An explainable strategy game multi-player style assessment method and device

Abstract

The present invention provides an interpretable strategy game multi-player style evaluation method and device, which relates to the field of data processing technology. The method comprises the following steps: obtaining game data of a player to be predicted, and extracting operator action features of the player to be predicted; obtaining the operation behavior features of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action features of each order; obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to the time series relationship, and obtaining the collaborative behavior features of the player to be predicted; inputting the operation behavior features and the collaborative behavior features into a player combat style model, and obtaining the estimated combat style of the player to be predicted output by the player combat style model. The present invention obtains the estimated combat style of the player to be predicted with a relatively low threshold, and the player can design more targeted game tactics under the map based on this, thereby gaining the upper hand in the war strategy game.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an interpretable strategy game multi-player style evaluation method and device. Background Art

[0002] War strategy games are "people-centered" and are designed to improve players' command, decision-making and action capabilities. If players want to master war strategy games proficiently, they need to have sufficient command knowledge on the one hand, and learn experience from a large number of battles on the other hand. These two requirements form a high threshold for players' learning and training. In order to solve this problem, a player style assessment solution needs to be proposed to help players learn and train. Summary of the invention

[0003] The present invention provides an explainable strategy game multi-player style evaluation method and device, which are used to solve the defect of high learning and training thresholds for players in the prior art, and to obtain the estimated combat style of the player to be predicted with a lower threshold. Players can design more targeted game tactics under the map based on this, and thus gain the upper hand in war strategy games.

[0004] The present invention provides an explainable strategy game multi-player style assessment method, comprising the following steps:

[0005] Obtaining game data of the player to be predicted, and extracting operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is the operator;

[0006] Based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action features of each order, the operation behavior features of the player to be predicted are obtained;

[0007] According to the time sequence relationship, at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted are obtained to obtain the collaborative behavior features of the player to be predicted;

[0008] The operation behavior characteristics and the collaborative behavior characteristics are input into a player combat style model to obtain an estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on the target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0009] According to the interpretable strategy game multi-player style evaluation method provided by the present invention, the operation behavior characteristics of the player to be predicted are obtained based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order, which specifically includes the following steps:

[0010] Encoding the operator action features in a manner of arranging them from front to back in action time sequence to form 0-order time sequence action data;

[0011] Based on the preset support, an association rule mining algorithm is used to mine N-1 order operator action feature frequent items and corresponding support in N-1 order time series action data; wherein N is a positive integer, and the N-1 order time series action data contains N-1 order operator action feature frequent items that are not less than the preset support;

[0012] The support corresponding to the frequent items of the operator action features of each operator at orders 0 to N is accumulated to obtain the accumulated support, and based on the accumulated support, the operation behavior features of the player to be predicted are obtained.

[0013] According to the interpretable strategy game multi-player style evaluation method provided by the present invention, the operator action features are encoded in a manner arranged from front to back in action time sequence to form 0-order time sequence action data, specifically including:

[0014] The encoding is based on the form of who-type-how, wherein who represents the execution operator type of the action feature, type represents the player action feature type, and how represents the attributes or parameters of each action feature.

[0015] According to the interpretable strategy game multi-player style evaluation method provided by the present invention, based on the preset support, the association rule mining algorithm is used to mine the N-1 order time series action data N-order operator action feature frequent items and corresponding support, specifically including:

[0016] In the process of mining association rule mining algorithm, frequent items of operator action features of each order less than or equal to N-2 are pruned.

[0017] According to the interpretable strategy game multi-player style evaluation method provided by the present invention, the collaborative behavior characteristics of the players to be predicted are obtained by obtaining at least one type of other operator action characteristics that appear continuously after a certain type of operator action characteristics of the players to be predicted according to the time sequence relationship, and specifically comprising the following steps:

[0018] Based on the preset time, at least one type of other operator action features that appear continuously after a certain type of operator action features are obtained in the 0th order time series action data according to the time series relationship, so as to obtain the collaborative behavior features.

