Account allocation method and system for mobile game

Abstract

The invention discloses a mobile game account number distribution method and system, and relates to the technical field of machine learning, and the method comprises the steps: obtaining the multi-dimensional behavior and state telemetry data of an online player in real time; generating a player activity state sequence based on data analysis; constructing and maintaining a multi-dimensional context switching cost model representing switching cost between active states, and collecting negative behavior feedback of players to perform dynamic self-learning optimization; predicting the short-term intention activity state of the player by using the state sequence, the player file and the real-time context; according to the method, server examples are periodically analyzed, micro-ecological portraits reflecting current activity states, transfer modes and the like of the server examples are generated, a core allocation decision engine calculates comprehensive scores of candidate examples by minimizing predicted total context switching cost, and optimal examples are selected to execute allocation. Adaptability of the game environment is enhanced and may facilitate server instances to form activity-based dynamic soft partitions.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and in particular to a method and system for allocating accounts for mobile games. Background Art

[0002] In existing mobile games, account allocation usually refers to the process in which the system allocates players to specific regions or servers based on server load, geographic location or preset rules when they create characters or log in to the game. The main purpose of this allocation method is to ensure balanced utilization of server resources and basic network connection quality. However, this static or system load-based account allocation model has significant limitations. First, it ignores the fact that players' gaming behaviors and interests are constantly changing during the game. Players may experience multiple activity states such as exploration, combat, and social interaction in one session. A server environment that seems suitable at the time of initial allocation may no longer match the player's activity state during the subsequent game process, resulting in a fragmented experience. Secondly, traditional account allocation fails to solve the problems faced by players when switching between different activity states. The increased cognitive load, difficulty in adjusting operations, and sudden changes in the social environment caused by such switching will seriously affect the player's immersion and game fluency, and is a potential factor leading to player fatigue and churn; Therefore, a more effective account allocation strategy should ensure the smoothest and lowest-friction interactive experience regardless of the activity status or transitions of the player account. To fundamentally improve the gaming experience tied to the player account, the concept of account allocation needs to be expanded and deepened beyond the initial access, extending to the player's entire gaming session, and dynamically and intelligently managing the actual gaming environment of the player account. Summary of the invention

[0003] The present invention provides a method and system for allocating accounts of mobile games to solve the defects of existing account allocation strategies that are static and passive and ignore the continuity of players' dynamic experience.

[0004] In a first aspect, the present invention provides a method for allocating accounts for mobile games, comprising: Obtain multi-dimensional behavior and status telemetry data related to online players in real time; Based on the telemetry data, parse and generate an activity state sequence with context information representing the player's activity; Predicting the player's short-term intended activity state and its probability distribution based on the activity state sequence, the player profile and the real-time context information; Maintain and dynamically learn a multi-dimensional context switching cost model for switching between activity states; Periodically analyze server instances to generate instance microecological portraits including activity distribution, transfer patterns, and player composition; In response to the environment allocation request, for each candidate server instance, the following steps are performed: calculating the predicted total context switching cost of switching from the current activity state of the online player and the predicted short-term intended activity state to the representative activity state of the candidate server instance according to the maintained cost model; obtaining the microecological portrait and server performance constraint information of the candidate server instance; obtaining a preset strategy factor; and calculating the comprehensive score of the candidate server instance based on the predicted total context switching cost, the microecological portrait, the server performance constraint information and the strategy factor; An optimal target server instance is selected based on the comprehensive score, and an operation of allocating players to the optimal target server instance is performed.

