Method and terminal for automatic identification of a player's key operation in a game

Abstract

This invention provides a method and terminal for automatically identifying key player operations in games, comprising: S1, responding to the start of the game, using a pattern matching algorithm to identify the game type; using decision tree logic to construct a multi-dimensional definition of key operations for the game type; and based on the multi-dimensional definition of key operations, using a game context analysis module, a player behavior analysis module, and a cross-game data processing module to comprehensively identify key operations in the game's replay video. This invention can adapt to different types of games, accurately identifying key operations in game replay videos from all angles and perspectives, providing reliable data support for game development, operation, and player analysis.

Description

Technical Field

[0001] This invention relates to the field of game development and design technology, and in particular to a method and terminal for automatically recognizing key player operations in games. Background Technology

[0002] In the current gaming industry, existing technologies for combat replay and intelligent annotation have many shortcomings in automatically identifying key player actions. Specifically, in real-world gaming scenarios, most existing methods lack a comprehensive and accurate approach to identify and annotate key player actions.

[0003] From the perspective of defining critical actions, there is no unified and clear standard to define what constitutes a critical action in the complex and diverse range of game operations. For example, in different types of games (such as role-playing games and competitive games), there are no clear and explicit rules for judging actions that are extremely important in the game process, such as unleashing a powerful skill or dodging a fatal attack. This makes it difficult to have a definite basis for subsequent identification.

[0004] While identifying the implementation methods of key operations is known to involve analysis from the perspectives of operation command frequency and corresponding game effects, current technology has failed to effectively integrate and optimize these factors to build a comprehensive and efficient analysis mechanism. Therefore, in game battle replay videos, it is difficult to accurately locate and label the truly key operations. Consequently, subsequent development and operation phases of the game, such as game balance adjustments and player personalization, cannot be carried out efficiently and with high quality due to the lack of accurate key operation data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an automatic identification method and terminal for key player operations in games, which can accurately identify key operations in game battle replay videos from all angles and perspectives.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for automatically recognizing key player actions in a game, comprising the following steps: S1. In response to the start of the game, a pattern matching algorithm is used to identify the game type of the game; S2. Use decision tree logic to construct multi-dimensional key operation definitions for the game type; S3. Based on the multi-dimensional definition of key operations, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in the replay video of the game.

[0007] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: An automatic recognition terminal for key player operations in a game includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the automatic recognition method for key player operations in a game as described above.

[0008] The beneficial effects of this invention are as follows: It provides an automatic identification method and terminal for key player operations in games. It identifies game types through pattern matching algorithms, constructs multi-dimensional key operation definitions, and combines multiple innovative modules, such as game context perception, player behavior prediction, and cross-game data fusion, to accurately identify key operations in game replay videos. It can adapt to different types of games and accurately identify key operations from all angles, providing reliable data support for game development, operation, and player analysis. Attached Figure Description

[0009] Figure 1 This is an overall flowchart of a method for automatically recognizing key player operations in a game, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an automatic recognition terminal for key player operations in a game, according to an embodiment of the present invention.

[0010] Label Explanation: 1. An automatic recognition terminal for key player operations in a game; 2. Memory; 3. Processor. Detailed Implementation

[0011] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0012] Please refer to Figure 1 A method for automatically recognizing key player actions in a game, comprising the following steps: S1. In response to the start of the game, a pattern matching algorithm is used to identify the game type of the game; S2. Use decision tree logic to construct multi-dimensional key operation definitions for the game type; S3. Based on the multi-dimensional definition of key operations, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in the replay video of the game.

[0013] As can be seen from the above description, the beneficial effects of the present invention are as follows: It provides an automatic identification method for key player operations in games, identifies game types through pattern matching algorithms, constructs multi-dimensional key operation definitions, and combines multiple innovative modules, such as game context perception, player behavior prediction, and cross-game data fusion, to accurately identify key operations in game replay videos. It can adapt to different types of games, accurately identify key operations from all angles, and provide reliable data support for game development, operation, and player analysis.

[0014] Furthermore, in step S1, a pattern matching algorithm is used to identify the game type, specifically as follows: The core features of the game are extracted using a pattern matching algorithm. These core features include the game interface layout, interactive elements, game mechanics, resource management mode, and task system. The extracted core features are compared with typical pattern feature sets of each game type in the pre-existing game database. The similarity between the extracted core features and the stored typical pattern feature sets is calculated using a similarity calculation function to determine the game type of the current game.

[0015] As described above, by extracting the core features of a game, such as interface layout, interactive elements, and game mechanics, and comparing them with a pre-stored set of typical pattern features to calculate similarity, the game type can be determined quickly and accurately, providing a foundation for subsequent loading of configuration files and definition of multi-dimensional key operations.

[0016] Furthermore, after step S1, the method further includes: The configuration management system retrieves the configuration files related to the game type from the configuration file repository; The configuration management system parses the found configuration file to obtain parameters containing key operation definitions and parameters of each analysis module, and stores the parsed parameter information in the data structure of the game memory.

[0017] As described above, after determining the game type, the system searches for relevant configuration files in the configuration file repository and parses the parameter information, such as the parameters defined for key operations and the parameters of the analysis module. This parameter information is stored in memory for easy access and use later, ensuring that the system can flexibly load configurations according to different game types, thus improving the system's adaptability and versatility.

[0018] Further, step S2 specifically includes: S21. Based on the definition of key operations and their parameter information in the determined game type, the judgment process of key operations is divided into multiple branch nodes and leaf nodes. A decision tree is constructed based on the multiple branch nodes and leaf nodes. Each branch node represents a decision point. Judgment is made based on the dimensions of game data, including skill damage, control effect and resource acquisition amount. The leaf node represents the final operation judgment result. S22. When using cross-game data, relevant features are extracted from the operation data warehouses of different games using data mining algorithms. Operation frequency features, resource usage pattern features, and skill release order features are extracted from key operation data of similar gameplay and character positioning in different games. The operation frequency features, resource usage pattern features, and skill release order features are converted into a unified measurement range using a normalization algorithm. The operation frequency features, resource usage pattern features, and skill release order features after the unified measurement range are integrated into the key operation definition in step S21 using feature combination or weighted average.

[0019] As described above, by using decision tree logic, the critical operation judgment process is broken down into multiple branch nodes and leaf nodes. Judgments are made based on the dimensions of game data, such as skill damage, control effects, and resource acquisition. When cross-game data is used, the system extracts relevant features, performs normalization processing, and integrates them into the current game's critical operation definition. This method can construct a multi-dimensional critical operation definition that fits the actual game, improving the accuracy of critical operation identification.

