Data attribution analysis system based on indexes

Abstract

The invention provides an index-based data attribution analysis system, which relates to the technical field of data analysis, and comprises a data fusion layer, a dynamic graph engine and an attribution calculation layer, the data fusion layer is used for uniformly processing and integrating data from different sources to form a triple knowledge graph, and the sources comprise a service database, an event stream and an expert knowledge base; the dynamic graph engine is used for processing and analyzing real-time data, generating a corresponding risk propagation model and updating the knowledge graph to obtain a dynamic graph; the attribution calculation layer is used for dynamically calculating the risk value and the influence of each node according to the dynamic graph; and carrying out dimension contribution degree fusion attribution calculation to obtain a dimension contribution value. According to the method, the problems of poor adaptability and insufficient interpretability of attribution analysis are solved through a dynamic reasoning mechanism and a risk propagation model of the knowledge graph, and more accurate and real-time risk assessment and contribution degree analysis can be provided.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a data attribution analysis system based on indicators. Background Technology

[0002] Attribution analysis is a method that uses data analysis to identify and quantify the key drivers (such as marketing channels, reach paths, and user behaviors) behind user behavior or business outcomes (such as conversions, purchases, and registrations). Its core objective is to solve the problem of "multi-touchpoint contribution allocation"—before a user completes a key action (such as placing an order), they may have been exposed to multiple channels (such as search ads, social media, email, and organic search). It is necessary to scientifically evaluate the contribution value of each channel and touchpoint to the outcome, thereby optimizing resource allocation and strategies.

[0003] Currently, attribution analysis is widely used as an important technical tool in enterprise data analysis, particularly in areas such as financial risk control, marketing effectiveness evaluation, and business decision-making. Traditional attribution analysis methods often rely on static rules or single algorithms, which suffer from insufficient adaptability and inefficiency when facing complex and constantly changing business environments. Specific problems include:

[0004] (1) Limitations of traditional attribution models: They typically rely on static rules or single algorithms, such as decision trees and Bayesian networks. These models lack sufficient adaptability and flexibility when facing complex business scenarios and dynamic changes in the relationships between indicators. For example, when an e-commerce platform shifts from a traditional model to diversified services, traditional static attribution models often cannot adapt to the changes in business models in a timely manner, which leads to a lag in analysis results and thus affects decision-making efficiency.

[0005] (2) The problem of fragmented multi-source data: Enterprises often use multiple data sources for decision support, including expert experience, historical data, and real-time streams. However, due to the lack of effective integration between these data sources, risk assessment in attribution analysis is lagging, and the calculation of contribution is biased. For example, in financial risk control scenarios, the impact of industry policy changes on bad debt rates is ignored, leading to misjudgments.

[0006] (3) It lacks sufficient interpretability and cannot intuitively show the causal relationship between indicator fluctuations and dimensions.

[0007] With the rapid development of big data technology and the application of artificial intelligence, how to process and analyze massive amounts of data has become an important issue. In particular, in the integration of multi-source data and dynamic business scenarios, how to perform real-time and accurate attribution analysis of indicator data and assess risk value and contribution has become a problem that many enterprises urgently need to solve. Summary of the Invention

[0008] To address the aforementioned issues, this invention provides a data attribution analysis system based on indicators, which aims to improve the accuracy of attribution analysis by quantifying the risk value and dimensional contribution value of indicators in real time through the dynamic reasoning mechanism of knowledge graphs.

[0009] To achieve the above objectives, the present invention provides an indicator-based data attribution analysis system, comprising: a data fusion layer, a dynamic graph engine, and an attribution calculation layer;

[0010] The data fusion layer is used for:

[0011] Data from different sources is processed and integrated in a unified manner to form a knowledge graph of triples. The sources include: business databases, event streams, and expert knowledge bases.

[0012] The dynamic graph engine is used for:

[0013] Real-time data is processed and analyzed to generate corresponding risk propagation models, and the knowledge graph is updated to obtain a dynamic graph.

[0014] The attribution calculation layer is used for:

[0015] The risk value and influence of each node are dynamically calculated based on the dynamic graph.

[0016] Perform dimensional contribution fusion and attribution calculations to obtain dimensional contribution values.