[0005] In a second aspect, the present invention further provides a system for allocating accounts for mobile games, the system comprising: A real-time multi-dimensional behavior perception module configured to capture and initially process telemetry data related to online player behavior, status, interaction, and environment in real time; A player activity state parsing and serialization engine, connected to the real-time multi-dimensional behavior perception module, configured to parse the behavior event stream into discrete activity states based on the telemetry data using predefined rules and / or machine learning models, and maintain an activity state sequence with context information for each player; A multi-dimensional context switching cost knowledge base and dynamic learning module configured to store the multi-dimensional base costs of switching between activity states, and dynamically learn and update the cost model based on collected player behavior feedback data; An enhanced short-term intention prediction service, connected to the player activity state parsing and serialization engine, configured to use the player activity state sequence, personal profile and real-time context information to predict the player's most likely next activity state and its probability distribution; Server instance micro-ecological profile analyzer, configured to periodically analyze player activity distribution, state transition patterns, etc. of active server instances, and generate and maintain dynamic profiles of instances; A core allocation decision engine, connected to the player activity state parsing and serialization engine, the cost knowledge base and dynamic learning module, the enhanced short-term intention prediction service and the server instance microecological portrait analyzer, is configured to comprehensively calculate the predicted total context switching cost, performance penalty and strategy adjustment items of each candidate instance when players need to be allocated, generate a comprehensive score, and select the optimal target server instance accordingly; The allocation execution and closed-loop feedback interface connects the core allocation decision engine and related server management modules, is configured to execute the final allocation decision, and collect allocation execution results and player subsequent behavior feedback for the cost knowledge base and dynamic learning module, the enhanced short-term intention prediction service and the core allocation decision engine to perform closed-loop optimization of parameters or models.

[0006] The technical solution provided by this application has at least the following technical effects or advantages: By taking the difficult-to-quantify context switching cost as the core optimization target and performing multi-dimensional modeling and minimization, the cognitive burden, operational barriers and time loss of players when switching between different activities and environments are significantly reduced, making the game process more coherent and natural. The environment can be intelligently allocated based on the real-time and dynamic behavior patterns and intentions of players, as well as the actual "micro-ecology" of server instances, so that the game world can better adapt to the actual needs of players and the evolution of gameplay, rather than relying on rigid preset partitions; The cost model and decision-making process take into account players' personalized factors such as proficiency, preferences, and fine-grained activity status, so that allocation decisions can better meet the needs of individual players and enhance personalized experience; promote efficient utilization of server resources and community self-organization: by guiding players with similar activity flows together, "soft partitions" based on activity patterns may be spontaneously formed, which helps to optimize the resource allocation of specific instances in a targeted manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A flow chart of a method for allocating accounts for mobile games according to the present invention; Figure 2 A system architecture diagram for account allocation for a mobile game. DETAILED DESCRIPTION

[0008] The present invention relates to a method and system for allocating accounts for mobile games. The purpose is to achieve intelligent allocation of game environments and optimize the fluency of player experience by dynamically evaluating and minimizing the multi-dimensional context switching costs generated by players switching between different game activity states. The system is usually integrated into the back-end architecture of the game server.

[0009] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0010] 1. As Figure 1 A method for allocating accounts for mobile games is shown in a flow chart, and the method includes the following steps: Obtain multi-dimensional behavior and status telemetry data related to online players in real time; Based on telemetry data, parse and generate activity state sequences with contextual information that represent player activities; Predict the player's short-term intention activity state and its probability distribution based on the activity state sequence, player profile and real-time context information; Maintain and dynamically learn a multi-dimensional context switching cost model for switching between activity states; Periodically analyze server instances to generate instance microecological portraits including activity distribution, transfer patterns, and player composition; In response to the environment allocation request, for each candidate server instance, the following are performed: based on the maintained cost model, the predicted total context switching cost of switching from the current activity state of the online player and the predicted short-term intention activity state to the representative activity state of the candidate server instance is calculated; the microecological portrait and server performance constraint information of the candidate server instance are obtained; the preset strategy factor is obtained; based on the predicted total context switching cost, the microecological portrait, the server performance constraint information and the strategy factor, the comprehensive score of the candidate server instance is calculated; The optimal target server instance is selected based on the comprehensive score, and the player is assigned to the optimal target server instance.

