Game data multi-dimensional association analysis and visualization method, system, equipment and medium

Abstract

The invention discloses a game data multi-dimensional association analysis and visualization method, system and device and a medium, and the method specifically comprises the steps: automatically recommending an index association scheme through a recommendation engine based on an association rule generated by a game industry knowledge graph; inputting the somatosensory gesture data and the index association scheme into a pre-trained gesture recognition model, and recognizing a standardized gesture operation; mapping the recognized standardized gesture operation into a specific operation instruction for the game economic data or the player behavior track through a preset mapping rule base; inputting a gesture sequence formed by the specific operation instruction into a gesture intention prediction engine, and pre-judging a subsequent operation target of the user; and according to the gesture sequence and a pre-judged subsequent operation target, automatically triggering and executing a corresponding analysis process, and carrying out real-time visual display through a 3D thermodynamic diagram. According to the method, efficient multi-dimensional association analysis and real-time visual display of the game data are realized, and the analysis efficiency and decision scientificity are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to methods, systems, devices and media for multidimensional correlation analysis and visualization of game data. Background Technology

[0002] In today's booming gaming industry, game data analysis plays a crucial role in optimizing game experience, improving player retention, and increasing game revenue. However, the field of game data analysis still faces many pressing issues that severely restrict the efficiency and quality of data analysis, thereby affecting the scientific rigor and timeliness of game operation decisions.

[0003] First, current game data analysis primarily relies on the traditional mouse and keyboard operation mode. This method becomes extremely cumbersome when dealing with complex, multi-dimensional data. For example, when analysts need to compare multiple game data metrics across different dimensions, such as player level distribution, completion rates for different levels, and spending frequency of paying players, switching and viewing data one by one using the mouse and keyboard is not only cumbersome but also prone to errors, significantly reducing interaction efficiency and making it difficult for analysts to quickly obtain the necessary information, thus affecting timely judgments about the game's status.

[0004] Secondly, complex relationships often exist between multiple metrics in game data. However, existing analytical methods struggle to visually represent these relationships. For instance, analyzing the connection between players' social behaviors (such as team-up frequency and number of friends) and spending behaviors (such as recharge amount and type of items purchased) requires analysts to perform complex configurations and extensive calculations, setting up various correlation rules and statistical models. This not only demands extremely high levels of professional skill from analysts but is also time-consuming and labor-intensive, making it difficult to quickly discover deep correlations between data and hindering timely insights into key factors affecting game operations.

[0005] Furthermore, most existing visualization tools are limited to two-dimensional charts, such as bar charts, line charts, and pie charts. While these two-dimensional charts are effective at presenting simple data relationships, they fall short when dealing with complex data relationships, such as those involving multiple variables changing simultaneously or spatial distribution. For example, when analyzing the relationship between player activity and resource distribution in different areas of a game, two-dimensional charts cannot clearly show the complex connections between these three dimensions: area, activity, and resources. This limits the depth of analysis and makes it difficult for analysts to fully and accurately understand the inherent characteristics of game data.

[0006] Finally, game data is dynamic data that changes constantly over time, such as changes in players' online time at different times and fluctuations in the currency circulation of the in-game economy. However, existing analytics tools struggle to provide smooth, dynamic analysis of data changes over time. Traditional methods often require manually switching between data views at different points in time when displaying dynamic data changes, failing to achieve real-time dynamic display and interaction. This makes it difficult for analysts to observe the continuity and trends of data changes and to promptly capture key changes in game data over time.

[0007] Furthermore, while existing 3D visualization tools offer richer visual effects, they lack intuitive operation methods and have high learning costs. Analysts need to spend a significant amount of time familiarizing themselves with complex operation processes and interaction logic, making these tools difficult to apply to real-time interactive analysis scenarios and unable to meet the real-time and efficiency requirements of game data analysis. Summary of the Invention

[0008] The purpose of this invention is to provide a method, system, device, and medium for multidimensional correlation analysis and visualization of game data, which realizes efficient multidimensional correlation analysis and real-time intuitive display of game data, improves analysis efficiency and scientific decision-making, and solves at least one of the above-mentioned problems in the prior art.

[0009] In a first aspect, the present invention provides a method for multidimensional correlation analysis and visualization of game data, the method specifically comprising: Based on the user behavior pattern library, the paid conversion path library, and the indicator association rule library, a knowledge graph of the game industry is constructed. Based on association rules generated from a knowledge graph of the gaming industry, a recommendation engine automatically recommends association schemes based on metrics. Collect users' somatosensory gesture data, input the somatosensory gesture data and indicator association scheme into a pre-trained gesture recognition model, and recognize standardized gesture operations; By using a pre-defined mapping rule library, the recognized standardized gesture operations are mapped into specific operation instructions for game economic data or player behavior trajectories. The gesture sequence formed by the specific operation instructions is input into the gesture intent prediction engine to predict the user's subsequent operation goals; Based on the gesture sequence and the predicted subsequent operation target, the corresponding analysis process is automatically triggered and executed, and the output data of the analysis process is visualized in real time through a 3D heat map.

[0010] Secondly, this invention provides a multidimensional correlation analysis and visualization system for game data, the system specifically comprising: The knowledge graph module is used to build a knowledge graph for the gaming industry based on a user behavior pattern library, a paid conversion path library, and a metric association rule library. The indicator recommendation module is used to automatically recommend indicator association schemes based on association rules generated from the knowledge graph of the game industry through the recommendation engine. The gesture operation module is used to collect the user's somatosensory gesture data, input the somatosensory gesture data and indicator association scheme into the pre-trained gesture recognition model, and recognize standardized gesture operations. The instruction mapping module is used to map the recognized standardized gesture operations into specific operation instructions for game economic data or player behavior trajectories through a preset mapping rule library. The intent prediction module is used to input the gesture sequence formed by the specific operation command into the gesture intent prediction engine to predict the user's subsequent operation target; The data display module is used to automatically trigger and execute the corresponding analysis process based on the gesture sequence and the predicted subsequent operation target, and to visualize the output data of the analysis process in real time through a 3D heat map.

[0011] Thirdly, the present invention provides a computer device, including: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements the game data multidimensional correlation analysis and visualization method as described in any of the above methods.

[0012] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the game data multidimensional correlation analysis and visualization method as described in any of the above methods.

[0013] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention enables efficient multidimensional correlation analysis and real-time intuitive display of game data, improving analysis efficiency and the scientific nature of decision-making.

[0014] 2. Based on target indicators and analysis context, this invention generates and optimizes recommendation indicator association schemes from knowledge graphs, helping users quickly obtain relevant analysis dimensions and improve analysis efficiency.

[0015] 3. This invention collects haptic gesture data and combines it with the analysis context of the indicator association scheme, using a pre-trained model to recognize standardized gesture operations, thereby achieving a natural and intuitive interaction method.

[0016] 4. This invention constructs a mapping rule base, which associates standardized gesture operations with system context as specific operation instructions, and precisely controls the analysis and operation of game economic data or player behavior trajectories.

[0017] 5. This invention extracts high-order intent features by recording gesture sequences, uses an attention mechanism model to predict subsequent operation targets, and performs system preloading or interface guidance in advance, thereby improving the smoothness of interaction.

[0018] 6. This invention dynamically assembles and analyzes a pipeline based on gesture sequences and predicted targets, and visualizes the results in real time through 3D heat maps, intuitively presenting the inherent relationships of the data, thereby enhancing the depth and real-time nature of the analysis. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for multidimensional correlation analysis and visualization of game data provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a game data multidimensional correlation analysis and visualization system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0022] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0023] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0024] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0025] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0027] In this application embodiment, the entity executing the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 A flowchart illustrating a method for multidimensional correlation analysis and visualization of game data disclosed in an embodiment of the present invention is shown below: S101 constructs a knowledge graph for the gaming industry based on a user behavior pattern library, a paid conversion path library, and a rule library for association of metrics.

[0028] In this embodiment, for the user behavior pattern library, player operation records in the game are extracted, including login time, operation type (such as combat, trading, social interaction), operation frequency, etc. Invalid or duplicate records are removed, and the time format and operation classification standards are unified. For the paid conversion path library, data on the conversion process from free users to paying users is collected, covering information such as the first payment time, payment amount range, and payment trigger events (such as level unlocking, item purchase), etc. The payment amount is normalized, and different currency units are uniformly converted into standard values. For the indicator association rule library, predefined indicator association rules are extracted, such as "positive correlation between player level and equipment purchase probability" and "correlation between online time and social interaction frequency", etc. The natural language expression in the rule description is converted into structured fields. For example, "positive correlation" is uniformly marked as "association type: positive correlation", and key elements such as indicator name, association direction, and association strength involved in the rule are extracted.

[0029] Based on the preprocessed data, entities and relationships between them are extracted from three databases. Entity types include player entities (such as player ID, player level, player payment status), game behavior entities (such as login behavior, combat behavior, transaction behavior), game metric entities (such as online time, completion rate, payment rate), and business rule entities (such as payment conversion rules, metric association rules). During the extraction process, natural language processing technology is used to identify entities and relationships in the text, and semantic verification is performed using a domain knowledge base (such as a game terminology dictionary) to ensure the accuracy of the extraction results.

[0030] The architecture of the game industry knowledge graph is designed, using a triple structure of "entity-relationship-entity" as the basic unit. The graph is divided into three layers: the bottom layer is the data layer, storing the raw entities and relationships extracted from three databases; the middle layer is the schema layer, defining entity types, relationship types, and their attributes. For example, the "player" entity is defined to have attributes such as "ID," "level," and "payment status," and the "payment conversion" relationship has attributes such as "trigger event," "conversion rate," and "time interval"; the top layer is the application layer, providing knowledge query and reasoning interfaces for specific analysis scenarios (such as player retention analysis and payment optimization). In the architecture design, the concept of ontology is introduced to clarify the hierarchical relationships of each entity type, and attribute constraints are used to ensure the consistency of knowledge. For example, the association rule between the "player" entity and the "game behavior" entity is designed as "players can perform multiple game behaviors," and the subclasses of "game behavior" (such as combat behavior and social behavior) and their attributes (such as the "damage value" of combat behavior and the "number of interactive objects" of social behavior) are defined.