[0020] Furthermore, the step between S2 and S3 also includes: S23. Using an event-driven architecture, collect operation data from the current game in real time, including skill release time, resource acquisition time, character movement events and their operation types, occurrence times, operation targets and operation effects, and store the operation data in a circular buffer or message queue; S24. Remove duplicate or erroneous data records from the operation data stored in the circular buffer or the message queue, perform a consistency check on the operation data, unify timestamps of different formats into a standard timestamp, and use a conversion algorithm to convert the string type operation type identifier in the operation data into an enumeration type identifier.

[0021] As described above, an event-driven architecture is used to collect operation data from the game in real time, such as skill release time, resource acquisition time, and character movement events. The data is stored in a circular buffer or message queue, and then the data is cleaned and standardized to remove duplicate or erroneous records, unify the timestamp format, and convert string-type operation type identifiers into enumeration types. This method can ensure the integrity and consistency of the data and provide high-quality data support for subsequent identification of key operations.

[0022] Furthermore, step S3 also includes: S30. Use statistical analysis methods to perform preliminary screening on the preprocessed operation data, including calculating the frequency of operations, calculating the average damage of skills, and evaluating the rate of resource acquisition. If the calculation results exceed a preset threshold, they are marked as candidate operations for key operations.

[0023] As described above, statistical analysis methods are used to analyze the preprocessed operation data and calculate indicators such as the frequency of operations, the average damage of skills, and the rate of resource acquisition. If the calculation results exceed a preset threshold, the operation is marked as a key operation. This method can quickly screen out possible key operations, provide candidate operations for subsequent fine identification, and improve identification efficiency.

[0024] Furthermore, in step S3, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in the game replay, specifically as follows: S31. The game scenario analysis module monitors game scene change data, enemy and friendly status data and battle rhythm data in real time. Based on the game type, the judgment threshold range of key operations is set. The threshold range is dynamically adjusted in combination with the real-time monitored game scene change data, enemy and friendly status data and battle rhythm data. The threshold range is optimized according to the behavior of the player group. Operations that fall within the threshold range in the replay video of the game are identified as key operations. S32. The player behavior analysis module uses a machine learning algorithm to build a player behavior model, uses the player behavior model to obtain real-time monitored player operation data to predict the player's next operation intention and operation data, and pre-identifies key operations based on the next operation intention. The machine learning algorithm is at least one of recurrent neural network and its variants, decision tree and random forest algorithms, and deep reinforcement learning algorithms. S33. Collect key operation data of similar gameplay and character positioning in different games through the cross-game data processing module, and normalize the key operation data to obtain a general key operation feature pattern. Identify the operation in the replay video of the game that conforms to the general key operation feature pattern as a key operation.

[0025] As described above, the game scenario analysis module can monitor changes in the game scene, the status of both allies and enemies, and the pace of combat in real time, dynamically adjusting the recognition thresholds and judgment logic for key operations to ensure that the recognition process closely matches the actual game scenario. The player behavior analysis module uses machine learning algorithms to predict the player's next action intention, anticipating possible key operations and improving the timeliness and comprehensiveness of recognition. The cross-game data processing module collects key operation data on similar gameplay and character roles in different games, extracts general key operation feature patterns, and assists in the recognition of key operations in the current game. The synergistic effect of these modules enables the system to accurately identify key operations in complex and ever-changing game scenarios.

[0026] Furthermore, step S3 also includes: S34. When cross-game data exists, transfer learning techniques are used to transfer knowledge learned from other games to the current game, and pre-trained models or feature mapping functions are used to map key operation patterns and features learned from other games of the same type as the current game onto the operation data of the current game.

[0027] As described above, when cross-game data exists, the system uses transfer learning technology to transfer key operation patterns and features learned from other games to the current game, assisting in the judgment of key operations. This can improve the system's recognition accuracy and adaptability when facing new games or new gameplay, and enhance the algorithm's versatility.

[0028] Furthermore, after step S3, the method further includes: S41. Use a data labeling tool to label the identified key operations, the labeling specifically being: Identify key operational information, add data tags and store them in the corresponding fields of the operational data, or store them in the form of a tag file. The data tags include the operation time, the operation role, and the operation effect. S42. The labeled operation data is stored in the database management system, which then classifies and stores the data according to game session, player ID, or date.

[0029] As described above, using data tagging tools to annotate identified key operations, adding metadata tags, and storing the annotated operation data in a database management system, categorizing and storing it according to game session, player ID, or date, effectively ensures efficient management and retrieval of key operation data, providing reliable data support for subsequent game analysis, operation, and optimization.

[0030] Please refer to Figure 2An automatic identification terminal for key player operations in a game includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the automatic identification method for key player operations in a game as described above.

[0031] As can be seen from the above description, the beneficial effects of the present invention are as follows: Based on the same technical concept, and in conjunction with the above-mentioned method for automatically identifying key player operations in a game, an automatic identification terminal for key player operations in a game is provided. This terminal identifies game types through a pattern matching algorithm, constructs multi-dimensional definitions of key operations, and combines multiple innovative modules, such as game context awareness, player behavior prediction, and cross-game data fusion, to accurately identify key operations in game replay videos. It can adapt to different types of games, accurately identify key operations from all angles and perspectives, and provide reliable data support for game development, operation, and player analysis.

[0032] This invention provides an automatic identification method and terminal for key player operations in games. It is mainly applied to scenarios where key player operations are accurately identified and annotated from all angles in game replay videos. The following detailed description is provided with reference to specific embodiments: Please refer to Figure 1 Embodiment 1 of the present invention is as follows: An automatic recognition method for key player actions in games, such as... Figure 1 As shown, the steps include: S1. In response to the game launch, a pattern matching algorithm is used to identify the game type, specifically: The core features of the game were extracted using a pattern matching algorithm. These core features included the game interface layout, interactive elements, game mechanics, resource management mode, and quest system.

[0033] The extracted core features are then compared with typical pattern feature sets for each game type pre-stored in the game database. A similarity calculation function is used to calculate the similarity between the extracted core features and the stored typical pattern feature sets to determine the game type. The similarity calculation function used in the pattern matching algorithm can be cosine similarity, edit distance, or other similarity metrics. Once the calculation result exceeds a predetermined similarity threshold, the game type is determined.

[0034] For example, when analyzing a new game, the pattern matching algorithm will find that it has a rich character attribute panel, a level-up system, and a story progression quest chain. These features are compared with the stored feature set of role-playing games (RPGs) and the similarity is calculated. If the similarity is 85% (assuming the similarity threshold is 75%), the game is therefore classified as an RPG.

[0035] This involves extracting the core features of a game, such as interface layout, interactive elements, and game mechanics, and comparing them with a pre-stored set of typical pattern features to calculate similarity, thereby determining the game type. This allows for quick and accurate identification of game types, providing a foundation for subsequent loading of configuration files and defining multi-dimensional key operations.