[0017] As a further improvement of the present invention, the data fusion layer accesses the business database to obtain time-series data of indicators, including market transaction data and financial data.

[0018] The data fusion layer accesses the event stream and receives market announcements, news events, and policy changes in real time.

[0019] The data fusion layer connects to the expert knowledge base, introduces the rule table of the expert knowledge base, extracts professional information, and supplements or enhances the interpretability of the data.

[0020] As a further improvement of the present invention, the dynamic graph engine includes an incremental update module, which dynamically adjusts the nodes and edges in the dynamic graph through an event-triggered mechanism.

[0021] As a further improvement of the present invention, a preset fluctuation threshold is set for each indicator. When the indicator fluctuation exceeds the preset threshold, the dynamic update module dynamically adjusts the corresponding nodes and edges in the graph.

[0022] As a further improvement of the present invention, the TransE graph embedding algorithm is used to adjust the node representation.

[0023] As a further improvement of the present invention, the incremental update module includes a dynamic map update mechanism, which drives a data drift trigger, an expert intervention trigger, and an external event trigger respectively through event monitoring;

[0024] The data drift trigger calculates the KL divergence, determines the KL divergence value and activates subgraph reconstruction, and performs subgraph embedding update based on the TransE model;

[0025] The expert intervention trigger determines whether manually labeled nodes are invalid or newly added. Based on the TransR model, the adjacent edges of the invalid or newly added manually labeled nodes are retrained to adjust the relationships between nodes in the knowledge graph.

[0026] The external event trigger performs NLP entity recognition, creates event nodes in the knowledge graph based on the recognition results, and links the newly created nodes to relevant metrics;

[0027] The dynamic graph is iterated based on the triggering results of each trigger, which is the real-time risk transmission.

[0028] As a further improvement of the present invention, the risk propagation model utilizes a graph convolutional network to propagate features of nodes in the dynamic graph, thereby achieving cross-node risk propagation.

[0029] As a further improvement of the present invention, the attribution calculation layer comprehensively evaluates the real-time risk value of each node by considering factors such as node centrality, historical volatility, and edge weight.

[0030] As a further improvement of the present invention, the attribution calculation layer performs dimensional contribution fusion attribution calculation to obtain dimensional contribution values, including:

[0031] The indicators are decomposed into hierarchical levels using a dimensional tree according to different dimensions, and the local contribution of each level is calculated.

[0032] Based on the causal paths in the dynamic graph, the local contribution is corrected according to the causal strength.

[0033] The weighted product of each causal path is used to obtain the final dimensional contribution value.

[0034] As a further improvement of the present invention, a closed-loop feedback mechanism is also included, which feeds back the results of the attribution calculation to the dynamic graph.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] This invention constructs a dynamically updatable domain knowledge graph by integrating three types of data sources: business rules, historical data, and expert experience. In this graph, nodes represent various entities, such as metrics, dimensions, and events, while edges represent the relationships between these entities, such as "influence" and "dependency." The knowledge graph enables rapid response to business changes by dynamically updating nodes and edges in real time. Simultaneously, a risk propagation model calculates the risk transmission weights between nodes in the graph, thereby updating the risk values ​​of each indicator in real time. Through the knowledge graph of business data, the transmission paths of risks can be accurately identified, and the risk level of each node can be precisely calculated. A contribution decomposition algorithm is provided, integrating hierarchical decomposition and causal reasoning techniques to generate multi-dimensional contribution values. By performing multi-level analysis of the attribution analysis results, the different roles and contributions of each dimension in indicator fluctuations can be explained. This invention fully considers the interrelationships between dimensions, thus providing more accurate contribution calculations. Furthermore, a feedback optimization mechanism allows the attribution analysis results to be fed back into the knowledge graph, forming a closed-loop learning system. Ultimately, this achieves real-time quantification of indicator risk values ​​and dimension contribution values, offering advantages such as adaptability to complex business environments, high interpretability, and significantly improved accuracy of attribution analysis.

