Enhancing graph explainability through graph analysis algorithm

By correlating GXAI and graph analysis algorithms and generating a visualization, the method enhances the interpretability and efficiency of graph analysis, enabling users to select optimal algorithms for graph structured data.

JP2024169352APending Publication Date: 2024-12-05FUJITSU LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
JP2024080914
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-23
Filing Date
2024-05-17
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Graph analysis algorithms provide insights but are often difficult to interpret, especially when using Graph Explainable Artificial Intelligence (GXAI) results, which are presented in complex and unfamiliar formats, making it challenging for users to effectively understand and utilize the insights.

Method used

A method involving GXAI techniques and graph analysis algorithms to determine correlations, rank algorithms based on similarity and effectiveness, and generate a visualization in a GUI to assist users in selecting appropriate algorithms for graph structured data analysis.

Benefits of technology

Improves the efficiency and interpretability of graph analysis by providing a ranked list and visualization of graph analysis algorithms, allowing users to select the most suitable algorithms for their datasets based on correlation and general characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024169352000001_ABST
    Figure 2024169352000001_ABST
Patent Text Reader

Abstract

To provide a method for enhancing graph explainability through graph analysis algorithms, a storage medium, and a system.SOLUTION: A method includes: obtaining a dataset of graph-structured data, the dataset including one or more subgraphs; determining a correlation between analyses of the dataset of the graph-structured data using graph explainable artificial intelligence (GXAI) techniques and using a plurality of graph analysis algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining general characteristics of the dataset of the graph-structured data, the general characteristics indicating similarities among the one or more subgraphs; assigning a threshold number of the graph analysis algorithms to the general characteristics based on the ranked list of the plurality of graph analysis algorithms; generating an assignment table including the general characteristics and the threshold number of the graph analysis algorithms; and generating a display within a graphical user interface (GUI) that visualizes the assignment table.SELECTED DRAWING: Figure 2
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] SUMMARY OF THE DISCLOSURE The embodiments described in this disclosure relate generally to improving the explainability of graphs through graph analysis algorithms. [Background technology]

[0002] Datasets in a graph structure can be analyzed using graph analysis algorithms. For example, graph analysis algorithms can provide information about the overall structure of a dataset, including patterns and anomalies. However, there can be many graph analysis algorithms, which can make it difficult to determine which graph analysis algorithm to use for a dataset.

[0003] The subject matter claimed in this disclosure is not limited to embodiments that solve any disadvantages or that operate only in environments such as those described above. Rather, this background is only provided to illustrate one example technology area where some embodiments described in this disclosure may be practiced. Summary of the Invention

[0004] In an exemplary embodiment, a method can include obtaining a dataset of graph structured data, the dataset including one or more subgraphs. The method can further include determining a correlation between an analysis of the dataset using graph explainable artificial intelligence (GXAI) techniques and an analysis using a plurality of graph analysis algorithms. A ranked list of the plurality of graph analysis algorithms can be generated based on the correlation. The method can further include determining a general characteristic of the dataset of graph structured data, the general characteristic indicating a similarity between the one or more subgraphs. The method can further include assigning a threshold number of graph analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms. An assignment table can be generated that includes the general characteristic and the threshold number of graph analysis algorithms. A display can be generated within a graphical user interface (GUI) that visualizes the assignment table.

[0005] The object and advantages of the embodiments will be realized and achieved at least by the elements, features, and combinations particularly pointed out in the claims.

[0006] Both the foregoing general description and the following detailed description are provided by way of example and explanation, and are not limitations of the invention as claimed. [Brief description of the drawings]

[0007] Exemplary embodiments will be described and explained with additional specificity and detail through the use of the accompanying drawings in which:

[0008] [Figure 1] FIG. 1 illustrates an exemplary process flow for classifying graph analysis algorithms.

[0009] [Diagram 2] 1 shows a flowchart of an exemplary method for classifying graph analysis algorithms.

[0010] [Diagram 3] 1 is a flowchart of an exemplary method for selecting a graph analysis algorithm for a dataset.

[0011] [Figure 4] 1 illustrates a flowchart of an exemplary method for generating a rank table for a graph analysis algorithm.

[0012] [Diagram 5] 1 illustrates an exemplary computing system, fully configured in accordance with at least one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Information may be included in a dataset, and the dataset may be presented in a graph format (referred to as "graph-structured data"). The graph-structured data may be analyzed using graph explainable artificial intelligence (GXAI) techniques or graph analytics algorithms. The graph analytics algorithms may provide insights and explanations about the graph-structured data, such as predicting behavioral patterns, identifying significant nodes or edges, and identifying anomalous characteristics of the corresponding dataset. The GXAI techniques may include providing the graph-structured data to a machine learning model or other artificial intelligence-enabled computing system to evaluate the corresponding dataset. The GXAI techniques may provide insights and analysis of the graph-structured data, which may cover different perspectives or provide a more comprehensive analysis than the results of the graph analytics algorithms. However, the results provided by the GXAI techniques may be difficult for users, such as data scientists, to understand because the GXAI results may be presented in an unfamiliar, unconventional, or complex format. Thus, such results may not be user-friendly. Additionally or alternatively, the GXAI results may cover a variety of metrics or details about the graph-structured data. As a result, a user reviewing the GXAI results may not be able to properly or effectively interpret the GXAI results and may need to drill down through many layers to arrive at the desired information to evaluate the dataset for an AI model. For example, analyzing a particular dataset of graph-structured data using a particular GXAI technique may result in a contribution score corresponding to each node or edge included in the particular dataset, which contribution score indicates how each node or edge contributes to a particular classification or regression analysis of the particular dataset.

[0014] The present disclosure relates to, among other things, methods and / or systems that can be used to identify graph analysis algorithms that may be more effective for particular types of graph-structured data datasets.

[0015] In some embodiments, the method can include determining a correlation between the GXAI results and graph analytic algorithm results associated with the same dataset of graph structured data. This correlation can provide a user with a better understanding of how well each graph analytic algorithm analyzes the dataset of graph structured data. The graph analytic algorithms can be ranked according to the correlation, and a threshold number of graph analytic algorithms can be assigned to a first general characteristic of the dataset of graph structured data. The first general characteristic of the dataset can indicate common characteristics among multiple subgraphs that can be included within the dataset of graph structured data. This assignment can allow a user to associate a particular graph analytic algorithm with the first general characteristic, thereby improving the efficiency of the selection of a graph analytic algorithm for analysis of other datasets. For example, in some instances, a second dataset of graph structured data can have a second general characteristic similar to the first general characteristic. Based on the graph analytic algorithms assigned to the first general characteristic, it can be determined which graph analytic algorithms are effective for the second dataset of graph structured data. Thus, in accordance with some embodiments of the present disclosure, by ranking graph analysis algorithms and assigning graph analysis algorithms to general characteristics of a dataset of graph structured data, the efficiency and interpretability of analysis of a dataset of graph structured data can be improved.

[0016] 1 illustrates an example process flow 100 for classifying graph analysis algorithms in accordance with at least one embodiment of the present disclosure. In some embodiments, the process 100 can include a characterization module 120 configured to obtain a dataset 110 of graph structured data.