[0036] Then, the configuration management system loads the configuration file required for the game type, specifically: The configuration management system retrieves configuration files related to the game type from the configuration file repository.

[0037] The configuration management system parses the found configuration files to obtain parameters containing key operation definitions and parameters of each analysis module, and stores the parsed parameter information in the data structure of the game memory.

[0038] The configuration management system is responsible for storing and managing various configuration information required for different game types. It retrieves the corresponding game type's configuration files from a dedicated configuration file repository. These files are stored in a structured format, such as key-value pairs, XML, or YAML, and contain parameters defining key operations, such as skill effect thresholds and operation frequency thresholds, as well as parameters for various analysis modules, such as machine learning model parameters and sensitivity adjustment parameters for context-aware modules. The configuration management system parses these configuration files and stores the information in an in-memory data structure, such as a hash table or tree structure, for easy access and use by subsequent programs.

[0039] For example, for a game classified as a real-time strategy (RTS) game, the configuration management system retrieves its configuration file from the repository. This file, stored in YAML format, contains parameters such as "resource gathering rate threshold: 500 units per minute" and "building construction time limit: 120 seconds." The configuration management system parses this file and stores these parameters in an in-memory hash table so that subsequent operations can quickly retrieve the corresponding value based on the key (e.g., "resource_gathering_rate_threshold").

[0040] Once the game type is determined, the system searches for relevant configuration files in the configuration file repository and parses the parameter information, such as the parameters defined for key operations and the parameters of the analysis module. This parameter information is stored in memory for easy access and use later, ensuring that the system can flexibly load configurations according to different game types, thus improving the system's adaptability and versatility.

[0041] S2. Use decision tree logic to construct multi-dimensional key operation definitions for game types.

[0042] In other words, for the diverse range of actions in games, a detailed and game-specific set of critical action definition standards will be developed based on different game types (such as the proportion of skill damage and control duration in role-playing games, and the impact of teamwork and the competition for key resources in competitive games). For example, in role-playing games, not only will the proportion of damage from a single skill release to the player's total output be considered, but also the duration of the negative effects brought by the skill and its impact on turning the tide of battle will be taken into account to comprehensively determine whether it is a critical action. When a skill's damage accounts for more than 30% of a player's total output, simultaneously inflicts crowd control on the enemy for more than 5 seconds, and grants the team a significant combat advantage in the following 10 seconds, it is precisely identified as a key play. In competitive games, in addition to focusing on a player's successful acquisition of key resources and actions that lead the team to a short-term advantage, we also analyze the player's role in restricting key enemy characters and the tactical space created for the team during this process. If a player successfully restricts the enemy's core damage dealer for more than 3 seconds while acquiring key resources, thereby creating a favorable situation of local numerical superiority, this is also considered a key play. Through this multi-dimensional and in-depth definition of key plays, we can more accurately align with the actual game situation and capture actions that truly have a significant impact on the game's progress.

[0043] Meanwhile, in identifying key operations, it not only integrates conventional methods such as analyzing the DC frequency of operations and the corresponding game effects, but also innovatively incorporates three core creative modules: game context analysis module, player behavior analysis module, and cross-game data processing module, namely the content of execution step S3.

[0044] S3. Based on the definition of key operations in multiple dimensions, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in the game replay video.

[0045] In this embodiment, a pattern matching algorithm is used to identify the game type, load the configuration file, construct multi-dimensional key operation definitions, and combine multiple innovative modules, such as game context awareness, player behavior prediction, and cross-game data fusion, to accurately identify key operations in game replay videos. This can adapt to different types of games, accurately identify key operations from all angles, and provide reliable data support for game development, operation, and player analysis.

[0046] Embodiment 2 of the present invention is as follows: An automatic identification method for key player operations in a game, based on the above embodiment one, specifically includes step S2 as follows: S21. Based on the definition of key operations and their parameter information in the determined game type, the judgment process of key operations is broken down into multiple branch nodes and leaf nodes. A decision tree is constructed based on the multiple branch nodes and leaf nodes. Each branch node represents a decision point. Judgment is made based on the dimensions of game data, including skill damage, control effect and resource acquisition. The leaf node represents the final operation judgment result.

[0047] For example, in a MOBA game, the root node of the decision tree might be "whether the skill caused damage". If so, it will further check "whether the damage exceeds 30% of the total health". If it does, it will continue to check "whether it caused crowd control", and finally decide whether to mark the skill release operation as a critical operation.

[0048] In this embodiment, the specific steps for constructing multi-dimensional definitions of key operations using decision tree logic are as follows: 1. Determine the root node of the decision tree: Based on the core features of the game, select a representative key indicator as the root node of the decision tree. Taking competitive games as an example, "Does the action affect the acquisition of core resources?" can be used as the root node. This is because in such games, the acquisition of core resources (such as crystals, energy cores, etc.) has a significant impact on the outcome of the game. Using this as the starting point allows for a quick preliminary screening of actions and determines whether an action has the potential to become a key action.

[0049] 2. Design intermediate nodes and branch conditions: Starting from the root node, set intermediate nodes and corresponding branch conditions based on game data and operational characteristics from different dimensions. For example, in a decision tree with "Does the operation affect the acquisition of core resources?" as the root node, if the operation affects the acquisition of core resources, the next intermediate node could be "Does the amount of core resources acquired reach a certain threshold?" The threshold setting here needs to be determined based on the actual game situation, through analysis of a large amount of game data and developer experience. If the acquisition amount reaches the threshold, a further branch condition could be "The degree of interference caused to the enemy when acquiring resources." The degree of interference can be measured by indicators such as the enemy player's operation restriction time and the proportion of health loss. If the operation does not affect the acquisition of core resources, then "Does it have a significant impact on key enemy characters?" can be used as an intermediate node for another branch. Subsequent branch conditions can be further subdivided according to different situations, such as the way the significant impact is caused (control, high damage, etc.) and the duration of the impact.

[0050] 3. Determining Leaf Nodes and Operation Decision Results: After a series of intermediate node judgments, the process reaches a leaf node to determine whether the operation is critical. If the operation passes all critical conditions, it reaches the leaf node representing a critical operation; if a condition is not met at any node, it reaches the leaf node representing a non-critical operation. For example, in the decision tree of the competitive game mentioned above, if an operation affects the acquisition of core resources, the acquisition quantity reaches a threshold, and it significantly interferes with the opponent during acquisition, then the operation will ultimately be determined as a critical operation, corresponding to the critical operation leaf node in the decision tree; conversely, if any of the conditions are not met, it will lead to the non-critical operation leaf node.