[0019] According to the explainable strategy game multi-player style evaluation method provided by the present invention, the player combat style model is trained by the following steps:

[0020] Obtaining the sample operation behavior characteristics and the sample collaborative behavior characteristics, and performing dimensionality reduction processing;

[0021] Taking the sample operation behavior characteristics and the sample collaborative behavior characteristics after the dimensionality reduction processing as units, clustering and dividing the sample operation behavior characteristics and the sample collaborative behavior characteristics after the dimensionality reduction processing according to relevant clustering algorithms;

[0022] The sample operation behavior characteristics and the sample collaborative behavior characteristics after clustering are used as training data, and machine learning is used to obtain the player combat style model for generating the estimated combat style of the player to be predicted; wherein the loss function of the player combat style model is determined based on the vector similarity between the sample operation behavior characteristics and the target style, and the vector similarity between the sample collaborative behavior characteristics and the target style.

[0023] The present invention also provides an interpretable strategy game multi-player style evaluation device, comprising:

[0024] A feature extraction module, used to obtain game data of the player to be predicted, and extract operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is an operator;

[0025] The first acquisition module is used to acquire the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order;

[0026] The second acquisition module is used to acquire, according to the time sequence relationship, at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted, so as to obtain the collaborative behavior features of the player to be predicted;

[0027] A style evaluation module is used to input the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model to obtain the estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on the target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0028] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described interpretable strategy game multi-player style evaluation methods are implemented.

[0029] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described interpretable strategy game multi-player style evaluation methods.

[0030] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the above-described interpretable strategy game multi-player style assessment methods are implemented.

[0031] The present invention provides an explainable strategy game multi-player style evaluation method and device, and proposes a frequent item data mining scheme for strategy games. By selecting game data generated after training of opponents, such as game replay information, the opponents are used as players to be predicted, and then the operational behavior characteristics of single operator operations and collaborative action characteristics of multi-operator collaborative operations of the players to be predicted are mined, so that the players can obtain the main tactics and operational styles imagined by the opponents for a certain map. Then, based on the trained player combat style model, the estimated combat style of the players to be predicted is obtained with a lower threshold. Based on this, the players can design more targeted game tactics under the map, thereby gaining the upper hand in war strategy games. During this process, on the one hand, in order to explore the important tactical points of the map and analyze the players' micro-operation habits and common tactical combinations, the operation behavior feature mining of single operator operation was carried out; on the other hand, in order to better support players without professional knowledge to quickly complete core strategy learning, the collaborative action feature mining of multi-operator collaborative operation was carried out, and feature mining was carried out on war strategy games. The mined features can assist players who do not understand tactical knowledge to quickly complete evolutionary ability training; in addition, in order to conduct attribution analysis on the players' abilities, the operation behavior features and collaborative behavior features are used as input data, and intelligent stereotyped evaluation is performed based on the player's combat style model to distinguish the combat styles of multiple players. At the same time, the evaluation results are highly interpretable. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 is a flow chart of an explainable strategy game multi-player style assessment method provided by the present invention;

[0034] Figure 2 It is a specific flow chart of step S200 in the interpretable strategy game multi-player style evaluation method provided by the present invention;

[0035] Figure 3 It is a flowchart of a player combat style model training process in an explainable strategy game multi-player style evaluation method provided by the present invention;

[0036] Figure 4 It is a schematic diagram of the structure of an interpretable strategy game multi-player style evaluation device provided by the present invention;

[0037] Figure 5 It is a specific structural schematic diagram of the first acquisition module in the interpretable strategy game multi-player style evaluation device provided by the present invention;

[0038] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] As a typical war strategy game, especially tactical land war game, the environment of war game has a high complexity. The present invention uses tactical land war game as an example. Similarly, if players want to master tactical land war game, they need to have sufficient command knowledge and learn experience from a large number of battles. According to the rules of tactical land war game, players of the two hostile camps of red and blue (for example, A players who manipulate war games as the red camp, and B players who manipulate war games as the blue camp) will deploy tactics around the important tactical points of the map, and players with command literacy will carefully study the map content, analyze the important tactical points of the map, and then make the most favorable tactical arrangements for their own military situation. However, for players who do not have professional knowledge, it is difficult to analyze the current situation through the game battlefield situation, so how to establish and simulate the opponent's tactics and operating style is of great significance for most players to learn and train. Usually experienced gamers will use their personal subjective judgment to analyze the opponent's style, such as being good at penetration, good at defense, etc. Some people also use the Turing test model to use other players to evaluate the specific abilities of a specific player on a scale, but the evaluation results are slightly less reliable due to the influence of other players' cooperation.

[0041] Assuming that a player with a high tactical level will make full use of terrain and combat tactics to gain advantages when deploying actions, it is understandable that this assumption is obviously in line with basic common sense. From this, it can be concluded that the player's in-game review data information will help model the player's tactical style and operating style.