[0031] The knowledge extracted from the three databases is integrated to resolve entity alignment and relationship conflict issues. For entity alignment, entities describing the same object from different databases are merged by comparing entity attributes (such as player ID, operation time) or semantic features (such as the similarity of behavioral descriptions). For example, "Player X - Login Behavior - Time T1" in the user behavior pattern database and "Player X - Free Status - Time T1" in the paid conversion path database point to the same player and need to be merged into "Player X (ID: 123, Status: Free, Last Login Time: T1)". For relationship conflict resolution, in cases where different relationship descriptions exist between the same pair of entities (such as "Player level is positively correlated with equipment purchase probability" in the indicator association rule database, while historical data shows that some high-level players have a low purchase probability), weighted voting or evidence-based reasoning methods are used to determine the final relationship. For example, if the support of the positive correlation rule (such as 80% of high-level players having a higher purchase probability than low-level players) is higher than the influence of the conflict evidence, the positive correlation is retained, and the source and confidence level of the conflict are marked in the graph.

[0032] The merged knowledge is stored in a graph database (such as Neo4j or Janus Graph) to construct a knowledge graph for the gaming industry. Simultaneously, indexes are created for the graph (e.g., by player ID or behavior type) to accelerate subsequent knowledge retrieval. During the construction process, an incremental update mechanism is employed. When the data in the three databases is updated (e.g., adding player behavior records or modifying payment rules), the knowledge extraction and graph update process is automatically triggered to ensure the timeliness of the graph.

[0033] In this embodiment, a game industry knowledge graph is constructed based on a user behavior pattern library, a payment conversion path library, and an indicator association rule library, providing structured knowledge support for subsequent multi-dimensional association analysis and visualization of game data.

[0034] S102, based on association rules generated from the knowledge graph of the game industry, automatically recommends indicator association schemes through the recommendation engine.

[0035] In this embodiment, all relation edges in the knowledge graph are traversed to filter out associations involving game metrics (such as player level, online time, payment rate, and completion rate). For example, from the association edge between the "player behavior" entity and the "game metrics" entity, the rule "daily login frequency - impact - percentage of daily active users" is extracted. Then, the extracted association rules are preprocessed, including rule deduplication, rule standardization, and rule confidence calculation. Through preprocessing, the quality and consistency of the association rules are ensured, providing a reliable foundation for subsequent recommendations.

[0036] Before the recommendation engine starts, the current user's analysis scenario is identified. By analyzing the user's historical operation records, currently opened data analysis interface, and entered query keywords, the game business areas the user focuses on (such as player retention, monetization optimization, and level design) and specific analysis goals (such as identifying key metrics affecting player retention and analyzing the behavioral characteristics of paying players). For example, if a user has recently frequently queried data related to "new player first-day retention rate" and filtered metrics such as "new player tutorial completion time" and "first payment time" in the interface, then the analysis scenario is identified as "analysis of factors affecting new player retention." Simultaneously, the scenario is further refined by considering the current operational stage of the game (such as new game launch period, version update period, and event period). For example, during a new game launch period, users may be more concerned about the relationship between early player behavior and long-term retention; during an event, they may be more concerned about the correlation between event participation and monetization conversion. By accurately identifying the analysis scenario, it is ensured that the recommended metric correlation scheme is highly matched with user needs.

[0037] The recommendation engine matches suitable metric association schemes from a preprocessed association rule base based on the identified analysis scenario. A multi-dimensional matching strategy is employed: First, it matches based on scenario keywords. For example, in the scenario of "analysis of factors affecting new player retention," it filters association rules containing entities related to "new player" and "retention rate," such as "new player onboarding completion time - positive correlation - new player first-day retention rate" and "first payment time - negative correlation - new player third-day retention rate." Second, it ranks rules based on their confidence levels, prioritizing rules with higher confidence. For example, if the confidence level of "new player onboarding completion time - positive correlation - new player first-day retention rate" is 0.85, then rules with higher confidence levels are prioritized. If the confidence level of "Equipment Type - Impact - New Player Day 1 Retention Rate" is 0.6, then the former should be prioritized. Finally, considering the coverage and complementarity of rules, avoid recommending too many duplicate or redundant rules. For example, if "Player Level - Positive Correlation - Equipment Purchase Probability" has already been recommended, then the semantically similar "Player Level - Impact - High-Level Equipment Purchase Rate" should not be recommended. At the same time, new rules with weak correlation to the already recommended rules but valuable for scenario analysis should be added. For example, in the player retention analysis scenario, the rule "Social Interaction Frequency - Positive Correlation - Long-Term Player Retention Rate" should be added. Through multi-dimensional matching, a set of targeted and comprehensive indicator correlation schemes are generated.

[0038] The matched metric association schemes are ranked and then personalized based on user preferences. Ranking criteria include rule confidence, rule-scenario matching degree, and historical usage frequency of the rules. For personalization, recommended schemes are weighted or filtered based on the user's historical selection records (e.g., the user previously focused more on paid-related metrics) or explicit feedback (e.g., the user marked certain metrics as "highly focused" on the interface). Through ranking and personalization, it is ensured that the recommended metric association schemes both meet the needs of the analysis scenario and align with the user's personal preferences.

[0039] The sorted indicator association schemes are presented to users in a visual manner, with interactive functions to allow users to further filter and adjust them. The display methods include list display (each scheme displays the rule content, confidence level, associated indicators, etc.) and graphical display (such as using a network diagram to show the relationship between indicators, where nodes are indicators, edges are association rules, and edge thickness represents the confidence level).

[0040] S103 collects the user's somatosensory gesture data, inputs the somatosensory gesture data and indicator association scheme into the pre-trained gesture recognition model, and identifies standardized gesture operations.

[0041] In this embodiment, user actions are monitored in real time. When a user begins to make a gesture related to game data analysis, gesture data is recorded. For example, a user might wave their arm to switch between different data analysis views, or clench their fist or open their fingers to select or deselect specific indicator association schemes. During data acquisition, a depth camera continuously acquires depth images of the user's hand, and image processing algorithms extract the positional information of key points on the hand (such as fingertips and knuckles). The IMU sensor records the hand's acceleration and angular velocity data, reflecting the hand's motion state. The system synchronously fuses multi-source data collected from different devices, for example, by temporally aligning the hand position information in the depth image with the motion parameters measured by the IMU sensor to obtain more comprehensive and accurate gesture data.

[0042] Key information is extracted from the metric association schemes generated by the recommendation engine, including the names of the involved metrics (such as player level, payment amount, completion rate, etc.), the correlation between metrics (such as positive correlation, negative correlation), and the confidence level of the association rules. For example, for the metric association scheme "player level - positive correlation - payment amount, confidence level 0.8", key information such as "player level", "payment amount", "positive correlation", and "0.8" are extracted. Then, this key information is converted into a feature vector suitable for model processing. One-hot encoding can be used to encode the metric names and correlations. For example, "player level" is encoded as [1, 0, 0, ...] (assuming there are N possible metrics, this is just an example), "payment amount" is encoded as [0, 1, 0, ...], "positive correlation" is encoded as [1, 0], and "negative correlation" is encoded as [0, 1]. The confidence level is directly used as a numerical element in the feature vector. By extracting features, the indicator association scheme is transformed into digital features that computers can understand and process, providing a foundation for subsequent fusion and recognition with somatosensory gesture data.

[0043] The preprocessed motion-sensing gesture data and feature vectors from the indicator association scheme are fused to construct a comprehensive input feature. The fusion can be achieved through simple concatenation, such as sequentially concatenating the feature vectors of the motion-sensing gesture data (e.g., coordinate sequences of hand key points, acceleration, and angular velocity sequences) with the feature vectors of the indicator association scheme (e.g., indicator encoding, association relationship encoding, and confidence level) to form a longer feature vector. This fusion method combines the user's gesture with the currently relevant indicator association scheme, enabling the gesture recognition model to consider the business intent behind the user's actions. For example, when a user makes a waving arm gesture, the model not only recognizes the gesture based on its physical characteristics but also considers the currently relevant indicator association scheme (e.g., whether it is related to payment-related indicators), thus more accurately determining whether the user's gesture intent is to switch to a payment data analysis view or perform other operations. Through data fusion, the gesture recognition model's ability to understand and accurately interpret user actions is improved.

[0044] The fused feature vector is input into a pre-trained gesture recognition model. This model employs a deep learning architecture, such as a convolutional neural network (CNN) combined with a recurrent neural network (RNN) or its variants (e.g., Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU)). The CNN part processes spatial features in the haptic gesture data, such as extracting hand shape and texture features from depth images; the RNN part processes temporal features in the gesture data, capturing the changing patterns of gesture actions over time. During model training, a large number of labeled gesture data samples are used, covering various common gestures (such as waving, grasping, rotating, etc.) and combinations with different metric association schemes. The model parameters are continuously adjusted through backpropagation, enabling the model to accurately map the input feature vector to standardized gesture operations. When the fused feature vector is input, the model outputs the corresponding gesture operation category based on its learned patterns and features, such as standardized gesture operation results like "switching views," "selecting metrics," or "adjusting parameters."