[0037] Compared to traditional attribution methods that require manual pre-setting of dimensional weights, this invention automatically constructs a multi-dimensional relational network through a graph and uses graph neural networks to calculate contributions in real time, thereby reducing manual maintenance costs and improving analysis efficiency. Compared to traditional static models that struggle to cope with sudden market changes, such as changes in consumer behavior caused by a pandemic, this invention dynamically adjusts the graph by injecting new nodes in real time through event triggers and updating risk transmission paths. Compared to traditional black-box models that cannot provide clear explanations of contributions and guide actual decision-making, this invention visualizes causal chains, allowing business stakeholders to adjust strategies based on these chains, and then uses verification to feed back into the graph weights, forming a closed loop of "attribution-decision-feedback." Compared to traditional data attribution analysis where expert experience is often disconnected from the data model, this invention transforms expert experience into graph relation edges, effectively integrating prior expert knowledge and improving the reliability of attribution in sparse data scenarios.

[0038] Compared to traditional attribution methods, this invention improves analysis efficiency. Traditional attribution methods require manual preset of dimension weights, while this invention automatically constructs a multi-dimensional association network through graphs and uses graph neural networks to calculate contribution in real time, thereby reducing the cost of manual rule maintenance and improving computational efficiency.

[0039] Compared to traditional attribution methods, this invention is dynamically adaptable. Traditional static models are difficult to cope with sudden market changes, such as changes in consumer behavior caused by the pandemic. This invention, however, injects new nodes in real time through event triggers, updates the risk transmission path, and thus dynamically adjusts the graph.

[0040] Compared to traditional attribution methods, this invention enhances interpretability. Traditional black-box models cannot provide clear explanations of contribution and cannot guide actual decision-making. In contrast, this invention visualizes causal chains, enabling business stakeholders to adjust strategies based on these chains. After verification, the attribution graph weights are updated, forming a closed loop of "attribution-decision-feedback".

[0041] Compared with traditional attribution methods, this invention achieves multi-source data fusion. In traditional data attribution analysis, expert experience and data models are relatively separated. However, this solution transforms expert experience into graph relationship edges, effectively integrating expert prior knowledge and improving the reliability of attribution in sparse data scenarios. Attached Figure Description

[0042] Figure 1 This is a flowchart of a data attribution analysis system based on indicators disclosed in one embodiment of the present invention;

[0043] Figure 2 This is a flowchart illustrating the implementation of dynamic graph updating in one embodiment of the present invention;

[0044] Figure 3 This is a flowchart of a multi-algorithm attribution collaboration implementation disclosed in one embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The present invention will now be described in further detail with reference to the accompanying drawings:

[0047] like Figure 1 As shown, the present invention provides a data attribution analysis system based on indicators, comprising: a data fusion layer, a dynamic graph engine, and an attribution calculation layer;

[0048] The data fusion layer is used for:

[0049] Data from different sources is processed and integrated in a unified manner to form a knowledge graph of triples. The sources include: business databases, event streams, and expert knowledge bases.

[0050] in,

[0051] The data fusion layer connects to the business database to obtain time-series data of indicators, including market transaction data, company financial data, etc.

[0052] The data fusion layer connects to the event stream, receiving market announcements, news events, and policy changes in real time to influence subsequent calculations.

[0053] The data fusion layer connects to the expert knowledge base, introduces the rule tables of the expert knowledge base, extracts professional information, and supplements or enhances the interpretability of the data;

[0054] All incoming data will be processed using natural language processing technology to form a triplet graph structure, providing raw data support for subsequent graph analysis.

[0055] Dynamic graph engine, used for:

[0056] Real-time data is processed and analyzed to generate corresponding risk propagation models and update the knowledge graph to obtain a dynamic graph.

[0057] in,

[0058] The dynamic graph engine includes an incremental update module, which dynamically adjusts the nodes and edges in the dynamic graph through an event-triggered mechanism.

[0059] The risk propagation model uses graph convolutional networks to propagate features of nodes in a dynamic graph, enabling cross-node risk propagation. Specifically, the features of a node are affected by its neighboring nodes during the propagation process. The risk propagation model is based on a graph neural network model to calculate the risk transmission weights between nodes in the graph, thereby updating the risk values ​​of each indicator in real time.

[0060] Furthermore,

[0061] Each indicator has a preset fluctuation threshold. When the fluctuation of an indicator exceeds the preset threshold, the dynamic update module dynamically adjusts the corresponding nodes and edges in the graph.

