A game user behavior data clustering method, system and storage medium

Abstract

This application relates to the field of data mining technology and discloses a method, system, and storage medium for clustering game user behavior data. The method includes: processing time-series interaction data using a graph neural network to construct an interaction network representing user decision nodes and their connections, and deriving decision propagation paths; labeling high-influence propagation chains based on the influence strength between nodes, extracting key nodes as candidate boundary points, and further performing cluster analysis to obtain a preliminary group classification structure; determining the final set of boundary points by analyzing the cross paths of decision propagation between clusters, thereby optimizing the user group model; and predicting user behavior patterns and making dynamic adjustments based on this optimized model. This application improves the accuracy of user group segmentation and the real-time performance of prediction.

Description

Technical Field

[0001] This application relates to the field of data mining technology, and in particular to a method, system and storage medium for clustering game user behavior data. Background Technology

[0002] Game user behavior data analysis is a crucial component of data-driven decision-making, widely applied in game design, operational optimization, and personalized recommendations. By deeply analyzing game user behavior patterns, game developers and operators can adjust game mechanics in real time, enhancing player experience, optimizing the virtual economy, and improving user retention and conversion rates. In these systems, user interaction goes beyond simple information exchange; it involves the dissemination of decisions and the dynamic evolution of group behavior, directly impacting game efficiency and player engagement. Traditional game user behavior analysis methods typically rely on static group segmentation and user behavior pattern identification, making them ill-suited to the highly dynamic changes in the game environment. In games, player behavior is influenced by time, context, or external events, leading to rapid changes in group structure and behavioral patterns. For example, in multiplayer online games, specific activities or game events can quickly alter player behavior, resulting in rapid reorganization of player groups. Traditional methods cannot capture these behavioral changes in real time, leading to lagging group segmentation and an inability to accurately predict player behavior and preferences, thus impacting personalized recommendations and user experience optimization.

[0003] Furthermore, existing technologies have significant limitations in capturing the dynamics of decision propagation paths in games. In games, players influence the decisions and behaviors of other players through information sharing, social interaction, and virtual item recommendations. Identifying the key nodes in these decision propagation paths and their influence is a technical challenge. This is especially true in massively multiplayer online games (MMORPGs), where player interactions are complex and dynamic. Some players may become key decision propagators through their words or actions, but traditional methods struggle to identify these key nodes and quantify their role in the evolution of group behavior. Consequently, game systems struggle to accurately predict player behavior, thus affecting the accuracy of group segmentation and behavior prediction. Summary of the Invention

[0004] This application provides a method, system, and storage medium for clustering game user behavior data, which can improve the accuracy of user group segmentation and the real-time performance of prediction.

[0005] Firstly, this application provides a method for clustering game user behavior data, the method comprising:

[0006] Step S1: Obtain the user's time-series interaction data in the interactive system, process the time-series interaction data using a graph neural network, construct an interaction network to represent the user's decision nodes and connection relationships, and obtain the decision propagation path;

[0007] Step S2: Determine the influence intensity between nodes based on the decision propagation path; mark high-influence propagation chains based on the influence intensity between nodes, extract key nodes from the high-influence propagation chains as boundary point candidates, and obtain a set of boundary points;

[0008] Step S3: Perform clustering and grouping processing on the set of boundary points to obtain a preliminary group classification structure; analyze the number of cross paths of inter-cluster decision propagation in the preliminary group classification structure to determine the final set of boundary points;

[0009] Step S4: Based on the final set of boundary points, update the node labels in the interaction network, and obtain the optimized user group model based on the updated node labels;

[0010] Step S5: Predict the behavior pattern using the optimized user group model, calculate the magnitude of change for the predicted behavior pattern, determine whether the magnitude of change exceeds a preset range, and obtain the dynamically adjusted classification boundary.

[0011] Secondly, this application provides a game user behavior data clustering system, which includes:

[0012] The data acquisition module is used to acquire the user's time-series interaction data in the interactive system, and to process the time-series interaction data using a graph neural network to construct an interaction network representing the user's decision nodes and connection relationships, thereby obtaining the decision propagation path;

[0013] The link marking module is used to determine the influence intensity between nodes based on the decision propagation path; mark high-influence propagation chains based on the influence intensity between nodes; extract key nodes from the high-influence propagation chains as boundary point candidates; and obtain a set of boundary points.