[0017] In some embodiments, the dataset 110 can provide a visual interpretation of the data and represent relationships between different variables of the dataset 110. For example, the dataset 110 can include a collection of financial transactions. The dataset 110 can visualize financial transactions involving different people or entities. In some embodiments, the dataset 110 can include one or more subgraphs. For example, each collection of financial transactions can be represented as a subgraph in the dataset 110. For example, an exemplary subgraph can include specific individuals and / or entities involved in the financial transactions as nodes. The subgraphs can represent different financial transactions involving the specific individuals and / or entities, with each transaction represented by an edge. For example, the edges can include purchases, refunds, payments, etc., among others. In some embodiments, the dataset 110 can be organized into one or more subgraphs. For example, the dataset 110 can include financial transactions related to a bank. In these examples, the dataset 110 can be organized into one or more subgraphs according to different parameters. For example, each of the one or more subgraphs can include a type of transaction involving a bank, such as a refund or a deposit. In another example, one or more subgraphs may each include transactions with a different type of customer, such as an individual or a business.

[0018] In some embodiments, one or more graph analysis algorithms may be obtained. The one or more graph analysis algorithms may include any analysis technique of graph structured data that may be used to represent one or more aspects of a dataset of graph structured data. For example, the one or more selected graph analysis algorithms may include predicting behavioral patterns, identifying significant nodes and edges, and / or identifying anomalous characteristics of the dataset 110. In some embodiments, the one or more graph analysis algorithms may include publicly available graph analysis algorithms. For example, in some embodiments, the one or more graph analysis algorithms may be obtained from a public repository and / or a public library that includes a collection of graph analysis algorithms. For example, NetworkX, a Python library that includes implementations of hundreds of graph analysis algorithms, may be used to obtain the one or more graph analysis algorithms. Any other suitable repository and / or library that includes a collection of graph analysis algorithms may be used.

[0019] In some embodiments, a subset of graph analysis algorithms may be selected from the set of graph analysis algorithms. For example, the one or more graph analysis algorithms may include a portion of the set of graph analysis algorithms. In some embodiments, the selection may be made based on the process 100 being configured to categorize the one or more graph analysis algorithms. For example, the process 100 may predetermine which of the one or more graph analysis algorithms is more effective for the data set 110. In some embodiments, the predetermination may be made based on different characteristics of the data set 110. For example, if the data set 110 has a large number of nodes, a graph analysis algorithm associated with the nodes may be selected. In another example, the data set 110 may have only a small number of nodes. In these examples, a graph analysis algorithm such as node index analysis may not be applicable. In some embodiments, the selection may be made based on the types of graph analysis algorithms available. For example, a certain number of algorithms of each type may be included. For example, five graph analysis algorithms associated with each type such as shortest path, clique detection, coloring, and clustering may be selected.

[0020] In some embodiments, types of graph analysis algorithms may include community detection (which may be configured to evaluate how groups of nodes are clustered), similarity (which may be configured to determine a degree of similarity between data in the dataset 110, such as between one or more subgraphs), centrality (which may be configured to identify central nodes), heuristic link prediction (which may be configured to predict future potential links based on a score function), coloring algorithms (which may be configured to assign different color labels to each vertex), and pathfinding and exploration (which may be configured to explore paths between nodes), among other types of graph analysis algorithms.

[0021] In some embodiments, the dataset 110 can be analyzed using a graph neural network (GNN). For example, a machine-learning (ML) model can be trained using the GNN. To analyze the graph-structured data, an ML model can be trained on the dataset 110. For example, the ML model can be trained to generate GNN predictions at the node, edge, and graph levels of the graph-structured data. For example, the GNN predictions performed by the ML model can include node labeling, node and edge predictions, clustering, among others. In some embodiments, different GNN architectures can be used to train the ML model. For example, a graph isomorphism network (GIN), a graph convolutional network (GCT), and / or a graph attention network (GAT) can be used. In some embodiments, any other suitable GNN architecture can be used.

[0022] In some embodiments, the GNN predictions can be evaluated to detect reliability, potential bias, and / or errors. Additionally or alternatively, the decision-making process of the GNN predictions can be detected. In some embodiments, one or more explanation methods can be utilized to explain the decision-making process. For example, the one or more explanation methods can include SA, Guided BP, CAM, Grad-CAM, GNNExplainer, PGExplainer, GraphMask, ZORRO, Causal Screening, SubgraphX, LRP, Excitation BP, GNN-LRP, PGM-Explainer, XGNN, among others. In some embodiments, the one or more explanation methods can include various design elements that characterize the one or more explanation methods. Some examples of design elements can include type, learning (e.g., presence or absence of a learning procedure), task (e.g., tasks to which each method can be applied), target (e.g., subject of explanation), black box (whether the GNN is treated as a block box during explanation), flow (computational flow for explanation), and design (whether the explanation method has a specific design for graph-structured data). In these and other embodiments, each of the one or more illustrative methods can include one or more design elements.

[0023] In some embodiments, process 100 can include a correlation module 140 configured to obtain GXAI analysis results 130 and graph analysis results 135. For example, GXAI analysis results 130 can be obtained from one or more description methods, and graph analysis results 135 can be obtained from one or more graph analysis algorithms. In some embodiments, GXAI analysis results 130 and graph analysis results 135 can be calculated by process 100. In some embodiments, GXAI analysis results 130 and graph analysis results 135 can be provided from another system or can be retrieved from a data storage device.

[0024] In some embodiments, the GXAI analysis results 130 can include a contribution score. For example, analyzing a particular dataset of graph-structured data using a particular GXAI analysis technique can result in a contribution score corresponding to each node or edge included in the particular dataset, which indicates how each node or edge contributes to a particular classification or regression analysis of the particular dataset (i.e., a particular edge or a particular node has a stronger influence on the GXAI analysis).

[0025] In some embodiments, the contribution scores can be stored as a vector representation, with each element of the vector corresponding to each edge or node of a particular data set of graph structured data. A larger contribution score for a particular edge or a particular node may indicate that the particular edge or the particular node provides a greater contribution to the classification or regression results of the GXAI analysis (i.e., the particular edge or the particular node has a stronger influence on the GXAI analysis). In these and other embodiments, the elements of the vector representation of the GXAI analysis (i.e., the contribution score of each node or edge) can be normalized to facilitate comparison of the vector representation of the GXAI analysis with the results of various graph analysis algorithms described below.

[0026] In some embodiments, the graph analysis results 135 may include insights and explanations regarding the graph structured data dataset 110. For example, the graph analysis results 135 may include the results of applying a graph analysis algorithm to the dataset 110.

[0027] In these and other embodiments, GXAI analysis results 130 and graph analysis results 135 may be stored as one or more vector representations of an analysis of the graph structured data. For example, a particular analysis based on the nodes of a particular data set of graph structured data may generate a node vector, where each element of the node vector corresponds to an analysis result associated with a particular node of the particular data set. In another example, a particular analysis based on the edges of a particular data set of graph structured data may generate an edge vector, where each element of the edge vector corresponds to an analysis result associated with a particular edge of the particular data set.

[0028] The correlation module 140 can determine a correlation score 145 based on the GXAI analysis result 130 and the graph analysis result 135. The correlation module 140 can, for example, obtain vector representations corresponding to each of the GXAI analysis result 130 and the graph analysis result 135 and determine a degree of similarity between the vector representations. For example, a Pearson correlation coefficient can be calculated based on the vector representations. Additionally or alternatively, a cosine similarity between the vector representations can be determined. In this and other examples, the degree of similarity between the GXAI analysis result 130 and the graph analysis result 135 can be represented as a correlation score 145. In some embodiments, the correlation score 145 can be represented in a tabular form. For example, the correlation module 140 can generate a correlation table including at least a first column including a graph analysis algorithm and a second column including a correlation score 145 corresponding to the graph analysis algorithm.