[0042] Combine the following Figure 1 The present invention describes an explainable strategy game multi-player style assessment method, which includes the following steps:

[0043] S100, obtaining the game data of the player to be predicted, and extracting the operator action features of the player to be predicted. The manipulable objects that the player to be predicted can act on in the game are operators. For war strategy games, operators can be combat units such as various combat arms, combat buildings, etc.

[0044] In this method, the player to be predicted can be a player serving as an opponent, and the subsequent player style evaluation is to evaluate the combat style of the player serving as the opponent in the war strategy game. It should be noted that the player serving as the opponent can be one player or multiple players.

[0045] Since the game review data of the tactical-level army war game simulation stores all operator information in each frame and the actions made by both players in the frame in units of frames, the game data of the player to be predicted obtained in step S100 can be game review data marked with information of both players and including the player to be predicted. The game review data can be game data of the player to be predicted playing against other players, or game data of the player to be predicted playing against game artificial intelligence (AI).

[0046] S200, obtaining the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order.

[0047] S300 , obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to a time sequence relationship, and obtaining collaborative behavior features of the player to be predicted.

[0048] S400, inputting the operation behavior characteristics and the collaborative behavior characteristics into the player combat style model, and obtaining the estimated combat style of the player to be predicted output by the player combat style model. In this method, the player combat style model is trained based on the target style, the sample operation behavior characteristics and the sample collaborative behavior characteristics.

[0049] In the explainable strategy game multi-player style evaluation method of the present invention, a frequent item data mining scheme for strategy games is proposed. By selecting the game data generated after the training of the opponent player, such as the game review information, the opponent player is used as the player to be predicted, and then by mining the operational behavior characteristics of the single operator operation and the collaborative action characteristics of the multi-operator collaborative operation of the player to be predicted, the player can obtain the main tactics and operating style of the opponent player for a certain map, and then based on the trained player combat style model, the estimated combat style of the player to be predicted is obtained with a lower threshold. Based on this, the player can design more targeted game tactics under the map, thereby gaining the upper hand in the war strategy game. During this process, on the one hand, in order to explore the important tactical points of the map and analyze the players' micro-operation habits and common tactical combinations, the operation behavior feature mining of single operator operation was carried out; on the other hand, in order to better support players without professional knowledge to quickly complete core strategy learning, the collaborative action feature mining of multi-operator collaborative operation was carried out, and feature mining was carried out on war strategy games. The mined features can assist players who do not understand tactical knowledge to quickly complete evolutionary ability training; in addition, in order to conduct attribution analysis on the players' abilities, the operation behavior features and collaborative behavior features are used as input data, and intelligent stereotyped evaluation is performed based on the player's combat style model to distinguish the combat styles of multiple players. At the same time, the evaluation results are highly interpretable.

[0050] Combine the following Figure 2 The explainable strategy game multi-player style evaluation method of the present invention is described, and step S100 specifically includes the following steps:

[0051] S110, encoding the operator action features in a manner of arranging them from front to back in the action sequence to form 0-order time sequence action data. Specifically, the extracted operator action features are encoded in the form of WTH (who-type-how), and the encoded value is consistent with the player training environment, where W (who) represents the execution operator type of the action feature, including a total of 8 types as shown in Table 1, T (type) represents the player action feature type, including 13 different action features as shown in Table 2, and H (how) represents the different attributes or parameters of each action feature.

[0052] Table 1 Tactical Army Wargame Operator Type Coding Table

[0053] Operator Type tank Chariot Soldier artillery Unmanned combat vehicle Drones Helicopters Loitering munitions serial number 0 1 2 3 4 5 6 7

[0054] Table 2: Tactical Army Wargame Action Feature Types and Attributes

[0055]

[0056] The WTH code contains important information about each action feature, such as 0-2-1-12-4938, which is translated as tank-shooting-chariot-deploy weapons-4938, which means: the tank deployed its weapons against the enemy tank at coordinate 4938. Based on the WTH code, operator action features were extracted and encoded for all battle replay data. The WTH code contains all valid actions made by players from both camps in the game, and is encoded and stored in order from front to back according to the action sequence.