[0014] The boundary determination module is used to perform clustering and grouping processing on the set of boundary points to obtain a preliminary group classification structure; and to analyze the number of cross paths in the inter-cluster decision propagation in the preliminary group classification structure to determine the final set of boundary points.

[0015] The label update module is used to update the node labels in the interaction network based on the final boundary point set, and to obtain an optimized user group model based on the updated node labels.

[0016] A change prediction module is used to predict behavioral patterns using the optimized user group model, calculate the magnitude of change for the predicted behavioral patterns, determine whether the magnitude of change exceeds a preset range, and obtain dynamically adjusted classification boundaries. Thirdly, a computer-readable storage medium is provided, storing instructions that, when executed on a computer, cause the computer to perform the aforementioned game user behavior data clustering method.

[0017] The technical solution provided in this application employs graph neural networks to process time-series interactive data. This technology can accurately capture user decision propagation paths and identify key nodes and their influence intensity in real time, improving the accuracy and dynamic adaptability of group segmentation compared to traditional static clustering methods. By calculating the propagation paths and influence intensity between nodes, this method effectively solves the problem of traditional techniques struggling to identify decision propagation paths and key nodes in complex interactive scenarios in real time. Especially in multiplayer online game environments, it can accurately reflect the mutual influence between players, providing more accurate data support for behavior prediction and personalized recommendations. Furthermore, this technology optimizes the construction of user group models by clustering boundary point sets and combining the number of cross paths in inter-cluster decision propagation, improving the accuracy and reliability of group segmentation. Based on dynamically adjusting classification boundaries, this method can make timely adjustments when user behavior changes, ensuring the system can respond quickly to changes, avoiding prediction errors caused by lag in group segmentation, and improving the system's real-time response capability. Overall, this technology not only improves the accuracy of game user behavior analysis but can also be widely applied in multiple fields such as social platforms, e-commerce, and intelligent recommendations, providing strong support for various interactive systems. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the game user behavior data clustering method used in this application;

[0020] Figure 2 This application includes a flowchart of game user segmentation based on multi-dimensional behavioral feature clustering and group boundary stability determination.

[0021] Figure 3 This is a schematic diagram of the structure of the game user behavior data clustering system in this application. Detailed Implementation

[0022] This application provides a method, system, and storage medium for clustering game user behavior data. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0023] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of a game user behavior data clustering method in this application includes:

[0024] Step S1: Obtain the user's temporal interaction data in the interactive system, process the temporal interaction data using a graph neural network, construct an interaction network to represent the user's decision nodes and connection relationships, and obtain the decision propagation path. Specifically, this includes:

[0025] The system collects time-series interaction data of users within a preset time period from the interactive system. The time-series interaction data includes at least the user operation sequence and the interaction object. A graph neural network is used to process the time-series interaction data to generate an interaction network, where each node represents the user's decision behavior and each edge represents the connection relationship between decisions. Based on the interaction network, the propagation path between nodes is calculated to obtain the decision propagation path. The decision propagation path represents the propagation order and correlation strength of user decisions from the source node to the target node.

[0026] Specifically, firstly, temporal interaction data of users within a preset time period is collected from the interaction system. This temporal interaction data includes at least user operation sequences and interaction objects. Taking a game interaction system as an example, the preset time period can be set as a continuous login period within a day, such as from 9:00 AM to 9:00 PM. The collected data includes the player's operation sequences in the level, such as choosing weapons and moving paths, as well as interaction objects, such as other players or virtual items. This data provides the basic data for subsequent network construction. This data collection method helps to capture temporal dynamics and ensure data integrity. A graph neural network is used to process the temporal interaction data to generate an interaction network, where each node represents the user's decision behavior, and each edge represents the connection relationship between decisions. A graph neural network is a deep learning model based on a graph structure that aggregates information from the node's neighbors through a message passing mechanism to update the node's representation.