[0029] In some embodiments, the process 100 can include a ranking module 150 configured to obtain the correlation scores 145 and the graph analysis algorithms. In these and other embodiments, the ranking module 150 can be configured to generate a ranked list 155 of graph analysis algorithms based on the correlation scores 145. For example, in some embodiments, the graph analysis algorithms can be listed in descending order of the corresponding correlation scores. For example, the ranking module 150 can generate a ranked list 155 having at least a first column and a second column. In some embodiments, the first column includes the graph analysis algorithms and the second column includes a rank of the corresponding graph analysis algorithm in the first column. In these and other embodiments, the ranking module 150 can be configured to sort the obtained graph analysis algorithms according to the correlation scores 145. For example, the graph analysis algorithms can be sorted in descending order of the correlation scores 145. In some embodiments, the graph analysis algorithms included in the ranked list 155 can be ranked according to an order and sorted according to the correlation scores 145. For example, the graph analysis algorithm with the highest correlation score can be given the highest rank (e.g., 1). Each of the graph analysis algorithms in the ranked list 155 may be ranked. In some embodiments, the generation of the ranked list 155 may be illustrated and described in more detail with respect to FIG. 4 of this disclosure.

[0030] In some embodiments, the process 100 may include a characterization module 120 configured to obtain the dataset 110. In some embodiments, the characterization module 120 may be configured to determine a general characteristic 125 of the dataset 110. The general characteristic 125 may indicate a similarity between one or more subgraphs included in the dataset 110. In some embodiments, the similarity between the one or more subgraphs may include structural similarity between the one or more subgraphs, such as graph topology, node connectivity, and node neighborhood. For example, a clustering algorithm may be applied to the dataset 110 to determine the general characteristic 125 of the dataset 110. For example, the clustering algorithm may divide the nodes in the one or more subgraphs into groups or clusters. The division may be based on topological criteria, such as the graph structure and / or the position of the nodes on the graph. For example, k-means clustering may be used, which may be configured to identify multiple centroids and match each node to the closest centroid. The centroids may be optimized by determining the average value of the nodes connected to each centroid. For example, the centroids may be optimized such that the nodes connected to each centroid are at a similar distance from the centroid. In some embodiments, the general characteristics 125 of the dataset 110 may include one or more of: a number of centroids, an average distance of a node from the centroid, and a number of nodes connected to the centroid.

[0031] In some embodiments, the general characteristic 125 may be represented numerically. For example, in some embodiments, the general characteristic 125 may be represented as a vector. In these and other embodiments, each of the one or more subgraphs may be converted into a vector representation. The vector representation may be generated using any suitable graph embedding technique. For example, the graph embedding technique may include any technique that may map each of the one or more subgraphs into a fixed-length vector that captures characteristics of the one or more subgraphs. The general characteristic 125 may be determined by calculating an average of the vector representations of each of the one or more subgraphs.

[0032] In some embodiments, the process 100 may include an assignment module 160. The assignment module 160 may be configured to obtain the general characteristics 125 and the ranked list 155. In some embodiments, the assignment module may be configured to generate an assignment table 165 that assigns or associates the general characteristics 125 with one or more graph analysis algorithms listed in the ranked list 155. For example, the best performing graph analysis algorithm may be assigned to the general characteristics 125. For example, the best performing graph analysis algorithm may be the graph analysis algorithm with the highest correlation score that can be derived from the ranked list 155. In some embodiments, the assignment table 165 may include at least a characteristic column and an algorithm column. In some embodiments, the characteristic column may include the general characteristics 125. In these and other embodiments, the algorithm column may include one or more graph analysis algorithms. For example, the algorithm column may include a first column of the ranked list 155. For example, the graph analysis algorithms may be listed in the algorithm column in ranked order based on the correlation score 145.

[0033] In some embodiments, a threshold number of graph analysis algorithms may be included in the allocation table 165. For example, in some embodiments, the threshold number may define the number of graph analysis algorithms to be included in the allocation table 165. For example, the number of graph analysis algorithms with the highest rank may be included in the allocation table 165. In some embodiments, the threshold number may be provided by a user (e.g., a data scientist) attempting to analyze the dataset 110. In some embodiments, the threshold number may be determined based on the size of the dataset 110. For example, a larger dataset may have a higher threshold number. In yet another embodiment, the threshold number may be determined based on the number of graph analysis algorithms listed in the ranked list 155. For example, the threshold number may be 10% of the total number of graph analysis algorithms in the ranked list 155.

[0034] In some embodiments, the allocation table 165 can be obtained by the GUI module 170, which can be configured to cause the GUI 175 to present information in a particular manner. For example, the GUI module 170 can be configured to instruct the GUI 175 to present the allocation table 165. Additionally or alternatively, the GUI module 170 can be configured to obtain and display the correlation scores 145. For example, the GUI 175 can include a drop-down menu (or any other graphical selection tool) with which a user of the GUI 175 can interact to select one or more subgraphs in the dataset 110. The subgraph summary can include summary information regarding the particular subgraph selected via the drop-down menu. For example, the subgraph summary can include a GXAI explainability score (e.g., an element of the contribution score associated with the particular subgraph), the number of nodes included in the particular subgraph, the number of edges included in the particular subgraph, or any other statistics associated with the particular subgraph. In these and other embodiments, the subgraph summary can further include graph analysis results generated by a particular graph analysis algorithm.

[0035] In some embodiments, the GUI 175 may display a summary of various statistics associated with the dataset 110. For example, the summary may include the number of subgraphs included in the dataset 110, the average number of nodes included with each of the subgraphs, the average number of edges associated with each of the subgraphs, the training accuracy of the machine learning model from which the dataset 110 was obtained, the testing accuracy of the machine learning model from which the dataset 110 was obtained, or any other statistics associated with the particular dataset. In some embodiments, the summary of various statistics associated with the dataset 110 may assist a user in identifying which graph analysis algorithm is most suitable for the dataset 110. For example, one or more graph analysis algorithms to be applied to the dataset 110 may be selected based on the presented summary. For example, the type graph analysis algorithm may be selected based on the type of data present in the dataset 110. In some embodiments, the summary may assist a user in determining a threshold number of graph analysis algorithms to be included in the assignment table 165. For example, a user may determine a higher threshold number for a more complex dataset.

[0036] Modifications, additions, or omissions may be made to process 100 without departing from the scope of this disclosure. For example, process 100 may include more or fewer elements than shown and described in this disclosure.

[0037] 2 illustrates a flowchart of an example method 200 for classifying graph analysis algorithms in accordance with at least one embodiment of the present disclosure. Method 200 may be performed by any suitable system, apparatus, or device. For example, computing system 500 of FIG. 5 may perform one or more of the operations associated with method 200. Although illustrated by separate blocks, steps and operations associated with one or more of the blocks of method 200 may be divided into additional blocks, combined into fewer blocks, or eliminated, depending on the particular implementation.

[0038] The method 200 may include, at block 202, obtaining a dataset of graph structured data. In some embodiments, the dataset may include one or more subgraphs. Each of the one or more subgraphs may include one or more nodes and one or more edges. In some embodiments, the one or more subgraphs may share one or more nodes. In these and other embodiments, the one or more subgraphs may not share one or more edges.