[0051] The specific content of the multi-dimensional definition can be as follows: 1. Impact of actions on game resources: This includes core resource acquisition (such as gathering basic resources like gold and wood in real-time strategy games, or collecting energy gems to upgrade towers in tower defense games), rare resource acquisition (such as acquiring specific rare items in role-playing games, which can significantly enhance character abilities), and resource consumption (such as mana and stamina consumed when releasing skills; unreasonable resource consumption may lead to negative results, while reasonable resource consumption for key skill releases may be a crucial action). By considering resource-related indicators, we can determine whether actions promote or hinder the game's progress.

[0052] 2. Impact of actions on game character status: This includes buffs to one's own character (such as increased health, attack power, defense, etc., or gaining special buff effects like invisibility or invincibility) and negative effects on enemy characters (such as dealing damage, control effects including stun, slow, silence, etc., and reducing enemy character attributes). For example, in MOBA games, a control skill that renders the enemy's core damage dealer unable to act for a long time during a crucial team fight, greatly impacting the enemy's combat power, is considered a key action. Similarly, in role-playing games, a priest character timely applying healing skills to teammates, significantly increasing their health and turning the tide of battle, is also a key action.

[0053] 3. Team Collaboration Dimension: This dimension is crucial in multiplayer online games. It includes team-enhancing actions (such as providing vision for teammates and sharing key information; for example, marking enemy positions in shooting games so teammates can prepare for battle), coordinated kill actions (working with teammates to successfully eliminate key enemy characters through skill combos and tactical cooperation; in competitive games, support characters using skills to control enemies and create kill opportunities for the main damage dealers), and tactical execution actions (executing team-planned tactics, such as pushing lanes and tower diving in tower defense games). Successful execution of team collaboration actions often enhances the team's overall combat effectiveness and significantly impacts the outcome of the game.

[0054] 4. Game Situation Changing Dimension: Observe whether actions can change the overall game situation, such as breaking a stalemate (when both sides are deadlocked, one player successfully seizes a key point, breaking the balance and creating an advantage for their team), reversing a disadvantage (when one team is at a disadvantage, a player's action reverses the situation, such as a player making a brilliant play in a match where they are behind, leading their team to win), and expanding an advantage (when one team already has an advantage, actions further solidify that advantage, such as organizing an effective team fight in a MOBA game where they have an economic lead, destroying a key enemy tower, and widening the economic gap). These actions play a crucial role at key moments in the game and are an important dimension to consider in critical actions.

[0055] S22. When using cross-game data, relevant features are extracted from the operation data warehouses of different games through data mining algorithms. Operation frequency features, resource usage pattern features, and skill release order features are extracted from key operation data of similar gameplay and character positioning in different games. Normalization algorithms are used to convert operation frequency features, resource usage pattern features, and skill release order features into a unified measurement range. Feature combination or weighted average is used to integrate the operation frequency features, resource usage pattern features, and skill release order features after the unified measurement range into the key operation definition in step S21.

[0056] For example, for a new team-based competitive game, resource usage characteristics of key operations are extracted from multiple similar games, and Z-score normalization is used to standardize the resource usage data across different games. For the characteristic of "peak resource usage periods," if the current player's resource usage falls within the normalized range and is close to the average of multiple games, it will be considered an important factor in determining key operations.

[0057] This method uses decision tree logic to break down the critical operation judgment process into multiple branch nodes and leaf nodes. It makes judgments based on dimensions of game data, such as skill damage, control effects, and resource acquisition. When cross-game data is used, the system extracts relevant features, performs normalization processing, and integrates them into the current game's critical operation definition. This method can construct multi-dimensional critical operation definitions that fit the actual game, improving the accuracy of critical operation identification.

[0058] Embodiment 3 of the present invention is as follows: An automatic identification method for key player operations in a game, based on the above-described embodiment two, further includes the following step between S2 and S3: S23. An event-driven architecture is adopted to collect operation data in real time from the current game. The game engine will trigger corresponding events when the player performs operations, including skill release time, resource acquisition time, and character movement events. The program registers listeners for these events. When an event occurs, the listener will capture relevant information. The collection of operation data is asynchronous and will not block the main process, which can ensure the smoothness of the game. The collected data includes operation type, occurrence time, operation target, and operation effect. This information is stored in a circular buffer or message queue for subsequent processing.

[0059] For example, when a player uses a skill in the game, the skill release event is triggered, and the corresponding listener will immediately capture the skill's detailed information, including the skill ID, release time, and skill release location, and store this information in the message queue, waiting for subsequent preprocessing steps.

[0060] S24. Data cleaning and transformation algorithms are used to process the operation data stored in the buffer or queue. The data cleaning algorithm checks the integrity and consistency of the data, removing duplicate or erroneous data records from the operation data stored in the circular buffer or message queue, such as duplicate operation records caused by network latency; it also performs consistency checks on the operation data to ensure that the data format and range meet expectations, for example, unifying timestamps of different formats to standard Unix timestamps; while the transformation algorithm converts the data into a format suitable for subsequent analysis, such as converting string-type operation type identifiers to enumeration types, and converting resource quantities to a unified numeric type.

[0061] For example, if there are two skill release operations recorded almost simultaneously due to network jitter, but they are actually the same operation, the data cleaning algorithm will mark one of them as a duplicate and delete it. For time records from different sources, such as "2025-01-09 10:10:10" and "10:10:10.000", they will be uniformly converted to Unix timestamp format, such as "1735383010".

[0062] In this embodiment, an event-driven architecture is used to collect operation data from the game in real time, such as skill release time, resource acquisition time, character movement events, etc. The data is stored in a circular buffer or message queue, and then the data is cleaned and standardized to remove duplicate or erroneous records, unify the timestamp format, and convert the string type operation type identifier into an enumeration type. This method can ensure the integrity and consistency of the data and provide high-quality data support for subsequent identification of key operations.

[0063] In this embodiment, step S3 also includes: S30. Use statistical analysis methods to perform preliminary screening of the preprocessed operation data, including the frequency of calculation operations, the average damage of calculation skills, and the rate of resource acquisition. If the calculation results exceed the preset threshold, they are marked as candidate operations for key operations.

[0064] For example, the frequency of a player's skill release within a certain time window is calculated, and moving averages or sliding window algorithms are used to smooth out data fluctuations. Based on these statistical results and pre-set thresholds (obtained from the configuration management system), potential key actions are marked as candidates.

[0065] For example, within a 10-second time window, the sliding window algorithm is used to calculate the frequency of player skill releases. If the frequency exceeds the threshold of once every 2 seconds set in the configuration, these skill release operations are considered as candidate operations for critical operations.

[0066] By using statistical analysis methods to analyze the preprocessed operation data and calculate indicators such as the frequency of operations, the average damage of skills, and the rate of resource acquisition, if the calculation results exceed a preset threshold, the operation is marked as a critical operation. This method can quickly screen out possible critical operations, provide candidate operations for subsequent fine identification, and improve identification efficiency.