[0045] After the gesture recognition model outputs standardized gesture operation results, post-processing and verification operations are performed. Post-processing mainly includes filtering and correcting the recognition results. For example, a moving average filtering algorithm is used to smooth the gesture recognition results of multiple consecutive frames, removing erroneous recognitions caused by accidental interference. Simultaneously, the reasonableness of the recognition results is checked according to business rules. For instance, when a "select indicator" gesture is recognized, it is checked whether the selected indicator is within the currently operable range. Verification is performed by comparing the results with the user's actual feedback. The system can provide visual feedback to the user (e.g., displaying the recognized gesture operation and its corresponding business meaning on the screen) to allow the user to confirm its correctness. If the user reports an error in the recognition result, the system records the error and uses it for subsequent model optimization and improvement. For example, it can retrain the model by adding corresponding error samples to improve the accuracy and robustness of the gesture recognition model. Through post-processing and verification, it is ensured that the recognized standardized gesture operations accurately reflect the user's actual intentions, providing a reliable basis for subsequent operation command mapping and analysis process triggering.

[0046] In this embodiment, the process of collecting users' haptic gesture data, inputting the haptic gesture data and indicator association scheme into a pre-trained gesture recognition model, and recognizing standardized gesture operations is realized, providing strong support for convenient interaction and efficient operation in game data analysis.

[0047] S104 uses a preset mapping rule library to map the recognized standardized gesture operations into specific operation instructions for game economic data or player behavior trajectories.

[0048] In this embodiment, the scope of game economic data is defined, including costs incurred during game development, such as server rental fees, game art design fees, and programmer salaries; and profits generated during game operation, such as player recharge amounts, advertising revenue, and in-game item sales revenue. Furthermore, the statistical methods and calculation criteria for each data point are clearly defined, for example, whether recharge amounts are calculated based on single recharge records or cumulative recharge amounts over a certain time period.

[0049] For player behavior patterns, a thorough analysis of various player behaviors is conducted. Regarding online rate, the statistical timeframe is determined, such as daily online rate, weekly online rate, etc.; retention rate is divided into different dimensions such as day-2 retention rate, 7-day retention rate, and 30-day retention rate. For online time, the start and end times and calculation methods for statistics must be clearly defined—whether it's based on the cumulative time of each login or on effective game time (excluding idle time, etc.). Social aspects encompass player team-up frequency, number of friend additions, number of social interactions (such as chatting, trading, etc.), and the type and frequency of participation in social activities. These analyzed elements will be compiled into a detailed document, serving as the foundation for subsequently building a mapping rule base.

[0050] The collected standardized hand gestures are categorized in detail. First, they are initially classified based on their morphological characteristics, such as fist-clenching gestures (single-handed fist, double-handed fist), extension gestures (finger extension, palm extension), rotation gestures (palm rotation clockwise, palm rotation counterclockwise), and waving gestures (arm waving forward, arm waving backward). Then, for each type of gesture, the different variations in force, speed, and direction are further analyzed. For example, fist-clenching gestures are categorized by the force of the clench: light, medium, and heavy; by the speed of the clench: fast and slow; and by the direction of the clench: upward, downward, leftward, and rightward. In this way, standardized hand gestures are subdivided into various specific operation types, and each operation type is assigned a unique identifier for accurate matching in the subsequent mapping rule base.

[0051] Based on the previously analyzed game economic data, player behavior trajectory elements, and standardized gesture operations after classification, a hierarchical mapping rule base is constructed. This rule base adopts a multi-level structure, with the first level divided into two main categories according to the operation object: game economic data operations and player behavior trajectory operations.

[0052] Under the category of game economic data operations, two subcategories are further defined: cost expenditure operations and profit revenue operations. The cost expenditure operations subcategory is further subdivided according to different cost items, such as server rental cost operations and art design cost operations. The profit revenue operations subcategory is subdivided according to revenue source, such as recharge revenue operations and advertising revenue operations. For each subcategory, specific mapping rules between gesture operations and operation commands are established. For example, for server rental cost operations, a light grip with one hand and a slow upward swing is defined as the command to query server rental costs; for recharge revenue operations, a fist clenched with both hands and a rapid forward swing is defined as the command to query the recharge revenue for the current day.

[0053] Under the category of player behavior tracking operations, further subcategories exist such as online rate operations, retention rate operations, online time operations, and social interaction operations. The online rate operation subcategory uses rules based on different statistical time ranges; for example, extending one hand and rotating it clockwise is the command to query the daily online rate. The retention rate operation subcategory uses rules based on retention days; for example, extending both hands and rotating them counter-clockwise is the command to query the 7-day retention rate. The online time operation subcategory uses rules based on the statistical method; for example, waving one's hand forward is the command to query the player's cumulative online time during this login session. The social interaction operation subcategory uses rules based on different social behaviors; for example, extending and rapidly waving one's fingers is the command to query the player's daily team-up frequency.

[0054] The constructed mapping rule base is stored in a dedicated database with an indexing mechanism for fast querying and matching. Simultaneously, the rule base undergoes rigorous testing and validation, with game data analysts invited to conduct practical tests. Based on feedback, the rules are adjusted and optimized to ensure their accuracy and usability.

[0055] Standardized gesture operations are input into a pre-trained gesture recognition model, which can identify the type of standardized gesture operation based on feature information. For example, by analyzing the shape features and force information of the gesture, it can determine whether it is a light grip with one hand or a heavy grip with both hands; based on the speed and direction information of the gesture, it can determine whether it is a fast forward wave or a slow backward wave. After identifying the type of gesture operation, the system further analyzes the relevant parameters of the gesture operation, such as the magnitude of the force, the speed, and the specific angle of the direction, and performs preliminary matching of this information with rules in the mapping rule base to prepare for subsequent precise mapping.

[0056] Based on the parsed gesture operation type and related parameters, the system performs precise matching within a pre-defined mapping rule base. The matching process employs a combination of multi-level indexing and fuzzy matching. First, a first-level index is determined based on the operation object (game economic data or player behavior trajectory) to quickly locate the corresponding category. Then, a second-level index is created based on the type of gesture operation to further narrow down the query scope.

[0057] In the specific matching process, considering that gesture operation parameters may have certain errors and variations, a fuzzy matching algorithm is adopted. For example, for the force parameter, a reasonable error range is set. If the identified force value falls within the error range set in the rule base, the match is considered successful. Similarly, a similar fuzzy matching method is used for speed and direction parameters. In this way, operation commands that match the gesture operation can be found more accurately.

[0058] If no exact match is found during the matching process, the system will use a similarity algorithm to find the closest rule and make appropriate adjustments and corrections. For example, if the found rule is to lightly grip and slowly wave upwards with one hand to query server rental costs, while the current gesture is to lightly grip with one hand but wave upwards slightly faster, the system will adjust the range or level of detail of the query results according to a preset adjustment strategy to ensure that it can provide users with reasonable operation instructions.

[0059] Based on the mapping rules obtained through precise matching or adjustment, the system generates specific operation instructions for game economic data or player behavior patterns. These instructions include operation type (query, statistics, analysis, etc.), operation object (specific cost items, revenue sources, player behavior indicators, etc.), and operation parameters (statistical time range, calculation method, etc.). For example, a generated operation instruction for game economic data might be "Query the server rental costs for this month and perform statistical analysis according to server type"; an operation instruction for player behavior patterns might be "Query the player retention rate for the past 7 days and analyze the differences in retention rates among players from different channels."

[0060] The system will provide analysts with clear and easy-to-understand visual feedback on the generated operation instructions. A pop-up dialog box can appear in a specific area of ​​the game data analysis interface, displaying the detailed content of the operation instructions, such as "You are about to query this month's server rental costs and categorize them by server type. Do you confirm execution?" Confirm and cancel buttons will also be provided for analysts to choose from.

[0061] If the analyst confirms the execution command, the system will send the command to the corresponding data processing module to perform the actual operation on the game's economic data or player behavior patterns. For example, a command to query server rental costs will be sent to the game's financial management module to obtain relevant cost data and perform statistical analysis; a command to query player retention rates will be sent to the player data management module to obtain player login records and calculate retention rates. If the analyst cancels the command, the system will return to the previous level, waiting for the analyst to perform the gesture operation again.

[0062] During game data analysis, the mapping rule base needs continuous dynamic updates and optimization as the game updates and user needs change. The system collects analyst feedback and actual usage data to analyze which mapping rules are frequently used and effective, and which rules are inadequate or need improvement.

[0063] For example, if analysts are frequently dissatisfied with the mapping results of a particular gesture, or if new game features need to be added to the mapping rule base, the system will promptly organize relevant personnel to adjust and add rules. Simultaneously, the system will automatically optimize the mapping rule base using machine learning algorithms. By analyzing large amounts of gesture data and corresponding command execution results, the machine learning model can learn a more accurate mapping relationship between gestures and commands, automatically adjusting parameters and conditions in the rules to improve the accuracy and efficiency of the mapping.

[0064] For example, based on analysts' actual feedback on the force and speed of gesture operations, the system automatically adjusts the correspondence between force and speed and operation commands, making the mapping more in line with analysts' operating habits and business needs. Through dynamic updates and optimizations, the mapping rule base is ensured to always adapt to changes in game data analysis, providing analysts with accurate and efficient operation command mapping services.

[0065] In this embodiment, a preset mapping rule base is used to map the recognized standardized gesture operations into specific operation instructions for game economic data or player behavior trajectories, providing strong support for convenient interaction and efficient operation in game data analysis.

[0066] S105, inputs the gesture sequence formed by the specific operation instructions into the gesture intent prediction engine to predict the user's subsequent operation target.

[0067] In this embodiment, a large amount of analyst operation data is collected during game data analysis, covering operation records of different analysts in various analysis scenarios. These operation records include specific operation instructions and corresponding gesture operations, such as specific gestures used by analysts when querying player level distribution data, such as slowly rotating a fist with one hand.

[0068] The collected gestures are arranged chronologically to form gesture sequences. For each gesture sequence, the corresponding operation instruction and its position and role in the entire data analysis process are recorded in detail. For example, a gesture sequence might start by making a fist with one hand and slowly rotating it to check the player level distribution, then extend both hands and quickly bring them together to filter players of a specific level, and finally tap with a finger to view the detailed information of the filtered players.