[0057] Due to the huge decision space of tactical war games, the number of theoretically 0-order actions is huge. However, not all actions are of practical significance. On the one hand, this is due to action legitimacy restrictions, such as tanks cannot cross terrain with large elevation differences; on the other hand, this is due to tactical task restrictions, such as map exploration at a distance from the capture point. The above two situations are extremely unlikely or even impossible to occur in the limited actual operations of players. This method can be based on high-quality games of tactical army war games, so it can be considered that the actions in the battle replay data are all valid and reasonable. By extracting all the first-order actions in the corresponding battle replay data and removing duplicates, a simplified version of the 0-order meta-action is obtained. The single actions in this method are all based on the 0-order meta-action.

[0058] In tactical-level army war games, the interaction between players has significant temporal characteristics. Therefore, when processing the battle replay data, the order of the players' decision-making actions is retained, and the single action is expressed using WTH coding. The same processing method is used for all replay data. Considering that when players are in different camps, the action attributes, especially the coordinates, are quite different, this mining task distinguishes the two hostile camps of red and blue for players. Therefore, the 0-order temporal action data and 0-order temporal action data table corresponding to the two hostile camps of red and blue are extracted from a game.

[0059] S220: Based on the preset support, the association rule mining algorithm is used to mine the N-1 order operator action feature frequent items and the corresponding support in the N-1 order time series action data. N is a positive integer, and the N-1 order time series action data contains the N-1 order operator action feature frequent items with a support not less than the preset support.

[0060] In this embodiment, the data of a game of a player is defined as a transaction, and each possible action is an item. As described above, tactical land war game simulation is a typical dynamic game confrontation environment, and the interaction between players and the environment is a temporal process. Therefore, temporality needs to be considered in the mining process of frequent items. Considering comprehensively, the interpretable strategy game multi-player style evaluation method of the present invention selects the temporal frequent item mining algorithm AprioriAll, and the high-order frequent items mined by it take temporality into account.

[0061] The time series frequent item mining algorithm AprioriAll first forms all feasible N-order operator action feature frequent item candidates by combining N-1-order operator action feature frequent items, and can filter out the less meaningful operator action feature frequent item candidates by setting the corresponding support threshold. Therefore, how to quickly and efficiently form N-order operator action feature frequent item candidates from N-1-order operator action feature frequent items is one of the main ways to improve performance. It can be deduced from the definition of frequent items that any sub-item of an N-order frequent item is a frequent item. Therefore, preliminary frequent item candidates can be obtained by combining N-1-order frequent items in pairs. The combination method is to find two N-1-order operator action feature frequent items with the same first N-2 operator action feature frequent items, and concatenate the two subsequent elements (items) to form two N-order first candidate operator action feature frequent items. As shown in Table 3, the 3-order operator action feature frequent items generate four 4-order first candidate operator action feature frequent items. Then, the N-1-order operator action feature frequent items are searched to see whether there is an N-1-order operator action feature frequent item corresponding to the last N-1 elements (items) of the N-order first candidate operator action feature frequent item. If so, it is an N-1-order second candidate operator action feature frequent item. As shown in Table 3, one 4-order second candidate operator action feature frequent item is finally generated.

[0062] Table 3 Frequent item candidate generation algorithm

[0063]

[0064] For the mined N-order second candidate operator action feature frequent items, by setting the preset support, i.e., the support threshold, traversing the N-1-order time series action data, the N-order operator action feature frequent items are obtained. The frequent item generation algorithm utilizes the definition of time series frequent items and avoids the huge traversal overhead caused by large-scale permutations and combinations.

[0065] During the mining process of the association rule mining algorithm, the frequent items of operator action features of each order less than or equal to N-2 are pruned, that is, the action feature set that does not contain the frequent items of operator action features of N order is deleted. How to recursively generate high-order frequent items? From Table 3, it can be seen that only the frequent items of operator action features of N order are needed to generate frequent items of operator action features of N order. Therefore, the frequent items of operator action features of each order less than or equal to N-2 can be saved, and the memory data can be pruned, which can reduce memory consumption. One implementation method is to replace the original frequent items of operator action features of N order, that is, replace the N-1 order time series action data, after each frequent item of operator action features of N order is generated, and then execute step 220 in a loop.