[0067] Step S3 then uses the game context analysis module, player behavior analysis module, and cross-game data processing module to comprehensively identify key operations in the game replay, specifically: S31. The game scenario analysis module monitors real-time game scene changes, enemy and friendly status data, and combat rhythm data to dynamically adjust the judgment threshold range and logic for key operations. This includes setting the judgment threshold range for key operations based on the game type, dynamically adjusting the threshold range based on real-time monitored game scene changes, enemy and friendly status data, and combat rhythm data, and optimizing the threshold range based on player group behavior. This allows the algorithm to identify operations falling within the threshold range in game replay videos as key operations. For example, when the game enters an intense team battle phase, the algorithm automatically increases its focus on team-based operations and appropriately lowers the judgment threshold for individual output operations, as teamwork has a greater impact on the battle outcome at this time. In the early resource contention phase, the algorithm becomes more sensitive to operations such as seizing key resources and controlling key map points. Once a player performs such an operation and it meets the judgment criteria for the corresponding scenario, it will be quickly identified as a key operation and marked. Moreover, the module can also adaptively adjust according to the characteristics of different game maps and the special rules of game modes. For example, in the "Domination Mode" game, it will focus on the operation of successfully capturing key points and continuously defending points, and dynamically optimize the recognition mechanism according to the intensity of the point competition in the game process to ensure that key operations are not missed.

[0068] The dimensions that are increased or decreased are subject to corresponding range limitations, which play a crucial role in ensuring the accuracy and stability of the algorithm.

[0069] 1. Setting Threshold Ranges for Key Operations Based on Game Type: Different types of games have significantly different combat rhythms, the importance of teamwork, and operational characteristics. This dictates that the threshold range for key operations must vary from game to game. In MOBA games, the increase in attention to teamwork operations during team fights is determined by factors such as the synergy of hero skills, map resource distribution, and average player operation frequency. For example, if a MOBA game's hero skill combinations require extremely high teamwork, and map resource distribution has a significant impact on the situation during team fights, then the attention increase range for teamwork-related operations during team fights might be set at 50%-80% of the base attention, while the threshold for key operations involving individual output would be correspondingly reduced by 30%-50%. In role-playing games, where character growth and story progression are emphasized, the attention increase range for teamwork operations during team fights might be 30%-60%, while the threshold for individual output operations would be reduced by 20%-40%. This game type-based range setting allows the algorithm to better fit the characteristics of different games, avoiding misjudgments caused by a "one-size-fits-all" approach.

[0070] 2. Dynamically Adjust Threshold Ranges Based on Real-Time Monitoring of Game Scene Changes, Enemy and Allied Status Data, and Combat Pace Data: Real-time game data is a crucial basis for dynamically adjusting threshold ranges. The system monitors game scene changes, enemy and allied status data, and combat pace data in real time. During team battles, if the actual effect of team collaboration differs significantly from expectations—for example, frequent teamwork actions failing to achieve the anticipated combat advantage—the algorithm will further fine-tune within a preset range. If the focus on team collaboration actions originally increased by 50%, it can be appropriately increased within the 50%-80% range, up to a maximum of 80%; the threshold for individual output actions, initially reduced by 30%, can be further reduced within the 30%-50% range, up to a maximum of 50%. Conversely, if the effect of team collaboration actions exceeds expectations, the increase in focus and the decrease in the threshold for individual output actions will be correspondingly reduced, ensuring the algorithm adapts promptly to changes in the game situation.

[0071] 3. Optimize Threshold Ranges Based on Player Group Behavior: Different player groups have different operating habits and strategic preferences in the game, which will also affect the threshold adjustment range. For high-level players, their operations are more refined and their teamwork awareness is stronger. Therefore, the range for increasing the focus on teamwork operations during team battles may be relatively smaller, for example, set at 40%-60%. The reduction in the threshold for individual output operations will also be more cautious, at 20%-30%. This is because high-level players' individual output operations are often of high quality, and excessively lowering the threshold may lead to misjudging key operations. For novice players, their teamwork ability is relatively weaker. Therefore, the range for increasing the focus on teamwork operations during team battles can be appropriately expanded, at 60%-80%, and the reduction in the threshold for individual output operations can be increased to 40%-50%, in order to better highlight the importance of teamwork in novice players' battles and guide them to improve their teamwork awareness.

[0072] Specifically, in this embodiment, the game context awareness and dynamic adjustment module utilizes adaptive filtering technology to monitor the real-time game context, such as combat phases, map areas, and resource distribution. Adaptive filtering automatically adjusts filtering parameters based on dynamic changes in game data to better track changes in the game state. Simultaneously, a state machine is used to represent different game contexts, switching states based on player actions and game events, and adjusting the judgment thresholds and conditions for key operations in different states.

[0073] For example, in an RTS game, when the game is in the "early development" stage, adaptive filtering will pay more attention to resource gathering operations, adjusting the judgment of critical operations based on changes in the resource gathering rate. When the game enters the "mid-game combat" stage, the state machine switches states, increasing the sensitivity to combat-related operations while reducing the weight of resource gathering operations as a critical operation.

[0074] S32. A player behavior model is constructed using machine learning algorithms through a player behavior analysis module. This model uses real-time monitored player action data to predict the player's next action intent and action data. Based on this intent, key actions are pre-identified. The machine learning algorithm is at least one of recurrent neural networks and their variants, decision trees and random forests, and deep reinforcement learning. For example, if a player frequently maneuvers towards the enemy's backline and their skill preparation animation matches the characteristics of targeting the enemy's core backline damage dealers, the algorithm can predict that the action is likely a key action even before the skill is released and prepare to label it in advance. When the actual action occurs and meets the expected effect, it is immediately and accurately labeled. This intention-based prediction mechanism greatly improves the timeliness and comprehensiveness of key action identification, avoiding missing some momentarily significant actions due to focusing only on actions that have already occurred.

[0075] In the player behavior analysis module, a variety of machine learning algorithms are used to accurately predict the player's operational intentions. These algorithms each have their own advantages, complement each other, and work together to improve the accuracy and reliability of the predictions.

[0076] 1. Recurrent Neural Networks (RNNs) and their variants: RNNs can process sequential data, storing information from previous time steps through memory units to analyze the temporal dependencies of player action sequences. In game scenarios, player actions occur sequentially over time, and RNNs can effectively capture this temporal sequence feature. For example, Long Short-Term Memory (LSTM), a variant of RNN, effectively solves the gradient vanishing problem when RNNs process long sequences by introducing a gating mechanism, more accurately learning long-term and short-term player action patterns. In MOBA games, player actions in a match may last for a long time and have a complex sequence. LSTM can learn the action sequences of players in different stages such as early laning, mid-game team fights, and late-game pushing, predicting whether the player's next action will be to continue attacking, defending, or collecting resources. Gated Recurrent Units (GRUs), another RNN variant, have a relatively simple structure and higher computational efficiency. They can also effectively capture key information in player action sequences through gating mechanisms, ensuring prediction accuracy while quickly processing large amounts of action data, making them suitable for game scenarios with high real-time requirements.