[0069] The collected gesture sequences are categorized and organized according to the type of operation command (such as querying, filtering, viewing detailed information, etc.) and the data analysis scenario (such as player behavior analysis, game economy analysis, etc.). The categorized gesture sequences are stored in a dedicated database, with each sequence assigned a unique identifier and an indexing mechanism established for quick subsequent querying and retrieval.

[0070] A certain number of gesture sequences are selected from the gesture sequence database as training samples. For each training sample, in addition to recording the gesture sequence itself and the corresponding operation command, a professional game data analyst needs to label its possible subsequent operation goals. For example, for a gesture sequence that queries the player level distribution and then filters players of a specific level, the analyst might label its subsequent operation goal as checking the spending or social behavior of the filtered players.

[0071] Machine learning algorithms, such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs), are used to construct a gesture intent prediction model. The model is trained by taking gesture sequences from training samples as input and corresponding subsequent operation target annotations as output. During training, the model learns the order, frequency, and association between each gesture in the gesture sequence and the operation command, gradually mastering the rules for predicting subsequent operation targets from gesture sequences.

[0072] To improve the model's accuracy and generalization ability, cross-validation is used for evaluation and optimization. The training samples are divided into multiple subsets, with one subset used as the validation set and the remaining subsets used as the training set for both training and validation. Based on the validation results, the model's parameters and structure are adjusted, such as the number of layers and neurons in the neural network, until the model's accuracy on the validation set reaches the expected level.

[0073] In the game data analysis system, a dedicated gesture sequence receiving module is set up. When analysts perform game data analysis operations, this module collects their motion gesture data in real time and arranges this data chronologically to form a real-time gesture sequence. The gesture sequence undergoes preprocessing, including noise and outlier removal. Simultaneously, the gesture sequence is standardized to unify the gesture operations of different analysts under the same scale and specifications for subsequent predictive analysis. The preprocessed gesture sequence is stored in a temporary buffer, ready for subsequent input into the gesture intent prediction engine. The buffer size is set according to actual needs to ensure that it can store a sufficiently long gesture sequence to accurately predict the user's subsequent action goals.

[0074] The preprocessed gesture sequence is read from a temporary buffer and input into a trained gesture intent prediction engine. The engine first extracts features from the input gesture sequence, including key features such as the shape, force, speed, direction, and time intervals between gestures. These extracted features are then fed into the prediction model, which predicts the subsequent action target based on learned patterns and rules. The prediction process is complex; by learning from a large number of training samples, the model can identify patterns and trends in the gesture sequence, thereby inferring the user's possible subsequent action target. For example, if the gesture sequence first includes an action to query player level distribution, followed by an action to filter players of a specific level, the prediction model might predict that the user's subsequent action target is to check the payment status or social behavior of the filtered players. The prediction engine outputs the prediction results as probabilities, representing the likelihood of each possible subsequent action target.

[0075] The prediction engine outputs multiple possible subsequent action targets, which are then sorted according to their probabilities. A reasonable probability threshold is set, and actions with probabilities greater than this threshold are considered candidate targets. For example, if the probability threshold is set to 0.7, only actions with probabilities greater than 0.7 will be included in the candidate range. The candidate targets are further filtered by combining the current game data analysis scenario and the user's historical action records. For example, if player behavior analysis is currently underway, and the user's historical actions frequently involve checking player level distribution followed by viewing player spending, then "viewing the spending of filtered players" can be prioritized as the subsequent action target. Finally, the most probable subsequent action target is determined and fed back to the game data analysis system. Based on this prediction, the system can prepare the corresponding data and analysis processes in advance, providing users with more efficient and personalized data analysis services. For example, if the predicted user's subsequent action target is to view the spending of filtered players, the system can preload relevant spending data and quickly display the results when the user issues the action command.

[0076] In this embodiment, the gesture sequence formed by specific operation instructions is input into the gesture intent prediction engine to predict the user's subsequent operation goals, thus providing a guarantee for intelligent interaction and efficient decision-making in game data analysis.

[0077] S106 automatically triggers and executes the corresponding analysis process based on the gesture sequence and the predicted subsequent operation target, and displays the output data of the analysis process in real time through a 3D heat map.

[0078] This embodiment comprehensively outlines common analytical processes in game data analysis, such as player behavior trend analysis, game economic system fluctuation analysis, and player performance difference analysis across different levels. For each outlined analytical process, the required input data types are recorded in detail, such as player login time, spending amount, and level completion time; the calculation methods and logic involved in the analysis process, such as the formulas and statistical methods used to calculate player retention rate; and the final output data content, such as the specific values ​​of player retention rate in different time periods and the average completion time of players for each level. Simultaneously, the gesture sequence characteristics that may correspond to each analytical process and the predicted subsequent operation goals are analyzed. For example, for the player behavior trend analysis process, if the predicted subsequent operation goal is to view the changing trend of player activity within a specific time period, the corresponding gesture sequence might be to first swipe a finger to select a time range, and then clench a fist to confirm. These mapping relationships are compiled into a detailed document and stored in a dedicated mapping relationship library.

[0079] The system reads gesture sequences and predicted operation targets from a temporary data buffer and inputs them into the mapping database query module. This module performs rapid matching within the mapping database based on the input information. The matching process employs multi-level indexing and intelligent search algorithms. First, it performs a general filtering based on the predicted operation target to narrow down the matching range; then, it performs precise matching by combining the specific characteristics of the gesture sequence. For example, if the predicted operation target is to analyze the spending behavior of a certain type of player, it first filters out analysis processes related to spending behavior analysis from the mapping database, and then further determines the specific analysis process based on features such as the frequency and location of finger clicks in the gesture sequence. Once a match is successful, the system automatically triggers the corresponding analysis process. The triggering process includes calling the computing resources required for the analysis process, loading the relevant datasets, and initializing the analysis environment. For example, if the matched analysis process is the analysis of fluctuations in the game's economic system, the system will automatically allocate sufficient computing resources to process a large amount of economic data, load relevant datasets such as in-game currency circulation and item trading, and initialize the algorithms and models used to calculate economic indicators.

[0080] Once the analysis process is triggered, the system begins execution according to the preset analysis logic and calculation methods. During execution, the progress and status of the analysis are monitored in real time to ensure smooth operation. For data processing and calculations involved in the analysis, efficient data structures and algorithms are employed to improve processing speed. For example, when calculating player retention rate, a quicksort algorithm is used to sort player login time data to quickly count the number of retained players in different time periods. After the analysis process is completed, the system obtains the output data. The output data includes various statistical indicators, trend charts, and analysis conclusions. For example, for player behavior trend analysis, the output data may include specific numerical values ​​of player activity in different time periods, activity change trend curves, and analysis of key factors affecting activity changes. The output data is then initially organized and formatted for subsequent visualization.

[0081] Determine the display dimensions of the 3D heatmap. Based on the characteristics of the output data from the analysis process, select three appropriate dimensions for display. For example, when analyzing the relationship between player activity and resource distribution in different areas of the game, you can choose area location as the X-axis, player activity as the Y-axis, and resource quantity as the Z-axis.

[0082] Design color mapping rules for 3D heatmaps, mapping different data values ​​to different colors based on the size and range of the data. For example, areas with high player activity can be represented by red, areas with low activity by blue, and intermediate transitional areas by yellow, green, etc., to visually display the distribution of the data.

[0083] Meanwhile, interactive features for the 3D heatmap, such as zooming, rotating, and panning, are designed to allow users to observe and analyze the data from different angles. For example, users can rotate the 3D heatmap by swiping their finger or using the mouse to view details of different areas; and they can adjust the display scale of the chart using the zoom function to observe the data details more clearly.

[0084] The constructed 3D heatmap visualization model is integrated into the game data analysis system to ensure seamless integration with other modules of the system, preparing for subsequent real-time visualization.

[0085] The output data from the analysis process is input into the constructed 3D heatmap visualization model. Based on the input data, the visualization model generates a 3D heatmap in real time according to preset color mapping rules and display dimensions. During the 3D heatmap generation process, efficient rendering techniques are employed to ensure both the speed of chart generation and display quality. For example, a graphics processing unit (GPU) is used to accelerate the rendering process, improving chart generation efficiency and avoiding stuttering or delays.

[0086] The generated 3D heatmap is displayed in real-time on the visualization interface of the game data analysis system. The interface design is simple and clear, making it easy for users to operate and view. Users can perform various operations on the 3D heatmap through interactive buttons on the interface, such as rotating, zooming, and selecting specific areas to view detailed information.

[0087] In some embodiments, step S101 above, which involves constructing a game industry knowledge graph based on a user behavior pattern library, a paid conversion path library, and an indicator association rule library, specifically includes: Collect multi-source heterogeneous data from game clients, servers and operation databases, clean and segment the multi-source heterogeneous data for sessions, and form standardized user behavior data. Based on user behavior data, core entities are extracted through a predefined domain ontology model, and semantic relationships between core entities are established. The core entities include players, game behaviors, and payment events. The core entities and their semantic relationships are stored in a graph database, and knowledge is fused through entity linking technology to construct an initial knowledge graph containing a user behavior pattern library and a paid conversion path library. Based on historical case analysis, graph mining algorithms are used to determine the relationships between indicators from the initial knowledge graph and construct an indicator association rule base. Inject the indicator association rule base into the initial knowledge graph to form a game industry knowledge graph.

[0088] In this embodiment, the data collected from the game client covers various operational information of the player during gameplay, such as the number of times buttons are clicked, the trajectory of the moving character, and the frequency of skill usage. Technicians capture the player's operational data in real time by embedding a specific data acquisition interface in the game client and transmit it to a designated data storage location.