[0027] Specifically, the temporal interaction data is transformed into a graph structure, where nodes correspond to each decision action, such as a player choosing to attack or defend. Edges are connected according to the order of the actions and assigned weights to represent the strength of the association. Through multi-layer graph convolution operations, the network learns node embedding vectors to generate a complete interaction network, effectively capturing complex dependencies between decisions and improving the accuracy of path analysis. Based on the constructed interaction network, the propagation paths between nodes are calculated to obtain the decision propagation paths. The decision propagation paths represent the propagation order and association strength of user decisions from the source node to the target node. Specifically, a depth-first search algorithm is used to traverse the interaction network, starting from the source node (such as the initial decision), enumerating all possible paths to the target node, and calculating the cumulative association strength of each path. The degree is used as the weight of the path to obtain the decision propagation path. In this way, the propagation impact of decisions in the game can be quantified, key propagation chains can be discovered, and behavior prediction can be optimized. In addition, for each decision propagation path, its path length and the frequency of interaction between nodes are recorded to generate path feature vectors. These feature vectors are used to quantify the intensity of subsequent impact. This method can simplify subsequent processing, improve the efficiency of labeling high-impact paths, and thus optimize the user group model. Using the generated decision propagation paths and path feature vectors, node labels are optimized through graph neural networks, and an optimized user group model is constructed based on the updated node labels. This model can effectively reflect the user group structure and provide accurate support for subsequent behavior prediction and group segmentation.

[0028] Further, in step S1, a graph neural network is used to process the temporal interaction data to generate an interaction network, including:

[0029] The temporal interaction data is preprocessed to extract key behavioral features from the user operation sequence, including operation type and interaction time. Based on the key behavioral features, node feature matrices and edge feature matrices are constructed. A graph neural network is used to perform convolution operations on the node feature matrices and edge feature matrices to generate an embedded representation of the interaction network. The embedded representation represents the decision propagation relationship between nodes. Based on the embedded representation, the propagation path between nodes is calculated to generate a set of decision propagation paths. The interaction network is then constructed based on the set of propagation paths.

[0030] Specifically, the temporal interaction data is preprocessed to extract key behavioral features from the user operation sequence. These key features include operation type and interaction time. Operation type describes the user's specific actions during the interaction, such as clicking, moving, and selecting. Interaction time records the timestamp of each action, reflecting the temporal order of the actions. Based on these key behavioral features, node feature matrices and edge feature matrices are constructed. The node feature matrix represents the user's behavioral features at each decision point, while the edge feature matrix represents the connection relationships and correlations between decisions. A graph neural network is used to convolve the node feature matrix and the edge feature matrix. The graph neural network aggregates the node features and edge features through graph convolution to generate an embedded representation of the interaction network. The formula can be expressed as:

[0031]

[0032] in, Represents a node In the Layer feature representation, It is a node The set of neighboring nodes, It is the normalization coefficient. It is the first The weight matrix of the layer, Neighboring nodes In the Layer feature representation, It is a bias term. It is an activation function; through this graph convolution operation, the graph neural network can update the embedding representation of nodes based on the feature information of nodes and edges, thereby capturing the decision propagation relationship between nodes; these embedding representations not only reflect the local features of nodes, but also convey global structural information; based on the obtained embedding representations, the propagation path between nodes is calculated, generating a set of decision propagation paths; the propagation path refers to the propagation chain from a decision source node to a target node, and the influence intensity is measured by the weights of the nodes on the path and their associated edges; based on these propagation paths, an interactive network is constructed and further analyzed to identify key nodes and paths with greater influence, forming an interactive network for decision analysis and group segmentation.