[0066] After the cyclic processing of step S220, the method obtains the maximum order (Nth order) operator action feature frequent items for the player to be predicted. Different from the time series frequent item mining algorithm AprioriAll, the explainable strategy game multi-player style evaluation method of the present invention saves the action feature support of all orders of the player to be predicted. At the same time, because WTH has a standard form, it can realize cluster analysis of multi-dimensional characteristics of player characteristics. For example, if you want to know the frequency of players operating tanks, you can accumulate the support of tank frequent items from the 1st order operator action feature frequent items. Compared with directly counting the player's action frequency, this method is equivalent to only focusing on the player's habitual actions, thereby eliminating the deviation caused by low-probability actions. In addition, by querying the high-order time series actions of tanks and infantry, special tactical actions such as infantry-tank coordination can also be mined.

[0067] S230, accumulating the support corresponding to the frequent items of the operator action features of each operator in the 0th to Nth order, obtaining the cumulative support, and obtaining the operation behavior features of the player to be predicted based on the cumulative support.

[0068] The results obtained by the temporal frequent item mining algorithm AprioriAll are data-driven and do not contain any rules or game prior knowledge. The interpretable strategy game multi-player style evaluation method of the present invention was experimented on a tactical-level army war game. The data used totaled 1332 games, with a total of 32 players, where the players were divided into two groups, i.e., one group of 16 players played against each other. All the above WTH coded actions were mined using the temporal frequent item mining algorithm AprioriAll. Considering that some actions are rare but extremely important, the preset support of the frequent items set in the experiment is 0.1. At the same time, considering that the main way for players to win comes from seizing control and attacking, the analysis mainly starts from these two aspects.

[0069] Each player's perception of the map scenario is not completely consistent, including the selection of important tactical points on the map, the mode of action and tactical organization, etc. Therefore, the single operator frequent item features mined by each player have certain differences. At the same time, due to the restrictions of the rules of war game deduction, each player must reach a consensus on some cognitions in order to win. Using single operator frequent item mining to mine player behavior characteristics, it mines the replay data of specific players, extracts the players' common behavior patterns, and then derives the players' tactical characteristics.

[0070] To explore the action habits and characteristics of each player, it is necessary to analyze the frequent items of their respective single operators. The interpretable strategy game multi-player style evaluation method of the present invention extracts the attack and control frequent item features of 10 players for analysis. As shown in Table 4, the cumulative support corresponding to the frequent items of the first-order operator action features of the 10 players in the red and blue camps is calculated respectively. The higher the support, the more frequently such actions are used. According to the different initiating operators and action types, they are divided into: tank attack support, chariot attack support, soldier attack support, unmanned combat vehicle attack support, cruise missile support, main control point control support, and secondary control point control support. Table 5 also shows the frequent item order and game winning rate of players in different camps. The content after the WTH code ":" indicates the support of the WTH code.

[0071] Table 4 Frequent item analysis table of red and blue camp operator action features

[0072]

[0073] By analyzing the players' usual actions, we can deduce their tactical styles. Here we take players A and B as examples:

[0074] When player A was fighting against the Red side, the tank and infantry squads had a high degree of attack support, and used cruise missiles to attack many times, and tended to seize the main control points. It can be seen that he fully utilized the mobile assault capabilities of the tank squad and the concealed attack and capture capabilities of the infantry squad, and used the coordination of infantry and tanks to complete the attack and capture operations, and also fully utilized tanks and infantry fighting vehicles. Therefore, it can be inferred that player A analyzed that this scenario is suitable for the Red side to win by quickly attacking and seizing the main control points, so he adopted a sharp knife breakthrough style decision-making method; while on the Blue side, he turned to using infantry squads and infantry fighting vehicles as the main source of firepower, and tended to seize secondary control points. It can be inferred that player A analyzed and determined that the terrain of this scenario was not conducive to the Blue side seizing the main control points, so he adopted a support and guard style to win the victory.

[0075] When player B is on the red side, he tends to use light squads that are easy to hide, and coordinate infantry, unmanned vehicles and cruise missiles to carry out long-range harassment on the enemy in order to achieve the combat objectives, without trying to seize the main control points throughout the process. Therefore, it can be inferred that player B analyzed that it is risky for the red players to seize the hypothetical control point, and that by harassing, avoiding frontal confrontation on the battlefield, and using special operations to inflict heavy damage on the enemy can gain greater benefits. This style belongs to the special coordination style; when on the blue side, player B uses the three main combat arms in a balanced manner and fully utilizes their firepower. At the same time, he evenly distributes the combat intentions to the tasks of seizing the main and secondary control points. Therefore, it can be inferred that player B analyzed that it is easier for the blue players to seize the hypothetical control point, and thus adopted a steady assault style decision.