[0039] At block 204, a correlation may be determined between the analysis of the dataset of graph-structured data using GXAI techniques and the analysis using one or more graph analysis algorithms. In some embodiments, the GXAI analysis may include any machine learning-assisted or artificial intelligence-assisted graph analysis technique that may analyze one or more aspects of the dataset of graph-structured data. Additionally or alternatively, the machine learning module may be trained using a training dataset of graph-structured data that includes one or more subgraphs that already include labels or annotations for the subgraphs and patterns identified in the GNN. For example, the GXAI analysis may be performed by a computing system that includes a machine learning module configured to perform a clustering algorithm or pattern recognition routine using a graph neural network or any other GXAI algorithm. The GXAI algorithm may be configured to analyze the topology of one or more datasets of graph-structured data and make predictions based on the analyzed topology. In some embodiments, the specifications of the GXAI algorithm may be determined using one or more explainers. For example, the one or more explainers may describe the type of explanation provided, whether a learning procedure is included, to what tasks each explainer is applicable, the target of the explanation, the computational flow or explanation, and whether the explanation method has a specific design for graph data.

[0040] In some embodiments, the GXAI analysis may determine edge-based or node-based contribution scores. The contribution scores may be stored as a vector representation, with each element of the vector corresponding to each edge or node of a particular dataset of graph-structured data. A larger contribution score for a particular edge or a particular node may indicate that the particular edge or the particular node provides a greater contribution to the classification or regression results of the GXAI analysis (i.e., the particular edge or the particular node has a stronger influence on the GXAI analysis). In these and other embodiments, the elements of the vector representation of the GXAI analysis (i.e., the contribution score of each node or edge) may be normalized to facilitate comparison of the vector representation of the GXAI analysis with the results of various graph analysis algorithms described below.

[0041] In some embodiments, the graph analysis algorithm may include any analysis technique for graph structured data used to represent one or more aspects of a dataset of graph structured data. The graph analysis algorithm may analyze the nodes of the graph structured data included in the dataset. Additionally or alternatively, the graph analysis algorithm may analyze the edges of the graph structured data included in the dataset. An example of a particular graph analysis algorithm may include network motif analysis that identifies one or more patterns of nodes and edges occurring in the dataset of graph structured data, which patterns occur more frequently than node and edge configurations present in a randomized network of graph structured data.

[0042] Another example of a particular graph analysis algorithm may include a node-based PageRank analysis in which each node of a dataset of graph-structured data is assigned a value indicating the influence each node has on the dataset. The influence of a particular node may be determined, for example, according to the number of edges connecting the particular node to other nodes. In a PageRank analysis, the values ​​determined for each node may be represented as a vector, with each entry of the vector corresponding to a node of the dataset of graph-structured data. Additionally or alternatively, the values ​​determined by the PageRank analysis may be normalized (e.g., by dividing all values ​​by a maximum value), thereby allowing the results of the PageRank analysis, represented by the normalized vector, to be more easily compared with the results of other graph analysis algorithms or the results of GXAI analysis techniques.

[0043] Another example of a particular graph analysis algorithm may include edge-based shortest path analysis, in which each edge of a data set of graph structured data is assigned a value indicative of the influence each node has on the data set. The influence of a particular edge may be determined, for example, based on the number of times that a particular edge appears in the shortest path between any two nodes of the graph structured data. Each edge may be represented as an entry in a vector, and the value of the entry may indicate the number of times that the edge is included in the shortest path between the two nodes. Additionally or alternatively, the vector values ​​determined by the shortest path analysis may be normalized to facilitate comparison between the shortest path vector and any other vector, such as a PageRank vector.

[0044] In some embodiments, the correlation between the first result and the second result may include a comparison of a first vector corresponding to the first result (e.g., a vector representation of the result of a GXAI analysis) with a second vector corresponding to the second result (e.g., a vector representation of the result of a graph analysis algorithm). The comparison between the first vector and the second vector may be accomplished using any vector comparison technique. In some embodiments, the correlation may include a similarity score based on the first vector and the second vector. For example, in these and other embodiments, the similarity score may include a cosine similarity or a Pearson correlation coefficient. For example, using cosine similarity, a similarity score of 0 may indicate no correlation and a similarity score of 1 may indicate a strong correlation. For the Pearson correlation coefficient, a similarity score of 0 may indicate no correlation and a negative similarity score or a positive similarity score may indicate a strong correlation.

[0045] At block 206, a ranked list of one or more graph analysis algorithms may be generated based on the correlation. In some embodiments, the correlation and the one or more graph analysis algorithms may be obtained. The ranked list may list the graph analysis algorithms in order of correlation. For example, the one or more graph analysis algorithms may be listed in order of decreasing correlation.

[0046] In some embodiments, the ranked list can be generated as a table. For example, the table can include a first column that includes one or more graph analysis algorithms. Additionally or alternatively, the table can include a second column that includes the correlations. For example, the correlation scores can be sorted in the table in a particular order. For example, the correlation scores can be sorted in descending order. In some embodiments, the correlation scores can be sorted in ascending order. In these and other embodiments, the one or more graph analysis algorithms can be sorted in the table according to their corresponding correlation scores.

[0047] At block 208, a general characteristic of the dataset of graph structured data may be determined. In some embodiments, the general characteristic may indicate a similarity between one or more subgraphs contained within the dataset. In some embodiments, the general characteristic may be determined numerically. For example, the general characteristic may be expressed as an average vector indicating the similarity between the one or more subgraphs. For example, a vector for each of the one or more subgraphs may be generated. The vector may be generated using any suitable technique, such as Graph2Vec. The average vector may be generated by calculating the average of the vectors for each of the one or more subgraphs.

[0048] At block 210, a threshold number of the best performing graph analysis algorithms may be assigned to the general property. In some embodiments, the assignment may be based on a ranked list of at least one or more graph analysis algorithms. For example, the ranked list may list one or more graph analysis algorithms in descending order of correlation score. In these and other embodiments, the threshold number may determine a particular number of graph analysis algorithms to be assigned. For example, a particular number of graph analysis algorithms having the highest correlation scores may be assigned to the general property.

[0049] In some embodiments, the threshold number can define a correlation score threshold. For example, the threshold number can define a particular correlation score that must be exceeded. For example, any number of graph analysis algorithms that have a correlation score above a particular correlation score set by the threshold number can be assigned to the general characteristic.

[0050] At block 212, an assignment table may be generated that includes the general properties and a threshold number of best performing graph analysis algorithms. For example, the assignment table may include at least a first column and a second column. The first column may include the general properties and the second column may include the graph analysis algorithms assigned to the general properties.

[0051] At block 214, a display in the GUI may be generated that visualizes the allocation table. For example, the GUI may be configured to display an allocation table where a first column includes general characteristics and a second column includes graph analysis algorithms. In some embodiments, the GUI may display the allocation table in different formats. For example, the GUI may display the allocation table as a list or any suitable graphical representation. Additionally or alternatively, the GUI may be configured to display a summary of various statistics associated with the dataset of graph structured data. In some embodiments, displaying the allocation table in the GUI may include one or more operations as described with respect to FIG. 1 with respect to displaying the allocation table 165 on the GUI 175.