[0033] Step S2: Determine the influence intensity between nodes based on the decision propagation path; mark high-influence propagation chains based on the influence intensity between nodes, extract key nodes from the high-influence propagation chains as candidate boundary points, and obtain a set of boundary points. Specifically, this includes:

[0034] For each decision propagation path, the weight values ​​between each node in the path are calculated. The weight values ​​are determined based on the interaction frequency between nodes and the path length. A weighted accumulation method is used to calculate the cumulative impact from the source node to the target node by accumulating the weight values ​​between each node in the path, thus obtaining the impact intensity. If the impact intensity is greater than a preset threshold, the path is marked as a high-impact propagation chain. For each high-impact propagation chain, its path feature vector is analyzed, and the key node with the highest weight value in the path is extracted. The key node represents the user behavior point that has the greatest impact on decision propagation. Based on the features of the key node, it is compared with the candidate conditions. If the features meet the candidate conditions, it is determined to be qualified as a candidate for boundary point. All key nodes in the high-impact propagation chains are summarized to generate a candidate set of boundary points. For the candidate set of boundary points, duplicate and redundant nodes are removed to obtain the set of boundary points.

[0035] Specifically, by collecting temporal interaction data of users in the game, including user operation sequences and interaction objects, an interaction network is constructed based on this data to represent the decision nodes of each player and their connections, thus obtaining the decision propagation path. In this process, by traversing all nodes in the decision propagation path, the weight value between each node is calculated to quantify the cumulative impact from the source node to the target node. The weight value of each node is determined by the interaction frequency and path length. The interaction frequency is determined by statistically analyzing the frequency of user interactions within a fixed time period, and the path length is adjusted by reflecting the distance of decision propagation. A decay factor is used to adjust the weight of the path. Based on these calculated weights, a weighted cumulative method is used to calculate the influence intensity from the source node to the target node. If the influence intensity is greater than a preset threshold, the path is marked as a high-influence propagation chain. For each high-influence propagation chain, the path feature vector is analyzed, and the node with the highest weight value is extracted as the key node. The path feature vector consists of the weight values ​​of each node in the path, representing the propagation characteristics of the path. By finding the node corresponding to the maximum value in the feature vector as the key node, this extraction method ensures that the user behavior points with the greatest impact on decision propagation can be accurately captured, providing effective support for the subsequent identification of boundary points.

[0036] Based on this, potential key nodes are further identified by calculating the local weight contribution of each node. A weighted cumulative method is used to quantify the weight contributions between nodes; if the product of node weights exceeds a predetermined multiple of the path's average, it is considered a potential key node. This method identifies key nodes with significant influence in the propagation path. By comparing the weight values ​​of all potential key nodes, the node with the highest weight value is selected as the final key node. For example, in a multiplayer online competitive game, if a node on the path has the highest weight value, that node is selected as the key node. This selection method ensures that the nodes most influential on decision propagation are extracted, providing a basis for generating the subsequent candidate set of boundary points. For each key node, its eligibility as a boundary point candidate is further evaluated based on its characteristics. The degree of a node represents the number of other nodes it connects to in the interaction network, and the coverage of the propagation path represents the path length influenced by that node. This feature evaluation ensures that candidate nodes have significant influence in decision propagation.

[0037] Based on this assessment, the degree of a node and the coverage of its propagation path are calculated, and then filtered according to a preset threshold. If both the degree and coverage of a node meet the conditions, the node is confirmed as a qualified candidate for a boundary point. To improve the accuracy of candidate nodes, the number of path branches is also considered. If the number of branches exceeds a preset value, the coverage of the node is expanded to ensure the influence of key nodes in the group structure. Key nodes in all high-influence propagation chains are summarized to generate a candidate set of boundary points. To ensure the accuracy of the set, a hash table is used to remove duplicate nodes, and cosine similarity is used to remove redundant nodes. The resulting candidate set of boundary points is input into a clustering algorithm for grouping, and nodes with similar propagation characteristics are grouped into the same cluster to obtain a preliminary group classification structure. In this way, key influencing nodes between groups can be accurately identified, and the stability and accuracy of subsequent group division can be ensured. Through the above method, key nodes in game user behavior can be identified efficiently and accurately, and group division can be optimized based on these nodes to improve the accuracy of behavior prediction, thereby supporting more refined user behavior analysis and personalized recommendations.