[0077] 2. Decision Tree and Random Forest Algorithms: Decision tree algorithms classify and make decisions based on the characteristics of player action data. Player action data is divided according to different attributes to construct a decision tree model. For example, in role-playing games, attributes such as skill release type, resource acquisition amount, and character movement direction can be used to determine the player's possible next action based on the different values ​​of these attributes. Random forest algorithms are ensemble learning algorithms based on decision trees. By constructing multiple decision trees and combining their prediction results, they improve the model's stability and generalization ability. In games, random forests can perform a more comprehensive analysis of various player action data, reducing the overfitting problem that may occur with a single decision tree. For example, in a multiplayer online competitive game, random forests can comprehensively consider player action data in different map areas and at different game stages to more accurately predict whether a player is preparing to ambush enemies, support teammates, or seize map resources.

[0078] 3. Deep Reinforcement Learning (DRL) Algorithms: Deep reinforcement learning combines the perceptual capabilities of deep learning with the decision-making capabilities of reinforcement learning, allowing an agent (in this scenario, the player behavior model) to continuously learn optimal strategies through interaction with the game environment. In a game, a player's actions affect the game state and reward feedback. Deep reinforcement learning algorithms can utilize this feedback information to learn the optimal action strategy for the player in different game states, thereby predicting the player's action intentions. Taking policy gradient algorithms (such as A2C, A3C, etc.) as an example, it can directly optimize the policy network, enabling the player behavior model to quickly learn action strategies in different game scenarios and predict the player's action intentions in specific situations. Value-based deep reinforcement learning algorithms (such as DQN and its variants) guide decision-making by learning a state value function, which can more effectively handle game scenarios with discrete action spaces and accurately predict the player's intentions to choose between different action options.

[0079] Specifically, in this embodiment, the player behavior analysis module uses a machine learning prediction model to analyze the player's operation history and current operation to predict the player's subsequent operation intentions. This model uses the player's past operation sequences as training data and is trained using deep learning algorithms (such as recurrent neural networks) or other machine learning algorithms (such as decision trees and support vector machines). The trained model then predicts the key operations the player might perform based on the player's current operation sequence and pre-marks potential key operations.

[0080] For example, for a player's action sequence in the game, such as frequent resource gathering and defensive construction, the machine learning prediction model will predict that the player may enter a defensive state based on this sequence. Actions that may involve defense, such as setting traps or strengthening defensive towers, will be marked as potential critical actions in advance, pending further confirmation.

[0081] S33. Considering the correlation and similarities among various games in today's game market, a cross-game data fusion and comparison function has been introduced. This involves collecting key operation data related to similar gameplay and character roles across different games through a cross-game data processing module. This key operation data is then normalized to obtain a general key operation feature pattern. Operations in game replay videos that conform to this general key operation feature pattern are then identified as key operations. For example, in multiple real-time strategy games, for key operations under the similar gameplay of "base expansion," by analyzing player operation data in resource allocation and building layout across different games, general key operation indicators are summarized. These indicators are then applied to new games of the same type to help optimize the key operation judgment for corresponding actions within that game. This allows the algorithm to identify key operations faster and more accurately when facing new games or new gameplay, while also enhancing the algorithm's versatility and adaptability across the entire game industry.

[0082] The normalization process includes data standardization, min-max normalization, and logarithmic transformation.

[0083] Data standardization involves processing the collected key operation data from different games using the Z-score standardization method. For each characteristic temperature, including operation frequency, resource acquisition amount, and skill release interval time, the mean avg and standard deviation std are calculated. For each sample x in the original data, the standardized result x_std is calculated using the formula x_std=(x-avg) / std.

[0084] For example, after collecting resource consumption data for the "constructing specific buildings" operation from multiple real-time strategy games, the mean and standard deviation of this data are calculated, and the resource consumption data for each game is standardized. After this processing, the same characteristic data from different games are transformed into standard normal distribution data with a mean of 0 and a standard deviation of 1, eliminating the differences in units and scales between the data from different games and making the data from different games comparable.

[0085] Min-max normalization involves mapping data to the [0,1] interval for features with a preset boundary range. The formula for min-max normalization is x_norm=(x-min_val) / (max_val-min_val), where min_val and max_val are the minimum and maximum values ​​of the feature with the preset boundary range in all data, respectively.

[0086] For example, when processing character skill cooldown time data from different games, min-max normalization is used to map the differences in skill cooldown times from different games to the [0, 1] interval, highlighting the relative size relationship of the data and facilitating subsequent analysis and comparison.

[0087] Logarithmic transformation includes applying a logarithmic transformation to feature data with exponential growth or uneven distribution, and taking the logarithm of the original feature data x to obtain y=log(x).

[0088] For example, when analyzing the speed at which players gain experience points in different games, since experience point acquisition may grow exponentially during the game process, logarithmic transformation can make the data distribution more uniform, reduce the influence of extreme values, and more accurately reflect the similarity and differences in the experience point acquisition operations between different games.

[0089] The extraction of general key operational feature patterns includes K-Means clustering algorithm, principal component analysis (PCA) algorithm, and association rule mining Apriori algorithm.

[0090] The K-Means clustering algorithm is as follows: The K-Means clustering algorithm is used to perform cluster analysis on the normalized data to determine the number of clusters K. The normalized key operational data is used as input, and the K-Means clustering algorithm divides the data into K clusters, so that the data similarity within the same cluster reaches a first preset threshold, and the data similarity between different clusters reaches a second preset threshold.

[0091] During the clustering process, the algorithm continuously adjusts the cluster centers until convergence conditions are met (such as the cluster centers no longer changing or changing very little). For example, when performing cluster analysis on the operation data related to "releasing skills to cause damage" in multiple role-playing games, several different clusters may be obtained, each representing different types of high-damage skill operation patterns, such as single-target burst type, area-of-effect damage type, etc. These clustering results are the preliminary extracted general feature patterns.

[0092] The Principal Component Analysis (PCA) algorithm is as follows: The normalized critical operation data matrix is ​​input into the principal component analysis (PCA) algorithm to calculate the covariance matrix of the critical operation data, obtain the eigenvalues ​​and eigenvectors, and then select the first few eigenvectors with larger eigenvalues ​​as principal components.

[0093] These principal components are linear combinations of the original data, representing the main direction of data change. For example, when processing high-dimensional data containing multiple operational features (such as operation frequency, skill damage, resource consumption, etc.), PCA can synthesize these features into a few principal components. The feature combinations represented by these principal components are a manifestation of a general feature pattern. Through PCA, not only can data dimensionality be reduced and computational complexity lowered, but common features in key operational data of different games can also be more clearly discovered.