[0076] The order of the operator action feature frequent item corresponds to the duration of a certain type of action, which can be used to represent the flexibility of the player's tactics. The higher the order, the longer the duration of the player's single action and the lower the tactical flexibility. For example, the frequent item of player C has an order of 8 on the red side and an order of 14 on the blue side. It can be considered that player C's tactics are relatively fixed when facing different opponents. By looking at the high-order frequent items, we can have a more detailed understanding of the player's tactical style. Here are the single operator frequent items of two players (referred to as Team 1 and Team 2) on the blue side for illustration. Team 1 often uses large direct-aiming guns to attack enemy operators near point 4731, and uses portable missiles, small direct-aiming guns and other weapons to attack near 4435. By looking at the assumptions, it is found that point 4731 is the elevation point of the map, indicating that Team 1 is accustomed to using long-range attacks to cover its own operators to occupy the main control points. Team 2 often uses large direct-aiming cannons to attack enemy operators near point 4626, and also uses portable missiles, small direct-aiming cannons and other weapons to attack near point 4435. It can be seen that Team 2's tactic is to attack and interfere with enemy operators occupying secondary control points while occupying the main control points.

[0077] Therefore, step S300 specifically includes the following steps:

[0078] Based on the preset time, at least one type of other operator action features that appear continuously after a certain type of operator action features are obtained in the 0-order time series action data according to the time series relationship, and the collaborative behavior features are obtained.

[0079] In step S300, players of tactical army war games often arrange multiple operators to collaborate to complete specific tasks. The multi-operator collaborative action features often include the player's observation (Observe), adjustment (Orient), decision (Decide) and action (Act) steps, that is, the important military theory OODA cycle process. Specifically, in this embodiment, the player's tactical style will be obtained by mining the multi-operator collaborative action features, that is, the collaborative behavior features, and four types of collaborative action features are proposed: intelligence support, fire support, defense support, backup and replacement. The following will introduce these four types of collaborative actions respectively, and analyze the results of mining the multi-operator collaborative action features.

[0080] First, the definition of the multi-operator collaborative action feature in this embodiment is explained:

[0081] Intelligence support: The operators on one side share the enemy situation. When an operator obtains new changes in the enemy situation, it causes other operators to change their action types. This can be considered as intelligence support coordination.

[0082] Fire support: When a friendly operator obtains the specific location of an enemy operator through observation and finds that the enemy operator is approaching the friendly operator, it opens fire before other friendly operators to stop the enemy operator from attacking the friendly operator. This can be considered as fire support coordination.

[0083] Defense support: When one's own operator defends a certain area, it can strengthen the defense of one's own operator or block the attack. This can be considered as defense support synergy.

[0084] Backup and replacement: When one of our operators is performing a task, it is suddenly damaged or its action is suddenly changed. At this time, other operators on our side quickly take over the action or position of the operator. This can be considered as backup and replacement coordination.

[0085] Similar to the frequent item mining of single operator action features, the mining of multi-operator collaborative action features, i.e., collaborative behavior features, also uses WTH coding to process player actions, and in the mining process of multi-operator collaborative action features, specific extraction rules negotiated with experts can be combined to search on the frame data of the time series. In order to analyze the effect of multi-operator collaborative action feature mining, experiments were also conducted on 1332 games, and the mining results of the replay data of 10 players were analyzed, as shown in Table 6, which records the number of collaborative actions completed by each player. The following is an analysis of players A and B in Table 6 in combination with the analysis conclusions of Table 5.

[0086] When player A is on the red side, the coordinated action feature of fire support is more obvious than other coordinated action features. It can be judged that player A emphasizes fire coordination in the decision-making of this hypothetical red side camp, which corresponds to the conclusion of Table 5. When player A is on the blue side, the coordinated action feature of defense support is more obvious than other features. It can be judged that player A emphasizes the combat intention of capturing point guard in the decision-making of this hypothetical blue side camp, which corresponds to the conclusion of Table 5.