[0038] Step S3: Cluster the boundary point set to obtain a preliminary group classification structure; analyze the number of cross paths in the inter-cluster decision propagation within the preliminary group classification structure to determine the final boundary point set. Specifically, this includes:

[0039] Clustering algorithms are used to group nodes in the boundary point set. The clustering algorithm is based on the similarity of propagation features between nodes, including node weights and path coverage. For each group, the average feature vector of nodes within the cluster is calculated to generate cluster features. A preliminary group classification structure is determined based on these cluster features, representing the initial segmentation of user groups. For each cluster, the number of propagation paths and inter-cluster connection strength are recorded. For each cluster in the preliminary group classification structure, the number of intersection paths between it and other clusters is calculated, representing the degree of interaction in decision propagation between different clusters. If the number of intersection paths is less than a preset threshold, the inter-cluster boundary is considered stable, and the boundary points of that cluster are marked as final boundary points. For each final boundary point, its position and weight value in the interaction network are recorded to generate a final boundary point set. This final boundary point set is used for subsequent node label updates in the interaction network to optimize user group segmentation.

[0040] Specifically, a clustering algorithm is used to group the nodes in the boundary point set. The clustering algorithm divides the nodes based on the similarity of their propagation characteristics, which include the node's weight value and path coverage. The weight value represents the node's decision influence, while the path coverage reflects the breadth of the node's influence in decision propagation. By calculating the similarity between nodes, such as using Euclidean distance to measure the similarity of node features, the clustering algorithm groups nodes with similar propagation characteristics into the same group, thus obtaining a preliminary group classification structure. For each group, the average feature vector of the nodes within the cluster is calculated to generate cluster features. The cluster features represent the overall propagation characteristics of the group. Combining the weight value and path coverage of each node, the average feature vector of the cluster is generated to describe the group's performance in decision propagation. Based on these cluster features, the preliminary group classification structure is determined, representing the preliminary division result of the user group. For each cluster, the number of propagation paths of the nodes within the cluster and the inter-cluster connection strength also need to be recorded. These data will be used for subsequent analysis of the stability of the inter-cluster boundaries.

[0041] Furthermore, for each cluster in the preliminary group classification structure, the number of cross paths between it and other clusters is calculated. Cross paths represent the degree of interaction in decision propagation between different clusters. Specifically, by traversing each cluster, the path segments extending to other clusters in its decision propagation path are identified. The number of path segments can be used as a measure of cross paths, reflecting the frequency and intensity of decision influence propagation from one cluster to another. After calculating the number of all cross paths, if the number of cross paths is less than a preset threshold, the boundary between the clusters is determined to be stable, and the boundary point of the cluster is marked as the final boundary point.

[0042] For each final boundary point, its position and weight value in the interaction network are recorded, and a final boundary point set is generated. This set is used for subsequent node label updates in the interaction network to optimize user group segmentation. (Reference) Figure 2 The figure illustrates the process of segmenting game users based on multi-dimensional behavioral features clustering and group boundary stability determination.

[0043] Step S4: Based on the final set of boundary points, update the node labels in the interaction network. Based on the updated node labels, obtain the optimized user group model. Specifically, this includes:

[0044] Based on the final boundary point set, the label of each node in the interaction network is updated, and the node label represents the group category to which the node belongs. For each node, its connection relationship with each point in the final boundary point set is analyzed to determine its group affiliation. The interaction network is iteratively updated using a label propagation algorithm to generate an updated node label set. Based on the updated node label set, an optimized user group model is constructed. The optimized user group model represents the dynamic group structure of users in the interaction system and records the number of nodes and propagation characteristics of each group.

[0045] Specifically, based on the final set of boundary points, the label of each node in the interaction network is updated, with the node label representing the group category to which the node belongs. The group affiliation of each node is determined by analyzing its connections to the points in the final set of boundary points. Specifically, the connection relationship is quantified by calculating the graph distance from each node to the boundary points and using the shortest path algorithm to find the minimum number of edges between nodes. If the distance between a node and a boundary point is less than a set threshold, the node is classified into the corresponding group; otherwise, the node is marked for further analysis. A label propagation algorithm is used to iteratively update the interaction network. During initialization, the labels of all nodes are set according to their group affiliation. Then, in each iteration… In this process, each node collects the labels of its neighbors and updates its own labels through a majority voting mechanism until the label changes stabilize or the upper limit of the number of iterations is reached. This process helps to capture dynamic changes between groups and enhances the accuracy and robustness of group segmentation. Based on the updated set of node labels, an optimized user group model is constructed. This model dynamically reflects the group structure of users in the interaction system and records the number of nodes and propagation characteristics of each group. Specifically, the number of nodes in each group is counted, and the average weight of the decision propagation path within the group is calculated as a propagation characteristic. These data provide effective support for subsequent behavior pattern prediction and system optimization and can be adjusted in real time according to changes in the group.

