normalization formula, which is: in, is the normalized event attribute value, is the original event attribute value, and The original event attributes The maximum and minimum values ​​of .

5. The data-driven event knowledge graph improvement method according to claim 1, characterized in that: When the conditional independence test meets the following conditions, the event pair is considered to have a direct causal relationship: in, For a given set of condition variables Under the conditions, the event and directly causally related events The probability of simultaneous occurrence, For a given set of condition variables Under the conditions, the event The probability of occurrence, For a given set of condition variables Under the condition of direct causal relationship The probability of occurrence, is the set of conditional variables used in the conditional independence test, For events, It is an event directly causally related events, For events , Direct causal related events The value of a property or variable.

6. The data-driven event knowledge graph improvement method according to claim 1, characterized in that: The causal strength value between calculated events is quantified using the causal strength function, where: in, For events and directly causally related events The causal strength value between Indicates directly causally related events About Events The degree of influence, Indicates an event Direct causal relationship degree of impact.

7. The data-driven event knowledge graph improvement method according to claim 1, characterized in that: The correlation test uses the significance level calculation formula: in, For events and directly causally related events The statistic for the significance test of the correlation between For events A variable value in Direct causal related events A variable value in For events The set of all variable values ​​in , Direct causal related events The set of all variable values ​​in , is the observed value, is the expected value.

8. The data-driven event knowledge graph improvement method according to claim 1, characterized in that: The conditions for event triple completion are: in, Indicates an event and directly causally related events There is a direct causal relationship. is the Markov frontier set, is the causal strength threshold, which determines whether to complete This event triple.

9. An electronic device, characterized in that: It includes a processor and a memory, and the processor is used to implement the steps of the data-driven event knowledge graph improvement method as described in any one of claims 1 to 8 when executing the computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the data-driven event knowledge graph improvement method according to any one of claims 1 to 8 are executed.

Images