[0094] The Apriori algorithm for association rule mining is as follows: By setting support and confidence thresholds, we can find itemsets in the dataset that appear at a frequency that reaches a third preset threshold, as well as the relationships between these itemsets.

[0095] For example, when analyzing player action data across multiple competitive games, the Apriori algorithm can discover association rules such as "players releasing specific control skills within a certain timeframe after capturing key resources." This is a general key action feature pattern. By mining these association rules, we can gain a deeper understanding of the intrinsic connections between key actions in different games, providing richer reference data for identifying key actions in current games.

[0096] Specifically, the game scenario analysis module monitors real-time changes in the game scene, the status of both allies and enemies, and the pace of combat, dynamically adjusting the recognition thresholds and judgment logic for key operations to ensure the recognition process closely matches the actual game context. The player behavior analysis module uses machine learning algorithms to predict the player's next action intention, anticipating potential key operations and improving the timeliness and comprehensiveness of recognition. The cross-game data processing module collects key operation data for similar gameplay and character roles across different games, extracting general key operation feature patterns to assist in the recognition of key operations in the current game. The synergistic effect of these modules enables the system to accurately identify key operations in complex and ever-changing game scenarios.

[0097] In addition, in this embodiment, step S3 further includes: S34. When cross-game data exists, transfer learning techniques are used to transfer knowledge learned from other games to the current game. Pre-trained models or feature mapping functions are employed to map key operation patterns and features learned from other games of the same type as the current game onto the operation data of the current game. For some difficult-to-determine operations, the results of transfer learning are used to assist in the judgment, improving the accuracy of the judgment.

[0098] For example, for a new tower defense game, the characteristics of "key actions during high-frequency monster waves" learned from several existing mature tower defense games can be mapped to the current game through transfer learning, helping to determine whether the player's actions during the peak of monster waves are key actions.

[0099] When cross-game data exists, the system uses transfer learning technology to transfer key operation patterns and features learned from other games to the current game, assisting in the judgment of key operations. This can improve the system's recognition accuracy and adaptability when facing new games or new gameplay, and enhance the algorithm's versatility.

[0100] Embodiment four of the present invention is as follows: An automatic identification method for key player operations in a game, based on any of the embodiments one to three above, further includes the following after step S3 in this embodiment: S41. Use data labeling tools to label the identified key operations. The labeling is as follows: Identify key operational information, add data tags and store them in the corresponding fields of the operational data, or store them in the form of a tag file. The data tags include the operation time, the operation role, and the operation effect.

[0101] The tool adds metadata tags based on identified key operational information, such as "key skill release," "key resource acquisition," and "key defensive operation." These tags contain detailed operational information, such as operation time, operator role, and operation effect. This tagging information is stored in the corresponding fields of the operation data or as an additional tag file for subsequent retrieval and analysis.

[0102] For example, for an operation that is determined to be a critical skill release, the data labeling tool will add a field named "operation_label" to the data record of that operation, with the value "Critical skill release: Skill name - Fireball, Time - 1735383010, Effect - Deals 50% health damage to the enemy".

[0103] S42. Store the labeled operation data in the database management system, which will then classify and store the data according to game session, player ID, or date.

[0104] In other words, the database management system stores data in different tables or collections according to storage strategies, such as categorizing and storing data by game session, player ID, or date. Simultaneously, it ensures efficient data querying and retrieval based on indexing and partitioning strategies, such as creating indexes for timestamp fields of key operations to enable quick retrieval of key operation data by time range.

[0105] For example, key operation data for a game session with ID "game_session_123" will be stored in a dedicated "key operation data" table, partitioned by player ID, making it convenient for operations staff to analyze the player's key operation history based on their performance.

[0106] In this embodiment, a data tagging tool is used to annotate the identified key operations, add metadata tags, and store the annotated operation data in a database management system. The data is categorized and stored according to game session, player ID, or date, which effectively ensures the efficient management and retrieval of key operation data and provides reliable data support for subsequent game analysis, operation, and optimization.

[0107] In addition, this embodiment provides feedback and optimization schemes. Specifically, the program uses a series of monitoring and evaluation metrics, such as processing latency, resource consumption (CPU, memory usage), and the false positive and false negative rates of critical operations, to continuously monitor system performance and the accuracy of critical operation identification. Real-time data of these metrics is recorded by a logging system and stored in a dedicated monitoring database or log file for subsequent viewing and analysis.

[0108] For example, the program calculates the average latency of processing operation data every 10 seconds and records it in the log, such as "2025-01-09 10:10:10, average latency: 50 milliseconds, CPU utilization: 20%, critical operation misjudgment rate: 3%".

[0109] It also includes the use of feedback control algorithms to adaptively adjust parameters, including adjusting various system parameters based on monitored performance and accuracy metrics. For example, if the misjudgment rate of critical operations is high, the feedback control algorithm will adjust the branching conditions of the decision tree or the hyperparameters of the machine learning model; if the processing latency is too high, it will adjust the data acquisition frequency or buffer size. This algorithm automatically adjusts various system parameters according to preset adjustment strategies and optimization functions to improve system performance and the accuracy of critical operation identification.

[0110] For example, when the misjudgment rate of critical operations exceeds 5%, the feedback control algorithm will lower the threshold of some conditions in the decision tree, or increase the amount of training data for the machine learning model and retrain the model to improve its accuracy.

[0111] Please refer to Figure 2 Embodiment five of the present invention is as follows: An automatic identification terminal 1 for key player operations in a game includes a memory 2, a processor 3, and a computer program stored on the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, it completes the steps in the automatic identification method for key player operations in a game according to any one of the embodiments 1 to 4 above.

[0112] In summary, the automatic identification method and terminal for key player operations in games provided by this invention have the following beneficial effects: 1. Significantly improved accuracy: By defining key operations in a multi-dimensional way that aligns with the actual gameplay and incorporating multiple innovative analysis modules, the system can accurately identify various key operations in complex and ever-changing game scenarios, greatly reducing misjudgments and omissions. Whether in game scenarios with intensive single-player operations or complex battle situations with frequent team collaborations, the system can ensure high accuracy in identifying key operations, providing a reliable data foundation for subsequent analyses based on key operations.

[0113] 2. Strong adaptability and versatility: By considering the characteristics of different game genres, real-time dynamic changes during gameplay, and cross-game data fusion, this algorithm can be widely applied to various game types, including traditional massively multiplayer online role-playing games (MMORPGs), real-time strategy games, and emerging multiplayer competitive mobile games. With simple parameter configuration and fine-tuning based on specific game rules and characteristics, it can deliver excellent key operation recognition and annotation capabilities. This not only meets the needs of different sub-sectors within the game industry but also allows for rapid adaptation to the continuous development of the game industry and the release of new games, maintaining its application value and greatly expanding its scope of application, thus enhancing its practical value within the game industry.