[0011] like Figure 2 As shown in the system architecture diagram of a mobile game account allocation, the system includes: A real-time multi-dimensional behavior perception module configured to capture and initially process telemetry data related to online player behavior, status, interaction, and environment in real time; A player activity state parsing and serialization engine, connected to the real-time multi-dimensional behavior perception module, configured to parse the behavior event stream into discrete activity states based on telemetry data using predefined rules and / or machine learning models, and maintain an activity state sequence with context information for each player; A multi-dimensional context switching cost knowledge base and dynamic learning module, configured to store the multi-dimensional base costs of switching between activity states, and dynamically learn and update the cost model based on the collected player behavior feedback data; Enhanced short-term intent prediction service, connected to the player activity state parsing and serialization engine, configured to use the player's activity state sequence, personal profile and real-time context information to predict the player's most likely next activity state and its probability distribution; Server instance micro-ecological profile analyzer, configured to periodically analyze player activity distribution, state transition patterns, etc. of active server instances, and generate and maintain dynamic profiles of instances; The core allocation decision engine connects the player activity status parsing and serialization engine, the cost knowledge base and dynamic learning module, the enhanced short-term intention prediction service, and the server instance micro-ecological portrait analyzer. It is configured to comprehensively calculate the predicted total context switching cost, performance penalty, and strategy adjustment items of each candidate instance when players need to be allocated, generate a comprehensive score, and select the optimal target server instance based on it; The allocation execution and closed-loop feedback interface connects the core allocation decision engine and related server management modules, is configured to execute the final allocation decision, and collects allocation execution results and player subsequent behavior feedback for closed-loop optimization of parameters or models by the cost knowledge base and dynamic learning module, enhanced short-term intention prediction service and core allocation decision engine.

[0012] 2. Detailed implementation and operation process of key components of this solution (2.1) Real-time multi-dimensional behavior perception module Data sources: Multi-dimensional behavior and status telemetry data including but not limited to: Player position movement data: continuous stream of coordinates, calculated velocity vectors, zone enter / leave events.

[0013] Skill and item usage records: the skill or item identifier used, the casting / usage time, the target object identifier, and the effect record produced (such as damage value, treatment value, status effect identifier).

[0014] User interface interaction events: click events of specific interface elements (such as buttons and panels), opening / closing events of interface panels, records of dwell time on specific panels, and mouse trajectory data on the interface.

[0015] Communication and social data: text messages from each chat channel (public, team, guild, etc.), keywords extracted from them, preliminary judgment of emotional tendencies, task status change records (accepted, completed, abandoned, etc.), team / guild operation records (creation, joining, leaving, etc.), transaction behavior records, and player interaction marks (such as adding / deleting friends).

[0016] Combat log data: Detailed combat records, including damage / damage events, damage sources and values, kill / kill events, and Player versus Player (PvP) or Player versus Environment (PvE) combat results.

[0017] Production and economic data: resource collection activities (types, quantities), item crafting activities (recipe identifiers, results), market transaction records.

[0018] Data preprocessing: Timestamp alignment: Ensure that event data from all sources are sorted based on a unified time base.

[0019] Data cleaning: Identify and process abnormal data points, such as invalid coordinates, illogical values, etc., by setting threshold ranges, logical checks, etc.

[0020] Preliminary feature extraction: Calculate derived metrics, for example, calculate the number of operations per minute in a rolling time window to quantify the intensity of interaction; calculate movement entropy to distinguish goal-oriented movement from staying in place; count the frequency of social interactions in a short period of time, etc.

[0021] Data flow: Preprocessed structured or semi-structured event data is distributed to downstream processing modules through high-throughput distributed message queues (such as systems implemented using Kafka or Pulsar technology).

[0022] (2.2) Player activity status parsing and serialization engine Hierarchical status system definition: pre-define a hierarchical system that can accurately describe the player's activity status.

[0023] Macro-activity states: Define high-level activity categories that are clearly distinguished in business logic, such as exploration, player-to-environment combat, player-to-player combat, social interaction, item crafting, and casual idle play, etc., which are customized according to the game content.

[0024] Micro-activity states: Subdivide some macro-states to capture finer context switching differences. For example, subdivide "player battle" into: waiting in the arena queue, fighting in the arena, patrolling the outdoor player battle area, fighting in the outdoor player battle area, etc.

[0025] Transition state: A state defined to describe a fuzzy behavior pattern or a state transition process, such as post-battle cooldown, movement between areas, etc., to avoid making premature state judgments when there is insufficient information.

[0026] State parsing logic (hybrid approach): Rule Engine: Maintains the current state for each player. Based on the received behavioral events and context information (such as player location area type, activated task identifier, team status, surrounding entity type, current user interface status, etc.), matches the preset "IF-THEN" rule set to determine the state transition. The rules need to cover the main state entry and exit conditions.

[0027] Confidence calculation: For each new state that may be triggered, a confidence score (in the range of 0-1) is calculated based on the strength of the triggering rule, the amount of supporting evidence, and its consistency. For example, if the location only enters the player-versus-player area, the confidence of the "player-versus-player" state is low; if it is subsequently detected that the target selects an enemy player and uses a player-versus-player skill, the confidence will increase significantly.