[0089] Server-side data collection primarily involves the game's operational status and player connection information, such as server load, player login and logout times, and network latency. Operations engineers use the server's built-in monitoring tools and log systems to periodically collect this data and perform initial organization and categorization.

[0090] The operations database stores various business data for the game, such as player personal information, payment records, and in-game item transaction information. Database administrators write data extraction scripts to extract the required data from the operations database and convert its format for subsequent unified processing.

[0091] The collected multi-source heterogeneous data suffers from issues such as inconsistent formats, data redundancy, and errors. Therefore, data cleaning and session segmentation are necessary. Data cleaners use specialized data cleaning tools to perform operations such as deduplication, filling in missing values, and correcting erroneous data. For example, duplicate data in player payment records is deduplicated by comparing unique identifiers; missing player level information is reasonably estimated and filled in based on player playtime and experience points.

[0092] Session segmentation divides a player's gameplay into multiple sessions according to certain rules. For example, based on a player's login and logout times, each login and logout session can be considered a single session. Session segmentation allows for clearer analysis of player behavior patterns and spending behavior across different sessions. The cleaned and segmented data forms standardized user behavior data, providing a high-quality data foundation for subsequent knowledge graph construction.

[0093] Based on standardized user behavior data, game data analysis experts and knowledge graph construction engineers collaborated to extract core entities and establish semantic relationships. First, a predefined domain ontology model was developed, which clarified the core entities involved in the game industry knowledge graph and the relationships between them. Core entities include players, game behaviors, and payment events.

[0094] Data extractors use natural language processing techniques and rule engines to extract information related to core entities from user behavior data. For example, by identifying player identifiers in the data, they extract basic player information such as player ID, nickname, and level; for game behavior, they extract behavioral information such as combat, quest completion, and social interaction based on players' in-game operation records; and for paid events, they extract information such as payment time, payment amount, and payment method from payment records.

[0095] While extracting core entities, semantic relationships between them are established. For example, there is an "execution" relationship between players and game actions, meaning a player performs a certain game action; there is a "trigger" relationship between players and paid events, meaning a player's action triggers a paid event; there may also be associations between game actions and paid events, such as a player making a payment after completing a specific task. By establishing these semantic relationships, various phenomena and patterns in the gaming industry can be described more accurately.

[0096] The extracted core entities and their semantic relationships are stored in a graph database. A suitable graph database management system, such as Neo4j, is selected, as this system has efficient graph data storage and query capabilities. Based on the graph database schema design, the data storage personnel construct the initial graph structure, using core entities as nodes and semantic relationships as edges.

[0097] During storage, entity linking technology is used for knowledge fusion. Entity linking technology can associate and merge the same entity described in different data sources. For example, player data collected from the game client and player data in the operations database may have some overlap. Entity linking technology can merge these data together to form a more complete and accurate player entity information.

[0098] Through knowledge fusion, an initial knowledge graph was constructed, comprising a user behavior pattern library and a paid conversion path library. The user behavior pattern library records various behavioral patterns and regularities of players in the game, while the paid conversion path library describes the conversion process and key factors from free users to paying users. The initial knowledge graph provides a basic framework for subsequent metric association rule mining.

[0099] Based on historical case studies, we conducted a rule mining study on the association of game metrics. These case studies covered various data analysis projects conducted during game operations, such as player retention rate analysis and conversion rate analysis. Graph mining algorithms were used to extract relationships between metrics from the initial knowledge graph. Graph mining algorithms can analyze the association patterns between nodes and edges in a graph structure to discover potential metric association rules. For example, by analyzing the relationship between player level, game duration, and payment amount, we discovered a rule that higher player levels, longer game durations, and higher payment amounts were correlated. During the mining process, the results were evaluated and filtered, removing unreasonable or coincidental association rules and retaining those with practical significance and statistical importance. The filtered metric association rules were compiled into a rule library, which records the relationships and influence levels between various game metrics, providing important reference for game operation decisions.

[0100] The constructed indicator association rule base is injected into the initial knowledge graph. The injection process must ensure that the indicator association rules are compatible with the core entities and semantic relationships in the initial knowledge graph. Data fusion personnel add the indicator association rules to the knowledge graph in an appropriate form by writing data fusion scripts, such as adding indicator association rules as new edges or attributes to relevant nodes.

[0101] After injecting the metric association rule base, the game industry knowledge graph is improved and optimized. The consistency and completeness of the data in the knowledge graph are checked to ensure the logical correctness between various entities and relationships. Simultaneously, the knowledge graph is visualized to facilitate more intuitive understanding and use by game operators and data analysts. Through continuous improvement and optimization, a complete game industry knowledge graph is ultimately formed, covering information on multiple aspects such as user behavior patterns, payment conversion paths, and metric association rules, providing strong support for precise operation and decision-making in the game industry.

[0102] In some embodiments, step S102 above, where the association rules generated based on the game industry knowledge graph are used to automatically recommend association schemes based on metrics through a recommendation engine, specifically includes: In response to the target metrics selected by the user, the current analysis context information, including business cycle and user role, is obtained through the recommendation engine; Based on the target metrics and current analysis context, multiple candidate metric association schemes are generated by traversing the game industry knowledge graph in real time. The candidate indicator association schemes are merged and deduplicated, and the recommendation score of each candidate indicator association scheme is calculated to generate a ranking list of recommendation schemes. The recommended suggestions are ranked and pushed to users, and the recommendation strategy is optimized and updated based on user feedback.

[0103] In this embodiment, a user interface is built where users select target metrics based on their analytical needs. Once the user selects the target metric, the system immediately triggers the recommendation engine. The recommendation engine quickly interacts with the game operation management system to obtain the contextual information for the current analysis. This contextual information primarily includes the business cycle and the user's role.

[0104] The determination of the business cycle depends on the actual situation of game operation, and can be based on days, weeks, months, or quarters. The system will automatically identify the current business cycle based on the regular rhythm of game operation and the user's possible analytical needs. User roles are determined based on the user's identity when logging into the system. Different user roles have different responsibilities and focuses in game operation. For example, operations personnel may focus more on overall operational metrics, while marketing personnel may focus more on marketing-related metrics. By obtaining this current analytical context information, the recommendation engine can more accurately provide users with metric association solutions that meet their needs.

[0105] After obtaining the target metric and the current analysis context, the recommendation engine immediately initiates a real-time traversal of the game industry knowledge graph. The game industry knowledge graph is a vast and complex knowledge system that contains various entities within the game and the relationships between them, such as the connections between players, game behaviors, payment events, and game metrics.

[0106] The recommendation engine starts with the target metric and performs a deep and broad search within the gaming industry knowledge graph. It filters relevant entities and relationships based on the business cycle and user roles within the current analytical context. For example, if the target metric is paid conversion rate, the business cycle is monthly, and the user role is operations personnel, the recommendation engine will search the knowledge graph for other metrics related to the monthly paid conversion rate, such as monthly payment amounts for different user groups and the monthly paid conversion rate for new users.

[0107] During the traversal, the recommendation engine continuously discovers various potential correlations between metrics and combines these relationships into multiple candidate metric association schemes. Each candidate metric association scheme contains a set of interrelated metrics with certain logical relationships, enabling analysis and interpretation of the target metric from different perspectives. For example, a candidate metric association scheme might include metrics such as paid conversion rate, the proportion of paying players at different levels, and the average game time of paying players. Through correlation analysis of these metrics, a deeper understanding of the factors influencing paid conversion rate can be achieved.

[0108] Because the real-time traversal of the game industry knowledge graph may generate a large number of duplicate or similar candidate indicator association schemes, it is necessary to merge and deduplicate these candidate schemes. The system will set a set of fusion rules. For example, when two candidate schemes contain most of the same indicators, but differ only in the order of the indicators or the level of detail of some indicators, the two schemes will be merged to retain the most comprehensive and reasonable combination of indicators.

[0109] After merging and deduplication, the system calculates a recommendation score for each candidate metric's associated solutions. The calculation of the recommendation score considers multiple factors, including the correlation between metrics, the explanatory power of the target metric, the fit with the business cycle, and the applicability to the user role. For example, if a candidate solution's metrics are highly correlated with the target metric, can effectively explain the reasons for changes in the target metric, and have high practical value within the current business cycle and user role, then that solution will receive a higher recommendation score.

[0110] Based on the calculated recommendation scores, all candidate indicator-related solutions are sorted to generate a ranked list of recommended solutions. The ranked list is arranged from highest to lowest recommendation score, placing the solutions most likely to meet the user's needs at the top for easy selection.

[0111] The generated list of recommended solutions will be pushed to users. The push method can be selected based on user settings and system functionality, such as through system notifications, emails, or pop-ups. When pushing the solution, a brief description of each recommendation will be displayed, including the key metrics included and their relationship to the target metric, helping users quickly understand the solution's content and value.

[0112] After receiving the recommended solutions, users will adopt or provide feedback on the solutions based on their own analysis and decision-making needs. The system will collect user feedback information in real time, such as which recommended solution the user chose and their level of satisfaction with the solution. Based on user feedback, the system will optimize and update the recommendation strategy.

[0113] If a recommendation is frequently adopted by users, the system analyzes its characteristics and advantages, applying the successful experience to the generation of other recommendations. Conversely, if a recommendation is repeatedly ignored or receives poor feedback, the system investigates the reasons and adjusts the parameters or algorithms of the recommendation engine to avoid generating similar low-quality recommendations. By continuously optimizing and updating the recommendation strategy based on user feedback, the system can gradually improve the accuracy and practicality of recommendations, providing users with more suitable indicator association solutions.