[0062] Furthermore,

[0063] The TransE graph embedding algorithm is used to adjust the node representation.

[0064] Attribution calculation layer, used for:

[0065] The risk value and impact of each node are dynamically calculated based on the dynamic graph.

[0066] Perform dimensional contribution fusion and attribution calculations to obtain dimensional contribution values.

[0067] in,

[0068] The attribution calculation layer comprehensively evaluates the real-time risk value of each node by considering factors such as node centrality, historical volatility, and edge weight, thereby enabling dynamic monitoring of nodes and real-time response to potential risk events.

[0069] The attribution calculation layer performs dimensional contribution fusion attribution calculation to obtain dimensional contribution values, including:

[0070] Dimension tree decomposition involves breaking down the indicators into hierarchical levels according to different dimensions, and calculating the local contribution of each level.

[0071] Causal path correction, based on causal paths in the dynamic graph, corrects local contributions according to causal strength to ensure the accuracy of attribution calculation;

[0072] Weighted product calculation involves multiplying each causal path by weight to obtain the final dimensional contribution value.

[0073] The incremental update module of this invention is equipped with a map update mechanism, which is the key to ensuring the real-time performance and adaptability of the system of this invention.

[0074] Specifically, such as Figure 2 As shown, the graph update mechanism adopts an event-driven update strategy, defining three types of triggers to drive the dynamic update of the graph.

[0075] (1) Data drift trigger: When the index distribution changes significantly, the reconstruction process of the relevant subplot is triggered by Kullback-Leibler divergence measurement to ensure that the dynamic map can adapt to the new data distribution.

[0076] (2) Expert intervention trigger: When manually labeled nodes fail or are added, the TransR model is used to retrain the relevant critical edges to reflect the knowledge changes brought about by expert intervention.

[0077] (3) External event triggers, such as news reports mentioning "raw material price increases", the dynamic graph will automatically generate the corresponding time nodes and connect them with relevant indicators such as "procurement costs" to ensure that external events can be quickly integrated into the system.

[0078] In this invention, the attribution calculation layer adopts multi-algorithm attribution collaboration, that is, multiple algorithms work together.

[0079] Among them, such as Figure 3 As shown, the attribution indicators are decomposed vertically and subjected to causal analysis horizontally. Finally, the results of the horizontal and vertical analyses are fused at multiple scales and the contribution values ​​are visualized to obtain business decision support data.

[0080] Furthermore,

[0081] (1) Vertical decomposition: along the dimension tree structure, the contribution of each level is calculated by using the binary tree contribution accumulation algorithm to obtain the local contribution of each level, so as to ensure that the contribution of each level is effectively attributed.

[0082] (2) Horizontal causal analysis introduces partial mutual information to replace traditional mutual information in order to solve the nonlinear correlation problem in high-dimensional data and accurately capture the causal relationship between dimensions and indicators; obtain the causal path of the graph based on the graph path retrieval, perform PMI intensity calculation, and obtain the causal correction factor to ensure the efficiency and accuracy of the attribution results.

[0083] (3) Feedback closed-loop design: The final business decision support data is used for business decision-making. The graph edge weights are fed back through the business decision effect to optimize the graph structure. That is, the result of the attribution calculation will be fed back to the graph storage to form a closed-loop feedback mechanism. At the same time, an attribution accuracy evaluation index is constructed to automatically eliminate inefficient nodes and continuously improve the accuracy and reliability of the model.

[0084] Furthermore,

[0085] High-contribution dimension enhancement: For high-contribution dimensions, the system will automatically increase the weight of its associated edges, thereby increasing the importance of that dimension in the dynamic graph;

[0086] Low-contribution dimension degradation: For low-contribution dimensions, the system will trigger a degradation of the node's importance and reduce related computational redundancy. In this way, the system can achieve continuous optimization and precise adjustment.

[0087] Overall, this invention, through a multi-layered, dynamically updated approach, combined with graph convolutional networks, event-driven mechanisms, and attribution computation, ensures the system's rapid response and accurate assessment of risks.

[0088] Explanation of relevant terms in this invention:

[0089] Attribution analysis is an analytical method used to understand the contribution of different factors or dimensions to a particular outcome (such as changes in business metrics). Through in-depth analysis of data, attribution analysis helps reveal causal relationships between dimensions and is typically used to assess which factors are most likely to cause changes in the target variable.