[0052] Modifications, additions, or omissions may be made to method 200 without departing from the scope of the present disclosure. For example, the operations of method 200 may be performed in a different order. Additionally, or alternatively, two or more operations may be performed simultaneously. Furthermore, the outlined steps and operations are provided as examples, and some of the steps and operations may be optional, combined into fewer steps and operations, or expanded into additional steps and operations without diminishing the essence of the described embodiments.

[0053] For example, method 200 may further include obtaining a second dataset of graph structured data and determining a second general characteristic of the second dataset of graph structured data. The second general characteristic may be compared to the general characteristic. In response to determining that the second general characteristic is similar to the general characteristic, a second assignment table may be generated that includes the second general characteristic and a threshold number of graph analysis algorithms that may be searched from the assignment table.

[0054] In some embodiments, the second general characteristic may be dissimilar to the general characteristic. In these and other embodiments, the second ranked list of graph analytics algorithms may be generated based on a second correlation between an analysis of a second dataset of graph structured data using GXAI techniques and an analysis using the graph analytics algorithms.

[0055] 3 illustrates a flowchart of an example method for selecting a graph analysis algorithm for a dataset, in accordance with at least one embodiment of the present disclosure. Method 300 may be performed by any suitable system, apparatus, or device. For example, computing system 500 of FIG. 5 may perform one or more of the operations associated with method 300. Although illustrated by separate blocks, steps and operations associated with one or more of the blocks of method 300 may be divided into additional blocks, combined into fewer blocks, or eliminated, depending on the particular implementation.

[0056] At block 302, a dataset of graph structured data may be obtained. In some embodiments, the dataset of graph structured data may include one or more subgraphs. The one or more subgraphs may be graphs in the dataset of graph structured data. For example, the one or more subgraphs may each include a subset of vertices and edges that are part of the dataset. In some embodiments, the dataset may be provided by a user. For example, a user may specify which dataset is to be analyzed. In some embodiments, the dataset may be retrieved from a data store. Additionally or alternatively, the dataset may be provided from another system. The dataset may include any type of data that can be represented in a graph format. For example, in some embodiments, the dataset may include a collection of financial transactions. Each collection may be represented as a subgraph of the dataset. For example, a node may represent a person or entity involved in a financial transaction, and an edge may represent a particular transaction between two nodes. In some embodiments, an edge may be represented as an arrow having a direction. For example, an edge may point from a particular node to another node. In some embodiments, an edge may be represented as a connection between nodes without a particular direction.

[0057] At block 304, it may be determined whether the general characteristics of the dataset are known. For example, in some embodiments, a user providing the dataset may be aware of the general characteristics of the dataset. In some embodiments, the general characteristics of the dataset may be determined in advance. In some embodiments, the general characteristics may be determined by another system. In some embodiments, the dataset may include a general characteristic associated with the dataset. For example, the general characteristic may be provided with the dataset. In response to determining that the general characteristics are known, the method 300 may proceed to block 306.

[0058] At block 306, a general characteristic of the dataset may be retrieved. For example, in some embodiments, the general characteristic may be retrieved from a data store. As another example, the general characteristic may be provided by a user providing the dataset. In some embodiments, the general characteristic may accompany the dataset.

[0059] In some embodiments, the general characteristics of the dataset may not be known. For example, the dataset may be an unknown dataset where the general characteristics may not have been previously determined. In these cases, the method 300 may proceed from block 304 to block 308. At block 308, the general characteristics of the dataset may be determined. In some embodiments, the general characteristics may indicate a similarity between one or more subgraphs in the dataset.

[0060] In some embodiments, the similarity between the one or more subgraphs may include structural similarity between the one or more subgraphs, such as graph topology, node connectivity, and node neighborhood. In these and other embodiments, a clustering algorithm may be used to determine the general characteristics. For example, a clustering algorithm may be used to divide the nodes in the one or more subgraphs into groups or clusters. The division may be based on topological criteria, such as the graph structure and / or the location of the nodes on the graph. Continuing with the example of a dataset that includes a collection of financial transactions, a particular cluster may include nodes connected to a particular node that represents a bank. For example, the cluster may include all individuals or entities that have direct financial transactions with the bank. In some examples, a particular cluster may be divided into separate clusters that represent incoming transfers at the bank and outgoing transfers from the bank. For example, a cluster may include all nodes that deposit money into the bank and another cluster may include all nodes that withdraw money from the bank. In some embodiments, any suitable clustering algorithm may be used. For example, the Markov Cluster algorithm, Molecular Complex Detection, Highly Connected Subgraphs, and Restricted Neighborhood Search Clustering, among others, may be used to determine the general characteristics of the dataset.

[0061] In some embodiments, the general property may be presented in vector form. For example, the general property may be presented as an average vector that averages one or more vectors corresponding to one or more subgraphs. In these and other embodiments, the one or more subgraphs may be converted into one or more vectors using any suitable graph embedding method, such as Graph2Vec. In these and other embodiments, the one or more vectors may be configured to be the same length. For example, the one or more vectors may be configured to be a predetermined fixed length.

[0062] At block 310, it may be determined whether the general characteristics of the data set are similar to existing general characteristics of a previous data set. In some embodiments, the previous data set may include one or more data sets for which a corresponding graph analysis algorithm was determined according to process 100 of FIG. 1. In some embodiments, the existing general characteristics may be stored in a data storage device or memory. In these and other embodiments, the general characteristics may be compared to each of the existing general characteristics. For example, in some embodiments, a comparison score may be determined corresponding to a comparison between the general characteristics and each of the existing general characteristics. In some embodiments, the general characteristics and the existing general characteristics may each be a vector representation. In these and other embodiments, the comparison between the general characteristics and the existing general characteristics may include any vector comparison metric, such as cosine similarity. For example, the comparison score may be represented as a value between 0 and 1, with 0 indicating no similarity and 1 indicating strong similarity. In some embodiments, a similarity threshold may be applied to the comparison score. For example, the similarity threshold may be a number between 0 and 1. If the comparison score meets the similarity threshold, the general characteristics may be determined to be similar to the corresponding existing general characteristics. The comparison score may satisfy the similarity threshold by being greater than or equal to the similarity threshold.

[0063] In some embodiments, the general characteristic may be similar to one or more existing general characteristics. In these cases, method 300 may proceed to block 312. In block 312, one or more ranked lists of graph analysis algorithms associated with similar existing general characteristics may be retrieved. The one or more ranked lists may each include one or more graph analysis algorithms ranked according to a corresponding correlation score. For example, the one or more ranked lists may have been generated with respect to a previous data set. In some embodiments, the one or more ranked lists may have been generated in a manner similar to ranked list 155 of FIG. 1 and / or method 400 of FIG. 4. In some embodiments, the graph analysis algorithms present in the one or more ranked lists may be associated with the general characteristic. For example, one or more ranked lists may be assigned to the general characteristic in the same manner that one or more ranked lists were assigned to the existing general characteristics. In some embodiments, the graph analysis algorithms present in the one or more ranked lists may include graph analysis algorithms suitable for analyzing the data set. For example, the graph analysis algorithm with the highest rank (e.g., 1) may be most suitable for the data set. In these and other embodiments, the ranked list may provide the system and / or user with a list of graph analysis algorithms that may be selected to analyze the dataset.

[0064] In some embodiments, the general characteristics may not be similar to any of the existing general characteristics. For example, none of the comparison scores may meet a similarity threshold. In these and other embodiments, the method 300 may proceed to block 314, where the data set may be used to process a graph analysis algorithm. For example, the method 200 of FIG. 2 may be performed using the data set as input.