[0114] In some embodiments, step S103 above, which involves collecting the user's somatosensory gesture data and inputting the somatosensory gesture data and indicator association scheme into a pre-trained gesture recognition model to identify standardized gesture operations, specifically includes: Collect users' motion-sensing gesture data, and based on the motion-sensing gesture data, extract gesture feature vectors including spatial configuration and motion trajectory through a feature calculation engine; A large number of gesture samples labeled with analysis context are obtained, and the gesture recognition model is trained by using the gesture samples to build a hybrid model based on convolutional neural network and recurrent neural network. The analytical context information represented by the indicator association scheme is encoded into a context feature vector. The context feature vector and the gesture feature vector are concatenated and input into the gesture recognition model. The gesture recognition model outputs standardized gesture operations that correspond to the current analysis context information.

[0115] In this embodiment, when the user performs an operation, the motion-sensing camera records the user's motion gesture data in real time. This data covers the user's hand, arm and even whole body movement information, including but not limited to hand extension, bending and grasping, arm swinging and rotation, and body movement and tilting.

[0116] After collecting the motion-sensing gesture data, it is processed using a feature calculation engine. This engine is a specially designed software module that employs a series of algorithms to perform in-depth analysis of the raw motion-sensing gesture data. First, for spatial configuration features, the engine calculates the shape, size, and relative positional relationships between the user's hand or arm in three-dimensional space. For example, by analyzing the coordinates of hand joints, it determines whether the hand's grasping shape is a fist, an open palm, or bent fingers. For motion trajectory features, the engine records the positional changes of the user's gesture at different points in time, forming a continuous motion path. For example, it records the trajectory of the user's arm from its starting position to its ending position, including information such as the trajectory's length, direction, and speed. Finally, these spatial configuration and motion trajectory features are integrated into a gesture feature vector, which precisely describes the key features of the user's motion-sensing gesture in digital form.

[0117] To generate a gesture recognition model capable of accurately identifying standardized gesture operations, it is first necessary to acquire a large number of gesture samples labeled with analytical context. These gesture samples come from multiple sources. On the one hand, user-generated haptic gesture data is collected in actual game operations or data analysis scenarios, and the analytical context information at that time is recorded, such as the type of metric being analyzed (player retention rate, paid conversion rate, etc.), business cycle (daily, weekly, monthly), and user role (operations personnel, marketing personnel), etc. On the other hand, professionals are invited to perform standardized haptic gesture operations according to preset analytical contexts in simulated analytical scenarios, and these operations are then labeled in detail.

[0118] After collecting a sufficient number of gesture samples, these samples are used to train a hybrid model built from convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs excel at processing image and spatial data, effectively extracting spatial features from gesture samples, such as hand shape and position information. RNNs, on the other hand, have powerful processing capabilities for sequential data, capturing the temporal trajectory features of gestures. Hybridizing these two types of neural networks allows for full utilization of their strengths, leading to a more comprehensive understanding of gesture characteristics.

[0119] During training, the labeled gesture samples are divided into a training set and a validation set. The training set is used to adjust the model's parameters, allowing the model to gradually learn the mapping relationship between gesture features and the analysis context. The validation set is used to evaluate the model's performance, ensuring that the model maintains high accuracy even on unseen data. Through continuous iterative training and optimization, a high-performance gesture recognition model is finally generated.

[0120] After obtaining the indicator association scheme, the analytical context information represented by the scheme needs to be encoded into a context feature vector. The analytical context information includes multiple dimensions, such as indicator type, business cycle, and user role. To transform this information into a digital form that can be processed by a computer, a set of encoding rules needs to be designed.

[0121] For metric types, they are categorized and coded according to common game metrics, such as player retention rate as 01 and paid conversion rate as 02. Business cycles are coded as 1, 2, and 3 for daily, weekly, and monthly periods, respectively. User roles are coded according to different roles such as operations personnel and marketing personnel. These coded information from different dimensions are combined into a contextual feature vector, which accurately reflects the current analytical context.

[0122] The context feature vector and the previously extracted gesture feature vector are concatenated. This concatenation operation joins the two vectors together in a specific order to form a longer feature vector. This concatenated feature vector contains both detailed features of the user's haptic gestures and environmental information about the analysis context, providing a more comprehensive input to the gesture recognition model. Finally, the concatenated feature vector is fed into the trained gesture recognition model. Upon receiving the input, the gesture recognition model uses its internally learned knowledge to analyze and process the input features.

[0123] After receiving the concatenated feature vector, the gesture recognition model uses its complex internal neural network structure to perform calculations and inferences. The model first extracts and integrates features from the input feature vector, and then further explores the deep relationship between gesture features and the analysis context.

[0124] By comparing and matching the model with a large number of gesture sample patterns learned during training, the model can accurately determine the standardized gesture operation corresponding to the current analysis context. For example, if the current analysis context is that operations personnel are analyzing player retention rate metrics during a monthly business cycle, the model may recognize the standardized gesture operation of "clenching a fist and slowly raising it upwards," which may represent confirmation of player retention rate data or an instruction for further analysis.

[0125] Ultimately, the gesture recognition model outputs the identified standardized gesture operations in digital form, and can also transform them into visual gesture displays, making it easier for users and related systems to understand and apply them. In this way, the goal of accurately recognizing standardized gesture operations based on different analytical contexts and user-sensory gestures is achieved, providing a more natural and efficient approach to interaction and analysis in the gaming industry.

[0126] In some embodiments, in step S104 above, mapping the identified standardized gesture operations to specific operation instructions on game economic data or player behavior trajectories through a preset mapping rule base specifically includes: Construct a mapping rule library, which includes multiple mapping rules that associate standardized gesture operations with system context states to specific operation instructions; When a standardized gesture operation is received, the system context state, including the current chart type and the focused data object, is determined by the rule matching engine. Based on standardized gesture operations and system context state, a joint query is performed in the mapping rule base to retrieve the matching target mapping rule; Based on the parameter mapping logic set in the target mapping rules, operation parameters for game economic data or player behavior trajectories are extracted from the dynamic attributes of standardized gesture operations to form specific operation instructions.

[0127] In this embodiment, a series of common and easily identifiable actions are defined for standardized gesture operations, such as clicking, swiping, pinching, and rotating fingers, and extending, contracting, and swinging arms. Each gesture operation is described and categorized in detail to ensure its uniqueness and distinguishability.

[0128] The system context state encompasses multiple aspects. The current chart type is a crucial factor. Commonly used charts in game operations and analysis include bar charts, line charts, pie charts, and scatter plots. Different chart types are used to display different types of data and for different analytical purposes. For example, bar charts are suitable for comparing the magnitude of different categories of data, while line charts are better at showing the trend of data changes over time. The focus data object refers to the specific data point or data series that the user is currently focusing on in the chart. For instance, when analyzing player spending data, the focus might be the average spending amount per player over a specific time period, or the spending distribution of a particular player group.

[0129] Specific operational instructions include querying, filtering, sorting, and calculating game economic data, as well as tracking, analyzing, and simulating player behavior. For example, querying the total amount recharged by players within a certain time period, filtering out players whose spending exceeds a certain threshold, and visualizing players' movement trajectories in the game.

[0130] This study investigates the characteristics of game economic data and player behavior patterns to analyze potential user actions in different scenarios. Based on this analysis, several mapping rules are developed to associate standardized gesture operations with system context states and specific operation commands. For example, when the system context state is a line graph displaying the number of daily active users, and the focused data object is the number of daily active users for a specific day, if the identified standardized gesture operation is a double-tap, the corresponding mapping rule might be to query the detailed composition of that daily active user count, including the number of new users and the number of returning users. These mapping rules are then organized and stored in a mapping rule library.

[0131] When a standardized gesture is received, the system context state needs to be accurately determined. This process is achieved through a rule matching engine. The rule matching engine first obtains the currently displayed game data chart information and compares it with preset chart type templates to determine the current chart type. For example, if the chart contains a time axis and a value axis, and the data points are connected by line segments to form a continuous curve, it can be identified as a line chart. To determine the focus data object, the rule matching engine monitors user interaction behavior. When a user operates on the chart, such as hovering the mouse or clicking, the engine records the location of the user's operation and determines the focus data object based on the chart's coordinate system and data mapping relationships. For example, if a user clicks on a bar in a bar chart, the engine determines the specific data object represented by that bar, such as the player payment rate for a certain month, based on the bar's category and time information. Through these methods, the rule matching engine can accurately determine the system context state, including the current chart type and the focus data object, providing the necessary conditions for subsequent mapping rule queries.

[0132] After determining the standardized gesture operation and system context state, a joint query needs to be performed in the mapping rule base to retrieve the matching target mapping rule. The joint query process is a precise matching and filtering process. The rule matching engine uses the characteristic information of the identified standardized gesture operation, such as the type, direction, and force of the gesture, as well as information about the system context state, such as the chart type and attributes of the focused data object, as query conditions, and iterates and compares them in the mapping rule base. For example, if the identified standardized gesture operation is a finger swipe to the right, the system context state is a pie chart showing the distribution of players at different levels, and the focused data object is the proportion of advanced players, the rule matching engine will search for rules in the mapping rule base that match these conditions. It will check the gesture operation description and system context state description in each mapping rule one by one. When a rule is found whose specified gesture operation is consistent with the currently identified gesture, and whose system context state also matches perfectly, that rule is determined as the target mapping rule.

[0133] If no exact match is found in the mapping rule base, the rule matching engine will use an approximate matching and priority ranking method. It will filter the closest rules based on the similarity between the gesture and the system context state, and then select the most suitable one as the target mapping rule according to preset priority rules. This ensures that a relatively reasonable mapping rule can be found in various situations, improving the system's adaptability and accuracy.

[0134] Based on the retrieved target mapping rules, operational parameters for game economic data or player behavior trajectories are extracted from the dynamic attributes of standardized gesture operations to form specific operation instructions. The target mapping rules include parameter mapping logic that clarifies how to extract useful information from the dynamic attributes of gesture operations. For example, for a standardized finger swipe gesture, its dynamic attributes include the swipe direction, distance, and speed. If the target mapping rule maps a rightward swipe to an incremental query for a certain indicator in the game economic data, then the parameter mapping logic might stipulate that the increment step size of the query is determined by the swipe distance, and the query priority is affected by the swipe speed.