[0028] Fuzzy processing: If the highest confidence is lower than the preset "state determination confidence threshold", or the confidence of multiple states is close, it is determined as the corresponding transition state or marked as "state uncertain".

[0029] Machine learning assistance: Models suitable for processing sequence data (such as models based on recurrent neural networks or transformer architectures) can be selected to analyze recent behavioral event sequences and assist in identifying complex activity patterns that are difficult to cover with rules. The output can be used as additional input features for confidence calculations.

[0030] Serialization and context storage: Maintain a time-ordered list of state sequence records for each player: [(entry time 1, state identifier 1, confidence 1, duration 1, context information 1), (entry time 2, state identifier 2, confidence 2, duration 2, context information 2),...].

[0031] The "context information" includes a snapshot of the key environment when the state occurs, such as area identifiers, task identifiers, team identifiers, key role attribute values, etc.

[0032] Sequence records retain data from the most recent period of time (such as 15 minutes) or the most recent number of states (such as 20). The specific window size is configurable.

[0033] (2.3) Multi-dimensional context switching cost knowledge base and dynamic learning module Data structure: Basic cost matrix: stores the basic switching cost from state A to state B, and stores evaluation values ​​(such as 1-5 points) according to dimensions such as cognitive load, operational complexity, social adaptation, and preparation time.

[0034] Personalization Factor Model: Stores the rules or model parameters used to adjust the base cost, such as a lookup table based on player proficiency level, or a small regression model trained on player profile characteristics.

[0035] Dynamic adjustment record: records the cost revision history triggered by the learning mechanism.

[0036] Cost calculation logic: Calculate the cost for a specific player to switch from state A (context A) to state B (context B): Get the basic cost value of each dimension from A to B from the basic cost matrix.

[0037] Apply the personalized factor model to calculate the personalized adjustment factors of each dimension based on the player profile and context A and B.

[0038] Apply the adjustment factor to the base cost to get the personalized dimension cost.

[0039] According to the preset dimension weights (which reflect the relative importance of each dimension and are configurable), the personalized dimension costs are weighted and summed to obtain the total personalized switching cost.

[0040] Dynamic learning process (periodic background tasks): Data aggregation: Collect state transition instances within a period (such as every week) and their subsequent behavior feedback (such as offline, fast switching back, negative evaluation labels, etc.).

[0041] Statistical analysis: Calculate the average negative indicator rate after each A to B transition and compare it with the baseline for statistical significance (e.g., using a T-test or chi-square test).

[0042] Cost update: If the negative indicator rate of a conversion is significantly higher than the baseline, it indicates that the current cost assessment may be too low. According to the preset "significance-cost increase" mapping rule, the corresponding cost value in the basic cost matrix is ​​adjusted up.

[0043] Record updates: record update operations in the dynamic adjustment record. Learning cycle, significance threshold, and mapping rules are all configurable parameters.

[0044] (2.4) Enhanced short-term intention prediction service Input feature engineering: Integrate multi-source information as prediction input: Sequence features: extracted from the player's activity state sequence, such as the last N states, state duration, state transition frequency, etc.

[0045] Static features: extracted from player profiles, such as level, occupation, game time, achievements, strength of social relationships, etc.

[0046] Real-time contextual features: extracted from the current game environment, such as location type, active task type, nearby friend status, world event information, etc.

[0047] All features are encoded, normalized, and concatenated into a comprehensive feature vector.

[0048] Multi-model fusion prediction: Model library: Maintains a set of prediction models, such as models suitable for processing sequence data, gradient boosting decision tree models suitable for processing tabular mixed features, and rule engines based on strong context triggering.

[0049] Fusion strategy: Run multiple models in parallel to obtain the probability distribution of their next states. Calculate the context strength score. Dynamically adjust the weights of each model based on the base weight and the context strength score. The final probability is obtained by weighted summing the output probabilities of each model: Final probability (state S_j) = Σ(adjusted weight (model k) * model k predicted probability (S_j)).

[0050] Output: Returns a list of the K (e.g. K=3) highest probability (predicted state, predicted probability) pairs.

[0051] (2.5) Server Instance Microecological Profile Analyzer Calculation cycle and sliding window: Configure the update cycle (such as 30-60 seconds) and statistical time window (such as 5-10 minutes) to balance real-time performance and stability.