[0046] Step S5: Predict behavioral patterns using an optimized user group model. For each predicted behavioral pattern, calculate its magnitude of change, determine whether the magnitude of change exceeds a preset range, and obtain the dynamically adjusted classification boundary. Specifically, this includes:

[0047] An optimized user group model is used to simulate user behavior patterns in future time-series interactions based on the historical propagation paths and node labels of the interaction network, generating predicted behavior patterns. For the predicted behavior patterns, the magnitude of change is calculated, which is determined based on the weight changes of the propagation paths between groups. If the magnitude of change exceeds the preset range, the classification boundary is adjusted to generate a dynamically adjusted classification boundary, which represents the latest classification result of the user group.

[0048] Specifically, based on the optimized user group model, the system first simulates user behavior patterns in future temporal interactions to generate predicted behavior patterns. This process is simulated using historical propagation paths and node labels of the interaction network, combined with the aggregation mechanism of graph neural networks to generate path embedding representations. By temporally encoding historical data, the model can predict the direction and intensity of future decision propagation. The simulation results are used to calculate the magnitude of changes in propagation paths between groups. Propagation paths between groups are extracted from the predicted behavior patterns, and the weight change of each path is calculated. The weight change is determined by comparing the difference between the predicted weight and the historical weight. The magnitude of changes in all inter-group paths is summarized, and the average value is calculated as the overall magnitude of change. If this magnitude of change exceeds a preset range, the classification boundary needs to be dynamically adjusted. For example, in multiplayer online competitive games, when the prediction shows that the influence between a certain player group is enhanced and the propagation intensity of the path is higher than in historical data, this may lead to changes in group behavior, requiring adjustments to the classification boundary to adapt to this new change.

[0049] When the magnitude of change exceeds a preset range, the critical paths causing the change are identified. These critical paths are those with the largest weight changes. By extracting candidate boundary points from these paths, the classification boundaries are further adjusted. The specific adjustment process is implemented using the K-means clustering algorithm. The clustering algorithm groups similar points into clusters based on the Euclidean distance between nodes, thereby optimizing the group partitioning, minimizing inter-cluster cross paths, and generating new classification boundaries. For each boundary point marked as high-influence, its position and weight value in the interaction network are recorded, and a final set of boundary points is generated. This set will be used for subsequent node label updates to further optimize user group partitioning. Through this dynamic adjustment process, the system can respond to changes in user behavior in real time, ensuring the accuracy of group partitioning. For example, suppose in a strategy game, a player's decision-making behavior changes rapidly, affecting the strategy of the entire group. The label update mechanism can adjust the classification boundaries in a timely manner according to this change, ensuring that player behavior patterns in the game can be effectively predicted. The adjusted boundary points and group characteristics will be recorded and saved in the database. This data can not only optimize the current interaction system but also be used for subsequent recommendation systems and matching algorithms, improving the overall user experience and system efficiency.

[0050] It is understood that the executing entity of this application can be a game user behavior data clustering system, a terminal, or a server; no specific limitation is made here. This application's embodiment uses a server as an example for illustration.

[0051] The above describes the game user behavior data clustering method in the embodiments of this application. The following describes the game user behavior data clustering system in the embodiments of this application. Please refer to [link / reference]. Figure 3 One embodiment of the game user behavior data clustering system in this application includes:

[0052] The data acquisition module is used to acquire the time-series interaction data of users in the interactive system. It uses a graph neural network to process the time-series interaction data, constructs an interaction network to represent user decision nodes and connection relationships, and obtains the decision propagation path.