[0090] Knowledge Graph: A knowledge graph is a structured semantic network that represents entities and relationships between them through nodes and edges. In data attribution analysis, knowledge graphs are used to build domain-specific knowledge structure models, helping to analyze different dimensions, metrics, and their relationships, and supporting dynamic reasoning and real-time adjustments.

[0091] Contribution: Contribution refers to the degree of influence of a specific dimension or indicator on the final result, usually calculated quantitatively through regression analysis or causal inference models. In this invention, by integrating hierarchical decomposition and causal inference techniques, the contribution of each dimension to the fluctuation of the indicator can be revealed at multiple levels, and its specific contribution value can be quantified.

[0092] Graph Neural Networks: Graph convolutional networks are a type of deep learning model designed for graph-structured data. By propagating features of nodes and their adjacencies, they can learn the potential relationships between nodes in a graph and calculate the attributes of each node.

[0093] Feedback optimization mechanism: This is a self-learning technique that refers to continuously optimizing and adjusting model parameters by feeding analysis results back into the model. In knowledge graphs and attribution analysis, this mechanism can help the system automatically adjust and optimize, improving the accuracy and adaptability of the analysis results.

[0094] Advantages of this invention:

[0095] This invention presents a dynamic knowledge graph update mechanism, a risk-contribution joint quantification model based on a multi-algorithm attribution collaborative architecture, and a feedback-driven self-optimizing closed loop, aiming to improve the efficiency and accuracy of knowledge graph updates. Specifically, for the dynamic knowledge graph update mechanism, an event-driven update strategy is designed, using three types of triggers—data drift, expert intervention, and external events—to achieve dynamic updates of the knowledge graph. Simultaneously, an incremental knowledge graph embedding training method is employed to ensure rapid convergence after topological changes, maintaining the real-time performance and stability of the knowledge graph. For the multi-algorithm attribution collaborative architecture, at the vertical level, a binary tree weighted accumulation algorithm is used to decompose the knowledge graph hierarchically along the dimensional tree, improving computational efficiency. At the horizontal level, PMI strength correction is performed based on the causal path of the knowledge graph, ensuring efficient and accurate attribution results. Furthermore, a dynamic weight allocation mechanism adjusts the contribution ratio of decomposition and causality according to the confidence level of the data, achieving more accurate attribution analysis. For the joint quantification model of risk contribution, the model utilizes indicator risk values ​​to calculate and integrate topological centrality, historical volatility, and associated edge weights to form an accurate risk contribution analysis. Simultaneously, a causal penalty factor is introduced into the dimensional contribution generation to suppress spurious dimensions, further improving the model's interpretability and accuracy. For the feedback-driven self-optimizing closed loop, the graph structure is optimized by feeding back graph edge weights based on the effects of business decisions. Furthermore, an attribution accuracy evaluation index is constructed to automatically eliminate inefficient nodes, continuously improving the model's accuracy and reliability.

[0096] In this invention, the knowledge graph constructs a knowledge framework between entities through the relationships between nodes and edges. This effectively integrates various data sources such as business rules, historical data, and expert experience, thereby improving the accuracy and interpretability of data analysis. In complex business scenarios, the knowledge graph not only helps analyze business data but also enables real-time calculation of risk values ​​and contribution levels by dynamically adjusting the graph structure.

[0097] This invention utilizes graph neural networks, a cutting-edge deep learning technology, which is particularly suitable for processing graph-structured data. It can effectively capture the complex relationships between nodes in the graph and perform information propagation and learning. Therefore, in this invention, the risk transmission weights between various indicators are calculated based on graph neural networks, thereby updating the risk value of each indicator in real time and providing accurate basis for subsequent attribution analysis.

[0098] The contribution decomposition algorithm proposed in this invention, by integrating hierarchical decomposition and causal reasoning, can provide multi-granularity contribution analysis, helping enterprises to more accurately understand the degree of influence of each dimension on target indicators. This overcomes the shortcomings of current attribution analysis methods, which lack sufficient interpretability and struggle to intuitively demonstrate the causal relationship chain between indicator fluctuations and dimensions.