[0052] Image dimension calculation: Periodically calculate the following for each active instance: Activity status distribution: the percentage of players in each macro or micro status.

[0053] Mainstream state transition matrix: the frequency or probability of transition from state A to state B within an instance.

[0054] Player composition: the proportion of player groups of different levels, professions, etc.

[0055] Social indicators: average team size, percentage of guild members, frequency of public channel messages, etc.

[0056] Activity stability: the frequency or magnitude of recent changes in the set of mainstream activity states.

[0057] Average switching cost friendliness (estimated): Samples and calculates the average cost for a player to switch from the current state to the mainstream state of the instance.

[0058] Storage and query: Store the portrait data in an efficient key-value storage system (such as an in-memory database) or a time series database for quick query by the decision engine.

[0059] (2.6) Core allocation decision engine: Request reception and input validation: Receive allocation requests, verify player identifiers, trigger reasons, etc. Check the availability of required input data (player state sequence, intent prediction, cost model, instance list and profile, performance data), and handle missing cases (such as using default values, rejecting requests).

[0060] Pre-screening of candidate instances: Apply hard constraints (load limit exceeded, latency too high, under maintenance, region / version mismatch, etc.) to remove unqualified instances and obtain a list of valid candidate instances. If the list is empty, the logic of not being able to allocate is executed.

[0061] Iteratively evaluate each valid candidate instance (Inst_i): (1) Determine the representative activity state of the instance (DS_i): Select the activity state with the highest current proportion and is relatively stable from the portrait of the instance Inst_i.

[0062] (2) Calculate the total context switch cost (TotalContextCost_i): Calculate the current switch cost (Cost_Current_i): Call the cost calculation logic to calculate the personalized cost of switching from the player's current state to DS_i. Calculate the expected future switch cost (ExpectedCost_Next_i): For each predicted future state PS_j (probability Prob_j), calculate the cost Cost_Predict_ji of switching from PS_j to DS_i, then accumulate Prob_j*Cost_Predict_ji, and combine the total cost: TotalContextCost_i=Weight_Current*Cost_Current_i+Weight_Future*ExpectedCost_Next_i (weights are configurable).

[0063] (3) Calculate performance penalty (LoadPenalty_i, LatencyPenalty_i): Apply a predefined nonlinear penalty function that distinguishes between comfort zone, warning zone, and rejection zone, and calculate the penalty score based on instance load and player network latency.

[0064] (4) Calculate the policy adjustment item (PolicyAdjustment_i): Calculate a small adjustment score based on the system strategy (such as promoting stability, encouraging exploration, and avoiding low-interaction traps).

[0065] (5) Calculate the final comprehensive score (FinalScore_i): Normalize TotalContextCost_i, LoadPenalty_i, LatencyPenalty_i, and PolicyAdjustment_i. Apply the preset weights W_Context, W_Load, W_Latency, and W_Policy to perform weighted summation on the normalized values ​​to obtain FinalScore_i.

[0066] (6) Store evaluation results: record instance Inst_i and its calculation results and final score.

[0067] Final decision generation: Sort the result list in ascending order by FinalScore_i.

[0068] Apply decision rules: Recheck the hard constraints of the top-ranked instances.

[0069] Select the instance Inst_best with the lowest score.

[0070] Check whether the TotalContextCost of Inst_best is lower than the "maximum acceptable context switching cost threshold". If it is higher, trigger the alternative logic (select suboptimal, allocate to neutral zone, fail).

[0071] Preferably, if the optimal and suboptimal scores are close, the final selection is made based on the stability or diversity strategy.

[0072] Record decision logs: record selection results, scoring details, and special logic triggered in detail.

[0073] Output decision: Returns the final selected instance identifier.

[0074] (2.7) Allocation execution and closed-loop feedback interface Execute: Passes the selected instance identifier and player identifier to the server management service to perform instance allocation or migration.

[0075] Different server instances will gradually evolve their unique "microecological portraits" and enrich the player groups with specific activity mode preferences. For example, there may emerge "exploration-type" instances mainly composed of players who conduct deep exploration and light combat, "hardcore PVE-type" instances that focus on high-intensity team dungeon challenges, "economic / social center-type" instances that focus on manufacturing, trading and socializing, or "competitive" instances that focus on specific types of PvP activities.