[0065] Modifications, additions, or omissions may be made to method 300 without departing from the scope of the present disclosure. For example, the operations of method 300 may be performed in a different order. Additionally, or alternatively, two or more operations may be performed simultaneously. Furthermore, the outlined steps and operations are provided as examples, and some of the steps and operations may be optional, combined into fewer steps and operations, or expanded into additional steps and operations without diminishing the essence of the described embodiments.

[0066] 4 illustrates a flowchart of an example method for generating a rank table for a graph analysis algorithm, in accordance with at least one embodiment of the present disclosure. Method 400 may be performed by any suitable system, apparatus, or device. For example, ranking module 150 of FIG. 1 and / or computing system 500 of FIG. 5 may perform one or more operations associated with method 400. Although illustrated by separate blocks, steps and operations associated with one or more of the blocks of method 400 may be divided into additional blocks, combined into fewer blocks, or eliminated, depending on the particular implementation. In some embodiments, method 400 may correspond to operations that ranking module 150 may take to generate ranked list 155 of FIG. 1.

[0067] At block 401, a correlation table may be obtained that includes at least the graph analysis algorithms and correlation scores corresponding to the graph analysis algorithms. For example, the correlation table may correspond to the correlation table generated by correlation module 140 of FIG. 1. For example, a first column may include the graph analysis algorithms and a second column may include the corresponding correlation scores. In some embodiments, the correlation scores may correspond to correlation scores 145 of FIG. 1.

[0068] At block 402, a rank table may be generated having at least a first column and a second column. In some embodiments, the first column may include graph analysis algorithms and the second column may include ranks corresponding to the graph analysis algorithms in the first column. In some embodiments, the first column may be initialized with all the graph analysis algorithms analyzed. In these and other embodiments, the first column may be initialized in any order. For example, the first column may be initialized in the order in which the graph analysis algorithms were obtained. In these and other embodiments, the correlation table and the rank table may include the same list of graph analysis algorithms. In some embodiments, the graph analysis algorithms in the rank table may be initialized in the same order as the graph analysis algorithms in the correlation table. In some embodiments, the second column may be initialized to 0. For example, the graph analysis algorithms may not be ranked.

[0069] At block 404, it may be determined whether any graph analysis algorithms remain in the correlation table. For example, a graph analysis algorithm may be described as remaining in the correlation table if the graph analysis algorithm is not ranked in the rank table. For example, a graph analysis algorithm may not be ranked in the rank table if the second column corresponding to the graph analysis algorithm remains at 0. In response to determining that no graph analysis algorithms remain, the method 400 may proceed to block 406. At block 406, the rank table may be determined and output.

[0070] In response to determining that at least one graph analysis algorithm remains in the correlation table, method 400 may proceed to block 408. At block 408, the graph analysis algorithms in the correlation table may be sorted according to their corresponding correlation scores. For example, the graph analysis algorithms may be sorted in descending order of correlation score.

[0071] At block 410, a rank may be added to a second column of the rank table corresponding to the graph analysis algorithm in the first column. For example, the graph analysis algorithm at the top of the sorted correlation table may be given a rank of 1. For example, the corresponding graph analysis algorithm may be placed in the rank table and the corresponding 0 in the second column of the rank table may be updated with the given rank. The second graph analysis algorithm in the correlation table may be given a rank of 2. In some embodiments, the process of giving ranks to the graph analysis algorithms in the rank table may be repeated until all graph analysis algorithms have been given ranks. For example, method 400 may return to block 404 until no graph analysis algorithms remain in the correlation table. In some embodiments, after each graph analysis algorithm has been given a rank, the rank table may be sorted according to rank. For example, the rank table may be sorted such that the graph analysis algorithm with the highest rank (e.g., 1) is at the top of the list. Modifications, additions, or omissions may be made to method 400 without departing from the scope of the disclosure. For example, the operations of method 400 may be performed in a different order. Additionally or alternatively, two or more operations may be performed simultaneously. Additionally, the outlined steps and operations are provided as examples, and some of the steps and operations may be optional, combined into fewer steps and operations, or expanded into additional steps and operations without detracting from the essence of the described embodiments.

[0072] 5 is a block diagram illustrating an example system 500 that can be used for optical signal amplification, in accordance with at least one embodiment of the present disclosure. The system 500 includes a processor 510, a memory 512, a communication unit 516, a display 518, and a user interface unit 520, all of which can be communicatively coupled. In some embodiments, the system 500 can be used to perform one or more of the methods described in this disclosure. For example, the system 500 can be used to aid in performing the method described in FIG. 1. For example, the system 500 can be used to classify graph analysis algorithms, in accordance with at least one embodiment of the present disclosure.

[0073] In general, the processor 510 may include any suitable special purpose or general purpose computer, computing entity, or processing device, including various computer hardware or software modules, and may be configured to execute instructions stored on any suitable computer-readable storage medium. For example, the processor 510 may include a microprocessor, a microcontroller, a parallel processor such as a graphics processing unit (GPU) or a tensor processing unit (TPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or any other digital or analog circuitry configured to interpret and / or execute program instructions and / or process data.

[0074] 5, it is understood that processor 610 may include any number of processors distributed across any number of networks or physical locations and configured to individually or collectively perform any number of operations described herein. In some embodiments, processor 510 may interpret and / or execute program instructions and / or process data stored in memory 512. In some embodiments, processor 510 may execute program instructions stored in memory 512.

[0075] For example, in some embodiments, the processor 510 may execute program instructions stored in memory 512 associated with performing a task such that the system 500 performs or directs the performance of operations associated therewith as directed by the instructions. In these and other embodiments, the instructions may be used to perform one or more blocks of the methods 200, 300, 200 of FIGS. 2-4 and / or the process flows described in FIG.

[0076] Memory 512 may include a computer-readable storage medium or one or more computer-readable storage media for carrying or having computer-executable instructions or data structures stored thereon. Such computer-readable storage media may be any available media that can be accessed by a general-purpose or special-purpose computer, such as processor 510.

[0077] By way of example, and not limitation, such computer-readable storage media may include non-transitory computer-readable media including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage, flash memory devices (e.g., solid-state memory devices), or any other storage medium that can be used to carry or store specific program code in the form of computer-executable instructions or data structures and that can be accessed by a general-purpose or special-purpose computer. Combinations of the above may also be included within the scope of computer-readable storage media.

[0078] Computer-executable instructions may include, for example, instructions and data configured to cause processor 510 to perform a particular operation or group of operations, as described in this disclosure. In these and other embodiments, the term "non-transitory" as described in this disclosure should be considered to exclude only those types of transitory media that are not within the scope of patentable subject matter in Federal Circuit decision of In re Nuijten, 500 F.3d 1346 (Fed. Cir. 2007). Combinations of the above may also be included within the scope of computer-readable media.

[0079] The communication unit 516 may include any component, device, system, or combination thereof configured to transmit or receive information over a network. In some embodiments, the communication unit 516 may communicate with other locations, co-located devices, or other components within the same system. For example, the communication unit 516 may include a modem, a network card (wireless or wired), an optical communication device, an infrared communication device, a wireless communication device (e.g., an antenna), and / or a chipset (e.g., a Bluetooth device, an 802.6 device (e.g., a metropolitan area network (MAN)), a WiFi device, a WiMax device, a cellular communication facility, etc.), etc. The communication unit 516 may enable data exchange with a network and / or any other device or system described in this disclosure.