[0099] This invention introduces a feedback optimization mechanism, feeding the results of attribution analysis back into the knowledge graph to form a closed-loop learning system. Through this mechanism, the system can learn and optimize its analysis results, further improving the accuracy and adaptability of attribution analysis and ensuring that enterprises can make accurate decisions even in complex and ever-changing business environments.

[0100] This invention proposes an innovative data attribution analysis technique through the dynamic reasoning mechanism of knowledge graphs, graph neural network models, and feedback optimization mechanisms. It can solve the problems of poor adaptability and insufficient interpretability of traditional methods in complex business environments, and provide enterprises with more accurate and real-time risk assessment and contribution analysis.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data attribution analysis system based on indicators, characterized in that, include: Data fusion layer, dynamic graph engine, and attribution calculation layer; The data fusion layer is used for: Data from different sources is processed and integrated in a unified manner to form a knowledge graph of triples. The sources include: business databases, event streams, and expert knowledge bases. The dynamic graph engine is used for: Real-time data is processed and analyzed to generate corresponding risk propagation models, and the knowledge graph is updated to obtain a dynamic graph. The attribution calculation layer is used for: The risk value and influence of each node are dynamically calculated based on the dynamic graph. Perform dimensional contribution fusion and attribution calculations to obtain dimensional contribution values.

2. The data attribution analysis system based on indicators according to claim 1, characterized in that: The data fusion layer accesses the business database to obtain time-series data of indicators, including market transaction data and financial data; The data fusion layer accesses the event stream and receives market announcements, news events, and policy change information in real time. The data fusion layer connects to the expert knowledge base, introduces the rule table of the expert knowledge base, extracts professional information, and supplements or enhances the interpretability of the data.

3. The data attribution analysis system based on indicators according to claim 1, characterized in that: The dynamic graph engine includes an incremental update module, which dynamically adjusts the nodes and edges in the dynamic graph through an event-triggered mechanism.

4. The data attribution analysis system based on indicators according to claim 3, characterized in that: Each indicator has a preset fluctuation threshold. When an indicator fluctuates beyond the preset threshold, the dynamic update module dynamically adjusts the corresponding nodes and edges in the graph.

5. The index-based data attribution analysis system according to claim 3 or 4, characterized in that: The TransE graph embedding algorithm is used to adjust the node representation.

6. The data attribution analysis system based on indicators according to claim 3, characterized in that: The incremental update module includes a dynamic map update mechanism, which drives data drift triggers, expert intervention triggers, and external event triggers through event monitoring. The data drift trigger calculates the KL divergence, determines the KL divergence value and activates subgraph reconstruction, and performs subgraph embedding update based on the TransE model; The expert intervention trigger determines whether manually labeled nodes are invalid or newly added. Based on the TransR model, the adjacent edges of the invalid or newly added manually labeled nodes are retrained to adjust the relationships between nodes in the knowledge graph. The external event trigger performs NLP entity recognition, creates event nodes in the knowledge graph based on the recognition results, and links the newly created nodes to relevant metrics; The dynamic graph is iterated based on the triggering results of each trigger, which is the real-time risk transmission.

7. The data attribution analysis system based on indicators according to claim 1, characterized in that: The risk propagation model utilizes graph convolutional networks to propagate features of nodes in the dynamic graph, thereby achieving cross-node risk propagation.

8. The data attribution analysis system based on indicators according to claim 1, characterized in that: The attribution calculation layer comprehensively evaluates the real-time risk value of each node by considering factors such as node centrality, historical volatility, and edge weight.

9. The data attribution analysis system based on indicators according to claim 1, characterized in that: The attribution calculation layer performs dimensional contribution fusion attribution calculation to obtain dimensional contribution values, including: The indicators are decomposed into hierarchical levels using a dimensional tree according to different dimensions, and the local contribution of each level is calculated. Based on the causal paths in the dynamic graph, the local contribution is corrected according to the causal strength. The weighted product of each causal path is used to obtain the final dimensional contribution value.

10. The data attribution analysis system based on indicators according to claim 1, characterized in that: It also includes a closed-loop feedback mechanism, which feeds back the results of the attribution calculation to the dynamic graph.

Images