[0076] Record whether the allocation is successful or not, the player's initial behavior in the new instance, the length of stay, the subsequent state sequence, the associated player's subjective feedback (ratings, complaints, etc.), and the macro soft partition effect indicators.

[0077] The collected feedback data is used to drive dynamic learning of cost models, optimize prediction models, tune decision engine weights / thresholds, and verify the effectiveness of the overall solution.

[0078] 3. Initialization, deployment and maintenance Initialization: It is necessary to define the state system, set the initial cost matrix (which can be done through expert evaluation or small-scale testing), configure the basic weights and thresholds, and train the initial prediction model.

[0079] Deployment: It is recommended to adopt a microservice architecture, deploy each component as an independent service, and interact through application programming interfaces and message queues.

[0080] Maintenance and monitoring: Establish a monitoring system to monitor module status, algorithm performance (prediction accuracy, cost convergence), business indicators (allocation success rate, session duration), and system resource consumption. Regularly retrain models and optimize parameters. Keep detailed logs for analysis and troubleshooting.

[0081] This specific implementation method elaborates on the internal operation logic of each main functional module, aiming to enable those skilled in the art to understand and implement the detailed basis and instructions. It should be emphasized that the above description constitutes a specific and preferred embodiment, but the concept of the present invention is not limited to this. Any equivalent transformation, modification or improvement based on the core spirit of the present invention and not departing from the technical principles and scope disclosed in this specification, as long as the same or similar technical effects can be achieved, should be considered to fall within the scope of protection claimed by the present invention.

Claims

1. A method for allocating accounts for mobile games, characterized in that: include: Obtain multi-dimensional behavior and status telemetry data related to online players in real time; Based on the telemetry data, parse and generate an activity state sequence with context information representing the player's activity; Predicting the player's short-term intended activity state and its probability distribution based on the activity state sequence, the player profile and the real-time context information; Maintain and dynamically learn a multi-dimensional context switching cost model for switching between activity states; Periodically analyze server instances to generate instance microecological portraits including activity distribution, transfer patterns, and player composition; In response to the environment allocation request, for each candidate server instance, executing: calculating, based on the maintained cost model, a predicted total context switching cost of switching from the current activity state of the online player and the predicted short-term intended activity state to the representative activity state of the candidate server instance; obtaining a microecological profile and server performance constraint information of the candidate server instance; Get the preset strategy factor; Calculate a comprehensive score of the candidate server instance based on the predicted total context switching cost, the microecological portrait, the server performance constraint information, and the policy factor; An optimal target server instance is selected based on the comprehensive score, and an operation of allocating players to the optimal target server instance is performed.

2. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: The analysis generates an activity state sequence with context information representing the player's activity, including: Predefine a multi-level player activity status system covering macro activities, micro activities and transition states; Applying parsing logic based on a rules engine or a machine learning model to convert the telemetry data stream into discrete activity state events; A time-ordered sequence record is maintained for each online player, wherein the sequence record contains key context information of the activity state, state confidence, duration, and triggering state, constituting the activity state sequence.

3. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: The multi-dimensional context switching cost model for maintaining and dynamically learning switching between activity states includes: Build a knowledge base to store the basic costs of switching between activity states, including cognitive, operational, social, and preparation time dimensions; Build a personalized model for adjusting base costs based on player personalization factors; A cost calculation function is defined, which outputs a multi-dimensional personalized switching cost based on the player profile, source state, target state and its context, combined with the basic cost and the personalized model; By collecting behavioral feedback data after players switch states, statistical analysis is performed to identify negative indicator changes related to switching costs; According to the significance of the change of the negative indicator, a preset cost adjustment mapping rule is applied to dynamically update the basic switching cost in the knowledge base.

4. A method for allocating mobile game accounts as claimed in claim 3, characterized in that: According to the significance of the change of the negative indicator, applying the preset cost adjustment mapping rule to dynamically update the basic switching cost in the knowledge base includes: Aggregate player state transition instances and their subsequent associated behavioral feedback data within a specified period, wherein the feedback data includes indicators reflecting changes in player experience or retention; Calculate the average incidence of negative indicators after a specific state transition type and compare it with the baseline for statistical significance; Establishing a mapping relationship from the significant increase in the incidence of the negative indicator to the increase in switching cost; According to the statistical significance comparison result and the mapping relationship, the basic cost value of the corresponding state transition in the knowledge base is adjusted.

5. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: The predicted player's short-term intention activity state includes: Extracting temporal features, static features and contextual features from the activity state sequence, player profile and real-time game context; Build a prediction model library that includes at least one of a sequence model, a table model, and a rule engine; Adopt a multi-model fusion strategy to dynamically select a model based on the input features or perform weighted fusion on the probability distribution of multiple model outputs, where the weights can be adjusted dynamically; Output a list containing several future activity states with the highest probability and their corresponding probabilities to determine the short-term intention activity state and its probability distribution.

6. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: Analyze the microecological portrait of the example, including: Calculate and update the following indicators for each active server instance within the preset statistical time window: the proportion of players in each activity state, the transition frequency or probability matrix between mainstream states, the composition ratio of player groups with different characteristics, indicators reflecting social activity, stability measures of activity patterns, and estimated average instance switching cost friendliness; The indicator set is stored as a dynamic microecological portrait of the server instance for query during allocation decision-making.

7. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: Calculate the comprehensive score of the candidate server instance, for each valid candidate instance, including: Determine the current representative activity state of the instance microecological profile from the instance microecological profile; Calling the cost calculation function to calculate the current switching cost of the player switching from the current state to the representative activity state; In combination with the predicted short-term intended activity states and their probability distributions, the expected future switching costs of switching from these predicted states to the representative activity state are calculated; The current switching cost and the expected future switching cost are combined according to a preset weight strategy to obtain a total context switching cost evaluation value of the candidate instance.

8. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: Calculating the comprehensive score of the candidate server instance further includes: Calculating a performance penalty score based on the real-time performance indicator of the candidate server instance; Calculate the strategy adjustment score based on the preset strategy; Normalizing the predicted total context switch cost evaluation value, the performance penalty score, and the strategy adjustment score; The preset dimension weights are applied to perform weighted summation on the normalized predicted total context switch cost evaluation value, the performance penalty score, and the strategy adjustment score to obtain the final comprehensive score.

9. A method for allocating mobile game accounts as claimed in claim 1, characterized in that: The method further comprises: Collect the success or failure records of the allocation execution, the initial behavior of the player after being assigned to the target instance, the length of stay, the subsequent state transition sequence, and possible player subjective feedback data; The collected feedback data is used to perform closed-loop optimization and adjustment on the multi-dimensional context switching cost model, the short-term intention prediction model, the instance microecological portrait analysis logic, and the weights and parameters in the comprehensive score.

10. A system for allocating accounts for mobile games, characterized in that: The system is configured to perform the method according to any one of claims 1 to 9, and the system comprises: A real-time multi-dimensional behavior perception module configured to capture and initially process telemetry data related to online player behavior, status, interaction, and environment in real time; A player activity state parsing and serialization engine, connected to the real-time multi-dimensional behavior perception module, configured to parse the behavior event stream into discrete activity states based on the telemetry data using predefined rules and / or machine learning models, and maintain an activity state sequence with context information for each player; A multi-dimensional context switching cost knowledge base and dynamic learning module configured to store the multi-dimensional base costs of switching between activity states, and dynamically learn and update the cost model based on collected player behavior feedback data; An enhanced short-term intention prediction service, connected to the player activity state parsing and serialization engine, configured to use the player activity state sequence, personal profile and real-time context information to predict the player's most likely next activity state and its probability distribution; Server instance micro-ecological profile analyzer, configured to periodically analyze player activity distribution, state transition patterns, etc. of active server instances, and generate and maintain dynamic profiles of instances; A core allocation decision engine, connected to the player activity state parsing and serialization engine, the cost knowledge base and dynamic learning module, the enhanced short-term intention prediction service and the server instance microecological portrait analyzer, is configured to comprehensively calculate the predicted total context switching cost, performance penalty and strategy adjustment items of each candidate instance when players need to be allocated, generate a comprehensive score, and select the optimal target server instance accordingly; The allocation execution and closed-loop feedback interface connects the core allocation decision engine and related server management modules, is configured to execute the final allocation decision, and collect allocation execution results and player subsequent behavior feedback for the cost knowledge base and dynamic learning module, the enhanced short-term intention prediction service and the core allocation decision engine to perform closed-loop optimization of parameters or models.

Images