[0080] Display 518 may be configured as one or more displays, such as an LCD, LED, Braille, or other type of display. Display 518 may be configured to present video, text captions, a user interface, and other data as directed by processor 510.

[0081] The user interface unit 520 may include any device that allows a user to interface with the system 500. For example, the user interface unit 520 may include a mouse, a trackpad, a keyboard, buttons, a camera, and / or a touch screen, among other devices. The user interface unit 520 may receive input from a user and provide input to the processor 510. In some embodiments, the user interface unit 520 and the display 518 may be combined.

[0082] Modifications, additions, or omissions may be made to system 500 without departing from the scope of the disclosure. For example, in some embodiments system 500 may include any number of other components that may not be illustrated or described. Further, depending on the particular implementation, system 600 may not include one or more of the components illustrated and described.

[0083] As mentioned above, the embodiments described herein may involve the use of a special purpose or general purpose computer (e.g., processor 510 of FIG. 5) including various computer hardware or software modules, as discussed in more detail below. Further, as mentioned above, the embodiments described herein may be implemented using a computer-readable medium (e.g., memory cell 512 of FIG. 5) having computer-executable instructions or data structures stored thereon.

[0084] In accordance with common practice, various features shown in the drawings may not be drawn to scale. The figures shown in this disclosure are not intended to be actual views of a particular device (e.g., apparatus, system, etc.) or method, but are merely idealized representations used to explain various embodiments of the disclosure. Thus, dimensions of various features may be arbitrarily expanded or reduced for clarity. In addition, some of the drawings may be simplified for clarity. Thus, the drawings may not show all components of a given device (e.g., apparatus) or all operations of a particular method.

[0085] Terms used in this specification, and particularly in the appended claims (e.g., the body of the appended claims), are generally intended as "broad" terms (e.g., the term "including" should be interpreted as "including, but not limited to," the term "having" should be interpreted as "having, but not limited to," etc.).

[0086] Moreover, where the recitation of a particular number of introduced claims is intended, such intent is expressly set forth in the claim, and in the absence of such recitation, no such intent exists. For example, as an aid to understanding, the following appended claims may include the use of the introductory phrases "at least one" and "one or more" to introduce the recitation of claims. However, the use of such phrases should not be construed to mean that the introduction of a claim recitation with the indefinite article "a" or "an" limits any particular claim that includes such an introduced recitation of claims to an embodiment that includes only one such recitation, even when the same claim includes the introductory phrase "one or more" or "at least one" and the indefinite article "a" or "an" (e.g., "a" and / or "an" should be interpreted to mean "at least one" or "one or more"). In other words, the same applies to the use of definite articles used to introduce a recitation of claims.

[0087] Moreover, where a particular number of introduced claim recitations are explicitly recited, it is understood that such recitation should be interpreted to mean at least the number recited (e.g., a recitation of "two enumerations" without other qualification means at least two enumerations, or an enumeration of two or more). Moreover, in instances where a recitation similar to "at least one of A, B, and C, etc." or "one or more of A, B, and C, etc." is used, such configurations are typically intended to include A alone, B alone, C alone, A and B together, A and C together, B and C together, or A, B, and C together, etc. For example, use of the term "and / or" is intended to be interpreted in this manner.

[0088] Furthermore, any disjunctive word or phrase expressing two or more alternative terms, whether in the description, claims, or drawings, should be understood to contemplate the possibility of including one of the terms, either one of the terms, or both terms. For example, the phrase "A or B" should be understood to include the possibilities of "A" or "B" or "A and B."

[0089] Additionally, the use of the terms "first", "second", "third", etc., is not necessarily used herein to imply a particular order or number of elements. Typically, the terms "first", "second", "third", etc., are used as general identifiers to distinguish between different elements. Absent an indication that the terms "first", "second", "third", etc., refer to a particular order, these terms should not be understood to refer to a particular order. Additionally, absent an indication that the terms "first", "second", "third", etc., refer to a particular number of elements, these terms should not be understood to refer to a particular number of elements. For example, a first widget may be described as having a first side, and a second widget may be described as having a second side. Use of the term "second side" with respect to a second widget is to distinguish such side of the second widget from the "first side" of the first widget, and does not imply that the second widget has two sides.

[0090] All examples and conditional language described herein are intended for teaching purposes to assist the reader in understanding the invention and the concepts that the invention contributes to further developing the art, and should not be construed as being limited to such specifically recited examples and conditions. Although the embodiments of the present disclosure have been described in detail, it should be understood that various modifications, substitutions, and alternatives can be made thereto without departing from the spirit and scope of the present disclosure.

[0091] In addition to the above-described embodiment, the following supplementary notes are further disclosed. (Appendix 1) obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of the dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including a graph analysis algorithm of the general characteristics and the threshold number; generating a display in a graphical user interface (GUI) that visualizes the allocation table; The method includes: (Appendix 2) Using the GXAI technology, training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; The method of claim 1, comprising: (Appendix 3) Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; calculating a similarity score based on the first vector and the second vector; The method of claim 1, comprising: (Appendix 4) generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithm; ranking the graph analysis algorithms with a rank number based on the correlation; sorting the rank table according to the rank numbers; outputting the rank table; The method of claim 1, comprising: (Appendix 5) The step of determining the general characteristics of the dataset of graph structured data comprises: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; The method of claim 1, comprising: (Appendix 6) obtaining a second data set of the graph structured data; determining a second general characteristic of a second dataset of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second allocation table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; 2. The method of claim 1, further comprising: (Appendix 7) obtaining a second dataset of graph structured data; determining a second general characteristic of the second dataset of graph structured data; determining that the second general characteristic is not similar to the general characteristic; generating a second ranked list of the plurality of graph analytic algorithms based on a second correlation between an analysis of the second dataset of graph structured data using GXAI technology and an analysis of the plurality of graph analytic algorithms; 2. The method of claim 1, further comprising: (Appendix 8) One or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause a system to perform an operation, the operation including: obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of the dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of best performing graph analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including the general characteristics and the threshold number of best performing graph analysis algorithms; generating a display in a graphical user interface (GUI) that visualizes the allocation table;