[0053] The link marking module is used to determine the influence intensity between nodes based on the decision propagation path; high-influence propagation chains are marked based on the influence intensity between nodes, and key nodes are extracted from the high-influence propagation chains as boundary point candidates to obtain the boundary point set;

[0054] The boundary determination module is used to cluster and group the boundary point set to obtain a preliminary group classification structure; it analyzes the number of cross paths in the inter-cluster decision propagation in the preliminary group classification structure to determine the final boundary point set.

[0055] The label update module is used to update the node labels in the interaction network based on the final boundary point set, and to obtain an optimized user group model based on the updated node labels;

[0056] The change prediction module is used to predict behavior patterns through an optimized user group model, calculate the magnitude of change for the predicted behavior patterns, determine whether the magnitude of change exceeds a preset range, and obtain the dynamically adjusted classification boundary.

[0057] Through the collaborative efforts of the aforementioned components, this system enables efficient analysis and optimization of game user behavior. The data acquisition module first acquires time-series interaction data and processes this data using a graph neural network to construct an interaction network that accurately represents user decision nodes and their connections, thereby obtaining the decision propagation path. The link labeling module calculates the influence intensity between nodes based on the decision propagation path and labels high-influence propagation chains, extracting key nodes as boundary point candidates to obtain a set of boundary points. The boundary determination module then performs clustering and grouping processing on the boundary point set to obtain a preliminary group classification structure, and determines the final set of boundary points by analyzing the number of decision propagation cross paths between clusters. The label update module updates the node labels in the interaction network based on this, constructing an optimized user group model. The change prediction module predicts user behavior patterns based on the optimized group model and calculates their change magnitude, determining whether they exceed the preset range, thereby dynamically adjusting the classification boundary to ensure that the system can respond promptly to changes in user behavior and optimize group segmentation. The orderly cooperation of each module improves the accuracy of group segmentation and real-time response capability, providing strong support for accurate behavior prediction and personalized recommendations.

[0058] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the game user behavior data clustering method.

[0059] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0060] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for clustering game user behavior data, characterized in that, The method includes: Step S1: Obtain the user's temporal interaction data, which includes at least a user operation sequence and an interaction object. Preprocess the temporal interaction data to extract key behavioral features from the user operation sequence, including operation type and interaction time. Based on the key behavioral features, construct a node feature matrix and an edge feature matrix. Perform convolution operations on the node feature matrix and edge feature matrix using a graph neural network to generate an embedded representation of the interaction network, which characterizes the decision propagation relationship between nodes. Based on the embedded representation, calculate the propagation path between nodes to generate a set of decision propagation paths. Construct the interaction network based on the propagation path set. Each node represents the user's decision behavior, and each edge represents the connection relationship between decisions. Calculate the propagation path between nodes based on the interaction network to obtain the decision propagation path, which characterizes the propagation order and correlation strength of user decisions from the source node to the target node. Construct the interaction network to represent the user's decision nodes and their connection relationships to obtain the decision propagation path. Step S2: Determine the influence intensity between nodes based on the decision propagation path; mark high-influence propagation chains based on the influence intensity between nodes, extract key nodes from the high-influence propagation chains as boundary point candidates, and obtain a set of boundary points; Step S3: Perform clustering and grouping processing on the set of boundary points to obtain a preliminary group classification structure; for each cluster in the preliminary group classification structure, calculate the number of cross paths between it and other clusters, where the cross paths represent the degree of interaction in decision propagation between different clusters; if the number of cross paths is less than a preset threshold, determine that the inter-cluster boundary is stable, and mark the boundary points of the cluster as final boundary points; for each final boundary point, record its position and weight value in the interaction network to generate a final boundary point set, which is used for subsequent node label updates in the interaction network to optimize user group segmentation; Step S4: Based on the final boundary point set, update the label of each node in the interaction network, where the node label represents the group category to which the node belongs; for each node, analyze its connection relationship with each point in the final boundary point set to determine its group affiliation; use a label propagation algorithm to iteratively update the interaction network to generate an updated node label set; based on the updated node label set, construct an optimized user group model, where the optimized user group model represents the dynamic group structure of users in the interaction system, and records the number of nodes and propagation characteristics of each group; Step S5: Predict the behavior pattern using the optimized user group model, calculate the magnitude of change for the predicted behavior pattern, determine whether the magnitude of change exceeds a preset range, and obtain the dynamically adjusted classification boundary.