[0135] When extracting operation parameters, the rule matching engine acquires the dynamic attribute values ​​of standardized gesture operations in real time. For example, it obtains the actual distance and speed data of finger swiping through a motion sensing device. Then, it processes and transforms this data according to the parameter mapping logic in the target mapping rules. If the swiping distance is 5 centimeters, and according to the rule that each centimeter corresponds to an incrementing step of 10 units, then the query incrementing step would be 50 units.

[0136] Finally, the extracted and transformed operation parameters are combined with the operation types specified in the target mapping rules to form specific operation instructions. For example, if the operation type is to query the sales volume of a certain virtual item in the game, and the operation parameter is that the sales volume has increased by 50 units in the past week, then the final specific operation instruction would be "Query the relevant data on the increase of 50 units in the sales volume of this virtual item in the past week." This specific operation instruction can be understood and executed by the game data analysis system or related applications, thereby achieving precise manipulation of game economic data or player behavior patterns.

[0137] In some embodiments, step S105 above, which involves inputting the gesture sequence formed by the specific operation instruction into the gesture intent prediction engine to predict the user's subsequent operation target, specifically includes: Record the specific operation instructions generated by the gestures and their corresponding system context to form a gesture sequence; Based on gesture sequences, high-order intent features reflecting user operation patterns and analysis focus shift paths are extracted through feature calculation units; The high-order intent feature sequence is input into the gesture intent prediction model based on the attention mechanism. By calculating the importance weight of different operations in the high-order intent feature sequence to the current prediction, a probabilistic prediction of the target of subsequent operations is generated. Based on the subsequent operation target with the highest weight in the probabilistic prediction, the corresponding system preloading or interface guidance operation is triggered.

[0138] In this embodiment, every specific operation command generated by the user through gestures is recorded, and the system context corresponding to each specific operation command is captured. The system context covers multiple aspects, such as the type of game data chart currently displayed (bar chart, line chart, etc.), the game business module being analyzed (economic system, player behavior, etc.), and the user's current operation permission level. Taking querying player payment rate as an example, the system context may include that the currently displayed monthly payment rate line chart is the payment module in the game economic system being analyzed, and the user's operation permission is that of a regular analyst. The specific operation commands and their corresponding system contexts are organized and stored in chronological order to form an ordered gesture sequence.

[0139] Based on the recorded gesture sequences, the system activates the feature calculation unit to extract high-order intent features reflecting user operation patterns and analysis focus shift paths. The feature calculation unit is a complex analysis module that utilizes various data analysis techniques to deeply mine the gesture sequences. For user operation patterns, the feature calculation unit analyzes the frequency, order, and combination of user gesture operations. For example, if a user frequently performs a data query operation first, and then filters and sorts the query results, this frequent operation sequence constitutes an operation pattern. The feature calculation unit identifies and records these patterns, incorporating them as part of the high-order intent features. Regarding the analysis of focus shift paths, the feature calculation unit focuses on the user's switching between different data metrics and business modules. For instance, a user might first focus on player online time data, and then quickly shift to player payment behavior data; this focus shift path reflects changes in the user's analytical thinking and focus. The feature calculation unit extracts features of the analysis focus shift path by analyzing the changes in data metrics and business modules involved in the operation commands within the gesture sequence. Through the above analysis, the feature calculation unit integrates and abstracts the extracted features reflecting user operation patterns and analysis focus shift paths, forming a high-order intent feature sequence. These higher-order intent features can more comprehensively and deeply reflect users' operational intentions and analytical needs, providing a crucial basis for subsequent intent prediction.

[0140] The extracted high-order intent feature sequence is input into a gesture intent prediction model based on an attention mechanism, which is the core of the entire prediction system. The attention mechanism is a technique that simulates human attention allocation, automatically calculating the importance weights of different operations within the high-order intent feature sequence for the current prediction. In the gesture intent prediction model, the attention mechanism analyzes and evaluates each feature in the high-order intent feature sequence. For example, if a user has recently frequently performed actions related to player payments, and the focus of analysis has gradually shifted from overall payment rates to the payment behavior of specific player groups, then the high-order intent features related to these actions will be assigned higher importance weights. The model then uses these importance weights to probabilistically predict the user's subsequent action goals. It comprehensively considers the influence of all high-order intent features and calculates the probability of each possible subsequent action goal occurring. For example, the model might predict that the user's next step has a 60% probability of conducting in-depth analysis of the payment behavior of a specific player group, a 30% probability of visualizing payment data, and a 10% probability of performing other related actions. In this way, the gesture intent prediction model can generate comprehensive and accurate probabilistic predictions of the user's subsequent operation goals, providing a reliable basis for the system to take corresponding measures.

[0141] Based on the subsequent action target with the highest weight in the probabilistic prediction, the system will promptly trigger corresponding system preloading or interface guidance operations. System preloading is a technique to optimize user experience. It loads data and resources related to the predicted subsequent action target in advance before the user actually performs an action. For example, if it is predicted that the user's next step will be to conduct in-depth analysis of the spending behavior of a specific player group, the system will load the player group's spending data from the database in advance, including detailed information such as spending amount, spending frequency, and payment channels, and store this data in the cache. When the user actually initiates the analysis operation, the system can directly retrieve the data from the cache, greatly shortening the data loading time and improving the system's response speed.

[0142] User-guided interface operations involve adjusting the interface layout and displayed content to guide users through subsequent predictive actions more easily. For example, if the system predicts that a user will perform a visualization of paid data, it will highlight relevant buttons and options on the interface, such as "Generate Bar Chart" and "Generate Line Chart," placing these buttons in easily accessible locations. Simultaneously, the system can display prompts on the interface to guide users on how to perform visualization operations, helping them complete their target actions more quickly.

[0143] By triggering system preloading or interface-guided operations, the system can prepare users for operations in advance, improving user efficiency and experience, and making the entire game data analysis process smoother and more efficient.

[0144] In some embodiments, step S106 above, which involves automatically triggering and executing a corresponding analysis process based on the gesture sequence and the predicted subsequent operation target, and visually displaying the output data of the analysis process in real time using a 3D heatmap, specifically includes: Based on the gesture sequence and the predicted subsequent operation target, a corresponding executable analysis pipeline is dynamically assembled by querying the pre-set analysis component library and process template; Get the current data view and filter status, use the current data view and filter status as input, trigger and execute the analysis pipeline, and obtain the analysis result dataset; The analysis results dataset is input into a 3D graphics rendering engine, and a 3D heatmap reflecting the inherent relationships of the data is generated through spatial mapping and color height mapping algorithms. The 3D heatmap is presented in a visual interface with smooth animation transitions.

[0145] In this embodiment, a pre-built analysis component library and a process template library are provided. The analysis component library includes various components for data analysis, such as data aggregation components, which can perform aggregation operations on game economic data or player behavior trajectory data from different dimensions, such as aggregating player spending amounts by time period (day, week, month); data filtering components, which can filter data that meets specific requirements based on specific conditions, such as filtering player data whose spending amounts exceed a certain threshold; and data analysis algorithm components, such as linear regression algorithm components for analyzing player behavior trends. The process template library stores a variety of common analysis process frameworks, such as process templates for player retention rate analysis, which include preset processes for data collection, data cleaning, retention rate calculation, and result display.

[0146] Once the gesture sequence and the predicted subsequent operation target are obtained, a query mechanism is initiated. For example, if the gesture sequence shows that the user has performed multiple operations related to player spending behavior, and the predicted subsequent operation target is to conduct an in-depth analysis of the spending preferences of high-spending players within a specific time period, the system will filter out components related to spending data analysis from a pre-set analysis component library, such as a data filtering component (used to filter high-spending players) and a spending preference analysis component (analyzing players' spending on different types of virtual items). Simultaneously, a suitable workflow template for spending preference analysis will be found in the workflow template library. This template specifies the execution order and data flow of each component. Then, based on the queried components and template, the system dynamically assembles these components according to the order specified in the template, forming a complete and executable analysis pipeline. This analysis pipeline is like a customized production line, capable of processing and analyzing game data according to specific needs.

[0147] After assembling the analysis pipeline, it's necessary to obtain the current data view and filter status as input to trigger and execute the analysis pipeline. The current data view refers to the data display format currently seen by the user in the game data analysis interface, such as a bar chart showing the distribution of player spending or a line chart showing changes in player online time. The filter status records the filtering conditions applied by the user, such as filtering for players of a specific level or from a specific region.

[0148] After integrating these current data views with the filtering status information, they are input into a dynamically assembled analysis pipeline. The analysis pipeline then executes each analysis component in a preset order. For example, the data filtering component first filters the raw data based on the filtering status, removing data that does not meet the criteria; then the data aggregation component aggregates the filtered data, calculating statistical information such as the total spending amount of high-spending players within a specific time period; next, the consumption preference analysis component analyzes this aggregated data to determine the consumption preference ratios of high-spending players for different types of virtual items. After processing by a series of components, the final analysis result dataset is obtained. This dataset contains valuable information after in-depth analysis, such as the distribution of consumption preferences of high-spending players and the trend of changes in spending amounts within a specific time period, providing a data foundation for subsequent visualization.

[0149] The resulting dataset is then input into a 3D graphics rendering engine. This engine is a powerful visualization tool that uses spatial mapping and color height mapping algorithms to process the dataset. The spatial mapping algorithm maps each data point in the dataset to a specific location in 3D space. For example, if the dataset contains consumer preference data for different types of virtual items, the algorithm will distribute the data points across different areas of 3D space based on the item type, creating a hierarchical and structured spatial layout.