[0114] 3. Supporting the comprehensive development of the gaming industry: For game developers, accurately labeled key operation data allows for in-depth analysis of game balance, pinpointing potential strengths or weaknesses in skills or gameplay, and enabling targeted adjustments and optimizations to improve overall game quality and competitiveness. For game operators, it enables the development of personalized player training programs and precise tournament replay analysis based on player key operation performance, improving player retention and activity, and enhancing player experience and engagement. From the perspective of the entire gaming industry, this efficient and accurate key operation identification and labeling algorithm helps drive the development of game data analysis, promoting the industry towards greater refinement and professionalism, and injecting new momentum into industry innovation and sustainable development.

[0115] 4. Innovative technical features set it apart from many similar technologies, enhancing the patent's uniqueness and novelty: The introduction of the game context analysis module, player behavior analysis module, and cross-game data processing module represents a significant breakthrough in traditional key operation recognition technology. It fills gaps in existing technologies regarding dynamic analysis combined with real-time game contexts, prediction of key operations based on player intent, and cross-game data application. This meets the patent application's innovation requirements and greatly increases the likelihood of success. The synergistic effect of these innovative modules forms a complete and unique key operation recognition system, demonstrating not only technological advancement but also superior performance and value in practical applications. This will attract considerable attention within the industry, laying a solid foundation for the successful patent application and subsequent promotion and application.

[0116] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for automatically recognizing key player operations in a game, characterized in that, Including the following steps: S1. In response to the start of the game, a pattern matching algorithm is used to identify the game type of the game; S2. Use decision tree logic to construct multi-dimensional key operation definitions for the game type; S3. Based on the multi-dimensional definition of key operations, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in the replay video of the game.

2. The method for automatically recognizing key player operations in a game according to claim 1, characterized in that, In step S1, a pattern matching algorithm is used to identify the game type, specifically as follows: The core features of the game are extracted using a pattern matching algorithm. These core features include the game interface layout, interactive elements, game mechanics, resource management mode, and task system. The extracted core features are compared with typical pattern feature sets of each game type in the pre-existing game database. The similarity between the extracted core features and the stored typical pattern feature sets is calculated using a similarity calculation function to determine the game type of the current game.

3. The method for automatically recognizing key player operations in a game according to claim 1, characterized in that, The process following step S1 also includes: The configuration management system retrieves the configuration files related to the game type from the configuration file repository; The configuration management system parses the found configuration file to obtain parameters containing key operation definitions and parameters of each analysis module, and stores the parsed parameter information in the data structure of the game memory.

4. The method for automatically recognizing key player operations in a game according to claim 3, characterized in that, Step S2 specifically involves: S21. Based on the definition of key operations and their parameter information in the determined game type, the judgment process of key operations is divided into multiple branch nodes and leaf nodes. A decision tree is constructed based on the multiple branch nodes and leaf nodes. Each branch node represents a decision point. Judgment is made based on the dimensions of game data, including skill damage, control effect and resource acquisition amount. The leaf node represents the final operation judgment result. S22. When using cross-game data, relevant features are extracted from the operation data warehouses of different games using data mining algorithms. Operation frequency features, resource usage pattern features, and skill release order features are extracted from key operation data of similar gameplay and character positioning in different games. The operation frequency features, resource usage pattern features, and skill release order features are converted into a unified measurement range using a normalization algorithm. The operation frequency features, resource usage pattern features, and skill release order features after the unified measurement range are integrated into the key operation definition in step S21 using feature combination or weighted average.

5. The method for automatically recognizing key player operations in a game according to claim 4, characterized in that, The step between S2 and S3 also includes: S23. Using an event-driven architecture, collect operation data from the current game in real time, including skill release time, resource acquisition time, character movement events and their operation types, occurrence times, operation targets and operation effects, and store the operation data in a circular buffer or message queue; S24. Remove duplicate or erroneous data records from the operation data stored in the circular buffer or the message queue, perform a consistency check on the operation data, unify timestamps of different formats into a standard timestamp, and use a conversion algorithm to convert the string type operation type identifier in the operation data into an enumeration type identifier.

6. The method for automatically recognizing key player operations in a game according to claim 5, characterized in that, Step S3 also includes: S30. Use statistical analysis methods to perform preliminary screening on the preprocessed operation data, including calculating the frequency of operations, calculating the average damage of skills, and evaluating the rate of resource acquisition. If the calculation results exceed a preset threshold, they are marked as candidate operations for key operations.

7. The method for automatically recognizing key player operations in a game according to claim 1, characterized in that, In step S3, the game context analysis module, player behavior analysis module, and cross-game data processing module are used to comprehensively identify key operations in game replays, specifically: S31. The game scenario analysis module monitors game scene change data, enemy and friendly status data and battle rhythm data in real time. Based on the game type, the judgment threshold range of key operations is set. The threshold range is dynamically adjusted in combination with the real-time monitored game scene change data, enemy and friendly status data and battle rhythm data. The threshold range is optimized according to the behavior of the player group. Operations that fall within the threshold range in the replay video of the game are identified as key operations. S32. The player behavior analysis module uses a machine learning algorithm to build a player behavior model, uses the player behavior model to obtain real-time monitored player operation data to predict the player's next operation intention, and pre-identifies key operations based on the next operation intention. The machine learning algorithm is at least one of recurrent neural network and its variants, decision tree and random forest algorithms, and deep reinforcement learning algorithms. S33. Collect key operation data of similar gameplay and character positioning in different games through the cross-game data processing module, and normalize the key operation data to obtain a general key operation feature pattern. Identify the operation in the replay video of the game that conforms to the general key operation feature pattern as a key operation.

8. The method for automatically recognizing key player operations in a game according to claim 7, characterized in that, Step S3 also includes: S34. When cross-game data exists, transfer learning techniques are used to transfer knowledge learned from other games to the current game, and pre-trained models or feature mapping functions are used to map key operation patterns and features learned from other games of the same type as the current game onto the operation data of the current game.

9. The method for automatically recognizing key player operations in a game according to claim 1, characterized in that, The step S3 is followed by: S41. Use a data labeling tool to label the identified key operations, the labeling specifically being: Identify key operational information, add data tags and store them in the corresponding fields of the operational data, or store them in the form of a tag file. The data tags include the operation time, the operation role, and the operation effect. S42. The labeled operation data is stored in the database management system, which then classifies and stores the data according to game session, player ID, or date.

10. An automatic recognition terminal for key player operations in a game, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the automatic identification method for key player operations in a game as described in any one of claims 1 to 9.

Images