[0023] In one or more non-transitory computer-readable storage media, (Appendix 9) Using the GXAI technology, training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; 9. One or more non-transitory computer-readable storage media as described in claim 8, (Appendix 10) Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; Calculating a Pearson correlation coefficient or a cosine similarity based on the first vector and the second vector; 9. One or more non-transitory computer-readable storage media as described in claim 8, (Appendix 11) generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithm; ranking the graph analysis algorithms with a rank number based on the correlation, with a lower rank number reflecting a stronger correlation; sorting the rank table according to the rank numbers; outputting the rank table; 9. One or more non-transitory computer-readable storage media as described in claim 8, (Appendix 12) The step of determining the general characteristics of the dataset of graph structured data comprises: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; 9. One or more non-transitory computer-readable storage media as described in claim 8, (Appendix 13) The operation includes: obtaining a second data set of the graph structured data; determining a second general characteristic of a second dataset of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second allocation table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; 9. The one or more non-transitory computer-readable storage media of claim 8, further comprising: (Appendix 14) The operation includes: obtaining a second dataset of graph structured data; determining a second general characteristic of the second dataset of graph structured data; determining that the second general characteristic is not similar to the general characteristic; generating a second ranked list of the plurality of graph analytic algorithms based on a second correlation between an analysis of the second dataset of graph structured data using GXAI technology and an analysis of the plurality of graph analytic algorithms; 9. The one or more non-transitory computer-readable storage media of claim 8, further comprising: (Appendix 15) 1. A system comprising: one or more processors; one or more non-transitory computer-readable storage media configured to store instructions; the instructions, in response to being executed, cause the system to perform an action, the action comprising: obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of the dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of best performing graph analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including the general characteristics and the threshold number of best performing graph analysis algorithms; generating a display in a graphical user interface (GUI) that visualizes the allocation table; Including, the system. (Appendix 16) Using the GXAI technology, training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; 16. The system of claim 15, comprising: (Appendix 17) Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; Calculating a Pearson correlation coefficient or a cosine similarity based on the first vector and the second vector; 16. The system of claim 15, comprising: (Appendix 18) generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithm; ranking the graph analysis algorithms with a rank number based on the correlation, with a lower rank number reflecting a stronger correlation; sorting the rank table according to the rank numbers; outputting the rank table; 16. The system of claim 15, comprising: (Appendix 19) The step of determining the general characteristics of the dataset of graph structured data comprises: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; 16. The system of claim 15, comprising: (Appendix 20) The operation includes: obtaining a second data set of the graph structured data; determining a second general characteristic of a second dataset of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second allocation table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; 16. The system of claim 15, further comprising: [Explanation of symbols]

[0092] 110 Datasets 120 Characterization Module 125 General characteristics 130 Graph Explainable Artificial Intelligence Results 135 Graph Analysis Results 140 Correlation Module 145 Correlation Score 150 Ranking Module 155 Ranked List 160 Allocation Module 165 Allocation Table 170 Graphical User Interface (GUI) Module

Claims

1. obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of said dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including a graph analysis algorithm of the general characteristics and the threshold number; generating a display in a graphical user interface (GUI) that visualizes the allocation table; The method includes:

2. The use of the GXAI technology includes: training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; The method of claim 1 , comprising:

3. Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; calculating a similarity score based on the first vector and the second vector; The method of claim 1 , comprising:

4. generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithms; ranking the graph analysis algorithms with a rank number based on the correlation; sorting the rank table according to the rank numbers; outputting the rank table; The method of claim 1 , comprising:

5. Determining the general characteristics of the dataset of graph structured data includes: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; The method of claim 1 , comprising:

6. obtaining a second data set of the graph structured data; determining a second general characteristic of a second data set of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second assignment table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; The method of claim 1 further comprising:

7. obtaining a second dataset of graph structured data; determining a second general characteristic of the second data set of graph structured data; determining that the second general characteristic is not similar to the general characteristic; generating a second ranked list of the plurality of graph analysis algorithms based on a second correlation between an analysis of the second dataset of graph structured data using GXAI technology and an analysis of the plurality of graph analysis algorithms; The method of claim 1 further comprising:

8. One or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause a system to perform an action, the action including: obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of said dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of best performing graph analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including the general characteristics and the threshold number of best performing graph analysis algorithms; generating a display in a graphical user interface (GUI) that visualizes the allocation table; [0023] In one or more non-transitory computer readable storage media,

9. The use of the GXAI technology includes: training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; 10. The one or more non-transitory computer-readable storage media of claim 8, comprising:

10. Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; Calculating a Pearson correlation coefficient or a cosine similarity based on the first vector and the second vector; 10. The one or more non-transitory computer-readable storage media of claim 8, comprising:

11. generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithms; ranking the graph analysis algorithms with a rank number based on the correlation, with a lower rank number reflecting a stronger correlation; sorting the rank table according to the rank numbers; outputting the rank table; 10. The one or more non-transitory computer-readable storage media of claim 8, comprising:

12. Determining the general characteristics of the dataset of graph structured data includes: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; 10. The one or more non-transitory computer-readable storage media of claim 8, comprising:

13. The operation includes: obtaining a second data set of the graph structured data; determining a second general characteristic of a second data set of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second assignment table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; The one or more non-transitory computer-readable storage media of claim 8 , further comprising:

14. The operation includes: obtaining a second dataset of graph structured data; determining a second general characteristic of the second data set of graph structured data; determining that the second general characteristic is not similar to the general characteristic; generating a second ranked list of the plurality of graph analysis algorithms based on a second correlation between an analysis of the second dataset of graph structured data using GXAI technology and an analysis of the plurality of graph analysis algorithms; The one or more non-transitory computer-readable storage media of claim 8 , further comprising:

15. 1. A system comprising: one or more processors; one or more non-transitory computer-readable storage media configured to store instructions; the instructions, in response to being executed, cause the system to perform an action, the action comprising: obtaining a dataset of graph structured data, the dataset including one or more subgraphs; determining correlations between an analysis of said dataset of graph structured data using Graph Explainable Artificial Intelligence (GXAI) techniques and an analysis using a plurality of graph analytics algorithms; generating a ranked list of the plurality of graph analysis algorithms based on the correlation; determining a general characteristic of the dataset of graph structured data, the general characteristic indicative of a similarity between one or more sub-graphs; assigning a threshold number of best performing graph analysis algorithms to the general characteristic based on the ranked list of the plurality of graph analysis algorithms; generating an allocation table including the general characteristics and the threshold number of best performing graph analysis algorithms; generating a display in a graphical user interface (GUI) that visualizes the allocation table; Including, the system.

16. The use of the GXAI technology includes: training a machine learning (ML) model based on the dataset of graph structured data and a graph neural network (GNN); generating, by a GNN explainer, a numerical contribution score indicating how each node or edge of the dataset of graph structured data contributes to training the ML model; The system of claim 15 , comprising:

17. Determining correlations between the analysis of the dataset of graph structured data using GXAI techniques and the analysis using a plurality of graph analytics algorithms, generating a first vector for a first analysis using the GXAI technique; generating a second vector for a second analysis using the plurality of graph analysis algorithms; Calculating a Pearson correlation coefficient or a cosine similarity based on the first vector and the second vector; The system of claim 15 , comprising:

18. generating the ranked list of graph analysis algorithms based on the correlations, generating a rank table having at least a first column and a second column, the first column including the graph analysis algorithms and the second column indicating a rank of the graph analysis algorithms; ranking the graph analysis algorithms with a rank number based on the correlation, with a lower rank number reflecting a stronger correlation; sorting the rank table according to the rank numbers; outputting the rank table; The system of claim 15 , comprising:

19. Determining the general characteristics of the dataset of graph structured data includes: generating a vector for each of the one or more subgraphs; calculating an average vector of the vectors of the one or more subgraphs; The system of claim 15 , comprising:

20. The operation includes: obtaining a second data set of the graph structured data; determining a second general characteristic of a second data set of the graph structured data; determining that the second general characteristic is similar to the general characteristic; reading the allocation table; assigning a threshold number of analysis algorithms present in the assignment table to the second general characteristic; generating a second assignment table including a graph analysis algorithm for the second general characteristic and the threshold number; generating a display within the GUI that visualizes the second allocation table; The system of claim 15 further comprising:

Citation Information

Cited By

  • Transaction information processing system, transaction information processing method, and program

    JP7879298B1

  • Information processing system and information processing method for evaluating the integrity of decision-making processes

    JP7917862B1