[0150] The color-height mapping algorithm assigns a corresponding color and height to each data point based on its size or importance. For example, for virtual items with a high consumer preference rate, the corresponding data point will be given a brighter color and a higher height to highlight its importance in the data. Through the synergy of these two algorithms, the 3D graphics rendering engine can transform the analysis result dataset into a 3D heatmap that intuitively reflects the inherent relationships within the data. In this 3D heatmap, regions of different colors and data points at different heights clearly demonstrate the distribution, trends, and relative relationships between the data, allowing users to more intuitively understand the analysis results.

[0151] After generating the 3D heatmap, the system needs to present it to the user in a visualization interface. To enhance the user's visual experience, the system uses smooth animation transitions to display the 3D heatmap. When the 3D heatmap is ready to be displayed in the visualization interface, the system does not display it abruptly, but instead achieves a smooth transition through a series of animation effects.

[0152] For example, the system can start with a small, blurry image and gradually zoom in and sharpen the 3D heatmap, giving users a visual experience of moving from far to near and from blurry to clear. Simultaneously, rotation and scaling animations can be added during the zoom-in process, allowing users to observe the 3D heatmap from different angles and gain a more comprehensive understanding of the data distribution. Furthermore, the system can add dynamic color changes or height adjustments based on changes in the data within the 3D heatmap, making the data presentation more vivid and intuitive. Through these smooth animation transitions, users can view the 3D heatmap more comfortably and clearly, better understand the information reflected in the analysis results, and thus improve the user experience throughout the entire game data analysis process.

[0153] Reference Figure 2 An embodiment of the present invention provides a game data multidimensional correlation analysis and visualization system 2, the system 2 specifically including: The knowledge graph module 201 is used to construct a knowledge graph for the game industry based on the user behavior pattern library, the paid conversion path library, and the indicator association rule library. The indicator recommendation module 202 is used to automatically recommend indicator association schemes based on the association rules generated by the game industry knowledge graph and through the recommendation engine. The gesture operation module 203 is used to collect the user's somatosensory gesture data, input the somatosensory gesture data and indicator association scheme into the pre-trained gesture recognition model, and recognize standardized gesture operations. The instruction mapping module 204 is used to map the recognized standardized gesture operations into specific operation instructions for game economic data or player behavior trajectories through a preset mapping rule library. The intent prediction module 205 is used to input the gesture sequence formed by the specific operation instructions into the gesture intent prediction engine to predict the user's subsequent operation target; The data display module 206 is used to automatically trigger and execute the corresponding analysis process based on the gesture sequence and the predicted subsequent operation target, and to visualize the output data of the analysis process in real time through a 3D heat map.

[0154] It is understandable that, such as Figure 1The content shown in the game data multidimensional correlation analysis and visualization method embodiment is applicable to this game data multidimensional correlation analysis and visualization system embodiment. The specific functions implemented by this game data multidimensional correlation analysis and visualization system embodiment are as follows: Figure 1 The example of multidimensional correlation analysis and visualization of game data shown is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the illustrated multidimensional correlation analysis and visualization method for game data are the same.

[0155] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0156] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0157] Reference Figure 3 The present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, it implements the game data multidimensional correlation analysis and visualization method as described in any of the above methods.

[0158] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0159] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0160] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0161] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the game data multidimensional correlation analysis and visualization method as described in any of the above methods.

[0162] In this embodiment, 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, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0163] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0165] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0166] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for multi-dimensional association analysis and visualization of game data, characterized in that, The method specifically comprises: According to the user behavior pattern library, the paid conversion path library and the index association rule library, a game industry knowledge graph is constructed; Based on the association rules generated by the game industry knowledge graph, an index association scheme is automatically recommended through a recommendation engine; Collecting the user's somatosensory gesture data, inputting the somatosensory gesture data and the index association scheme into a pre-trained gesture recognition model, and identifying standardized gesture operations; Through a pre-set mapping rule library, the identified standardized gesture operations are mapped into specific operation instructions for game economic data or player behavior trajectory; The gesture sequence formed by the specific operation instructions is input into a gesture intention prediction engine to predict the user's subsequent operation target; According to the gesture sequence and the predicted subsequent operation target, the corresponding analysis process is automatically triggered and executed, and the output data of the analysis process is visualized in real time through a 3D heat map.

2. The method of claim 1, wherein, The game industry knowledge graph is constructed according to the user behavior pattern library, the paid conversion path library and the index association rule library, specifically comprising: Collecting multi-source heterogeneous data from game clients, servers and operation databases, cleaning and session dividing the multi-source heterogeneous data to form standardized user behavior data; Based on the user behavior data, core entities are extracted through a pre-defined domain ontology model, and semantic relationships between the core entities are established, the core entities including players, game behaviors and payment events; The core entities and the semantic relationships between them are stored in a graph database, and knowledge fusion is performed through entity linking technology to construct an initial knowledge graph containing the user behavior pattern library and the paid conversion path library; According to historical analysis cases, the index relationship is determined from the initial knowledge graph through graph mining algorithm to construct an index association rule library; The index association rule library is injected into the initial knowledge graph to form the game industry knowledge graph.

3. The method of claim 1, wherein, Based on the association rules generated by the game industry knowledge graph, an index association scheme is automatically recommended through a recommendation engine, specifically comprising: In response to the target index selected by the user, the current analysis context information including the business cycle and the user role is obtained through the recommendation engine; Based on the target index and the current analysis context information, multiple candidate index association schemes are generated by real-time traversal of the game industry knowledge graph; The candidate index association schemes are fused and de-duplicated, and the recommendation scores of each candidate index association scheme are calculated to generate a recommendation scheme ranking list; The recommendation scheme ranking list is pushed to the user, and the recommendation strategy is optimized and updated according to the user's adoption feedback.

4. The method of claim 1, wherein, The somatosensory gesture data of the user is collected, the somatosensory gesture data and the index association scheme are input into a pre-trained gesture recognition model, and standardized gesture operations are identified, specifically comprising: Collecting the user's somatosensory gesture data, based on the somatosensory gesture data, extracting gesture feature vectors including spatial configuration and motion trajectory through a feature calculation engine; A large number of gesture samples labeled with analysis context are obtained, and a hybrid model constructed based on convolutional neural network and recurrent neural network is trained through the gesture samples to generate a gesture recognition model; Encode the analysis context information represented by the index association scheme into a context feature vector, splice the context feature vector and the gesture feature vector, and input them into a gesture recognition model; Output the standardized gesture operation corresponding to the current analysis context information through the gesture recognition model.

5. The method of claim 1, wherein, The standardized gesture operation is mapped to a specific operation instruction for the game economic data or the player behavior trajectory through a preset mapping rule library, specifically including: Construct a mapping rule library, which includes a plurality of mapping rules for associating the standardized gesture operation and the system context state to a specific operation instruction; When receiving the identified standardized gesture operation, determine the system context state including the current chart type and the focus data object through a rule matching engine; Based on the standardized gesture operation and the system context state, jointly query in the mapping rule library to retrieve the matched target mapping rule; According to the parameter mapping logic set in the target mapping rule, extract the operation parameters for the game economic data or the player behavior trajectory from the dynamic attributes of the standardized gesture operation to form a specific operation instruction.

6. The method of claim 5, wherein, The gesture sequence formed by the specific operation instruction is input into a gesture intention prediction engine to predict the subsequent operation target of the user, specifically including: Record the specific operation instruction generated by the gesture and its corresponding system context to form a gesture sequence; Based on the gesture sequence, extract high-order intention features reflecting the user operation mode and the analysis focus shift path through a feature calculation unit; Input the high-order intention feature sequence into a gesture intention prediction model based on an attention mechanism, calculate the importance weight of different operations in the high-order intention feature sequence for the current prediction, and generate a probabilistic prediction of the subsequent operation target; According to the subsequent operation target with the highest weight in the probabilistic prediction, trigger the corresponding system preloading or interface guiding operation.

7. The method of claim 1, wherein, According to the gesture sequence and the predicted subsequent operation target, automatically trigger and execute the corresponding analysis process, and visualize the output data of the analysis process in real time through a 3D heat map, specifically including: According to the gesture sequence and the predicted subsequent operation target, query the preinstalled analysis component library and process template to dynamically assemble a corresponding executable analysis pipeline; Get the current data view and filtering state, take the current data view and filtering state as input, trigger and execute the analysis pipeline to get the analysis result dataset; Input the analysis result dataset into a 3D graphics rendering engine to generate a 3D heat map reflecting the internal relationship of the data through space mapping and color height mapping algorithms; Present the 3D heat map through a smooth animation transition in the visualization interface.

8. A multidimensional correlation analysis and visualization system for game data, characterized in that, The system specifically includes: A knowledge graph module for constructing a game industry knowledge graph according to a user behavior mode library, a paid conversion path library, and an index association rule library; An index recommendation module for automatically recommending an index association scheme through a recommendation engine based on the association rules generated by the game industry knowledge graph; The gesture operation module is configured to collect somatosensory gesture data of a user, input the somatosensory gesture data and an index association scheme into a pre-trained gesture recognition model, and identify a standardized gesture operation; The instruction mapping module is configured to map the identified standardized gesture operation into a specific operation instruction on game economic data or a player behavior track by using a pre-set mapping rule library; The intention prediction module is configured to input a gesture sequence formed by the specific operation instruction into a gesture intention prediction engine, and predict a subsequent operation target of the user; The data display module is configured to automatically trigger and execute a corresponding analysis process according to the gesture sequence and the predicted subsequent operation target, and visually display output data of the analysis process in real time by using a 3D heat map.

9. A computer device, comprising: The game data multi-dimensional association analysis and visualization method comprises: A memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, the game data multi-dimensional association analysis and visualization method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and when the computer program is executed by a processor, the game data multi-dimensional association analysis and visualization method according to any one of claims 1 to 7 is implemented.

Images