2. The method for clustering game user behavior data according to claim 1, characterized in that, S2 includes: For each decision propagation path, the weight values ​​between each node in the path are calculated. The weight values ​​are determined based on the interaction frequency between nodes and the path length. A weighted accumulation method is used to calculate the cumulative impact from the source node to the target node by accumulating the weight values ​​between each node in the path, thus obtaining the impact intensity. If the impact intensity is greater than a preset threshold, the path is marked as a high-impact propagation chain. For each high-impact propagation chain, its path feature vector is analyzed, and the key node with the highest weight value in the path is extracted. The key node represents the user behavior point that has the greatest impact on decision propagation. Based on the features of the key node, it is compared with the candidate conditions. If the features meet the candidate conditions, it is determined to be qualified as a boundary point candidate. All key nodes in the high-impact propagation chains are summarized to generate a boundary point candidate set. For the boundary point candidate set, duplicate and redundant nodes are removed to obtain the boundary point set.

3. The method for clustering game user behavior data according to claim 1, characterized in that, In step S3, the boundary point set is grouped to obtain a preliminary group classification structure, including: Clustering algorithms are used to group the nodes in the boundary point set. The clustering algorithm is based on the similarity of propagation features between nodes, and the propagation features include the weight value of the node and the path coverage. For each group, the average feature vector of the nodes in its cluster is calculated to generate cluster features. Based on the cluster features, a preliminary group classification structure is determined, which represents the preliminary division result of the user group. For each cluster, the number of propagation paths of the nodes in its cluster and the inter-cluster connection strength are recorded.

4. The method for clustering game user behavior data according to claim 1, characterized in that, S4 includes: Based on the final set of boundary points, the label of each node in the interaction network is updated, whereby the node label represents the group category to which the node belongs. For each node, its connection relationship with each point in the final set of boundary points is analyzed to determine its group affiliation. The interaction network is iteratively updated using a label propagation algorithm to generate an updated set of node labels. Based on the updated set of node labels, an optimized user group model is constructed, which represents the dynamic group structure of users in the interaction system and records the number of nodes and propagation characteristics of each group.

5. The method for clustering game user behavior data according to claim 1, characterized in that, S5 includes: Using the optimized user group model, based on the historical propagation path and node labels of the interaction network, the behavior pattern of users in future time-series interactions is simulated to generate the predicted behavior pattern; for the predicted behavior pattern, its change magnitude is calculated, and the change magnitude is determined based on the weight change of the propagation path between groups; if the change magnitude exceeds a preset range, the classification boundary is adjusted to generate a dynamically adjusted classification boundary, and the dynamically adjusted classification boundary represents the latest division result of the user group.

6. A game user behavior data clustering system, used to implement the game user behavior data clustering method as described in any one of claims 1 to 5, characterized in that, The game user behavior data clustering system includes: The data acquisition module is used to acquire the user's time-series interaction data in the interactive system, and to process the time-series interaction data using a graph neural network to construct an interaction network representing the user's decision nodes and connection relationships, thereby obtaining the decision propagation path; The link marking module is used to determine the influence intensity between nodes based on the decision propagation path; mark high-influence propagation chains based on the influence intensity between nodes; extract key nodes from the high-influence propagation chains as boundary point candidates; and obtain a set of boundary points. The boundary determination module is used to perform clustering and grouping processing on the set of boundary points to obtain a preliminary group classification structure; and to analyze the number of cross paths in the inter-cluster decision propagation in the preliminary group classification structure to determine the final set of boundary points. The label update module is used to update the node labels in the interaction network based on the final boundary point set, and to obtain an optimized user group model based on the updated node labels. The change prediction module is used to predict behavior patterns through the optimized user group model, calculate the magnitude of change of the predicted behavior patterns, determine whether the magnitude of change exceeds a preset range, and obtain the dynamically adjusted classification boundary.

7. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements a game user behavior data clustering method as described in any one of claims 1 to 5.

Images