[0087] When player B is on the red side, the characteristic values ​​of the three coordinated actions of intelligence support, fire support and defense support are similar. It can be judged that player B fully utilized the complementary advantages and disadvantages of operators in the decision-making of this hypothetical red camp. From the data in Table 5, it can be seen that when player B made decisions, some of the squad combat technical performances were not fully utilized or fully utilized, resulting in all its coordinated action characteristic values ​​being at a relatively low level. However, the horizontal comparison of its own coordinated action characteristic values ​​is relatively balanced, which is exactly consistent with the conclusion of Table 5. Player B used special and light squads to carry out effective harassment; when player B is on the blue side, the characteristics of the two coordinated actions of fire support and intelligence support are the same and higher than the other two values. It can be judged that player B fully utilized the firepower advantage under the coordination of good intelligence support, and was able to reasonably allocate the firepower of the main combat arms operators to maximize the firepower output and coordination, which is exactly consistent with the conclusion of Table 5.

[0088] Tables 5 and 6 analyze the single operator action features and multi-operator coordinated action features respectively. The two tables complement each other and analyze the decision-making style and strategy selection habits of each player. It should be noted that coordinated action features are not mutually exclusive. Multiple coordinated action features may occur at the same time, and they will be recorded in parallel. The following are several pieces of continuous coordinated action feature information mined:

[0089] 1176: Intelligence support: Own operator 400 <Unmanned Combat Vehicle> discovered enemy operator 10000 <Tank>, and Own operator 0 <Tank> in the field of view of the enemy operator took action at frame 1191 (0-shooting-10000-medium direct-aiming gun-4436); Fire support: Enemy operator 10000 <Tank> is approaching own operator 400 <Unmanned Combat Vehicle>, and own operator 0 <Tank> took action at frame 1191 (0-shooting-10000-medium direct-aiming gun-4436);

[0090] 1201: Backup and replacement: The friendly operator 0 <tank> was performing a task at point 4436. After a sudden transfer, the friendly operator 400 <unmanned combat vehicle> took over its position at frame 1282.

[0091] 1202: Fire support: Enemy counter 10000 <tank> is approaching friendly counter 0 <tank>, friendly counter 400 <unmanned combat vehicle> takes action at frame 1221 (400-shooting-10000-medium missile-4837)

[0092] 1220: Defense support: The friendly operator 200 <Soldier> is in a defensive state, the enemy operator 10000 <Tank> is trying to break through the defense, and the nearby operator 400 <Unmanned Combat Vehicle>, which is also in a defensive state, takes action at frame 1221 (400-Shooting-10000-Medium Missile-4837)

[0093] By summarizing the above information and combining it with the map assumption, it can be concluded that at frame 1176 of the game, the self-unmanned combat vehicle operator (400) discovered that the enemy tank operator (10000) was approaching the main control point defended by the self-tank operator (0), the soldier operator (200), and the unmanned combat vehicle operator (400). At frame 1191, the self-tank operator (0) attacked the enemy tank operator (10000) at coordinate 4436 with a medium-sized direct-aiming gun. At frame 1201, the self-tank operator (0) left the coordinate 4436. At frame 1202, the self-unmanned combat vehicle operator (400) attacked the enemy tank operator (10000) at coordinate 4837 with a medium-sized missile, and arrived at 4436 at frame 1282 to complete the defensive position. The above series of actions include the four types of collaborative action features mined by this patent, involving three types of operators: soldiers, tanks, and unmanned combat vehicles. It can be seen that the players have a high collaborative decision-making ability.

[0094] Table 5 Frequent item analysis table of player operator action features

[0095]

[0096] Table 6 Analysis of player collaborative behavior characteristics (multi-operator collaborative action characteristics)

[0097]

[0098] Players with command knowledge can obtain the above information by watching the battle replay, while players without command knowledge simply watching the battle replay is often time-consuming and does not obtain useful information. The multi-operator collaborative action feature mining provided by the explainable strategy game multi-player style evaluation method of the present invention can quickly scan the information of a game, capture the important moments of player decision-making, and assign labels with game command knowledge for players to complete training and learning.

[0099] Players with a high level of tactical skills will make full use of terrain and combat tactics to gain advantages when deploying actions. The interpretable strategy game multi-player style evaluation method of the present invention can also determine whether a certain coordinate point has tactical significance when deploying actions.

[0100] The player behaviors are summarized from two dimensions: single operator action feature frequent item mining and multi-operator collaborative action feature mining, and the operation behavior features and collaborative behavior features of the players to be predicted are obtained. The above-mentioned mined features can effectively describe the performance of the players in the game confrontation environment. In the interpretable strategy game multi-player style evaluation method of the present invention, corresponding preference labels can be marked for the operation behavior features, and corresponding preference labels can be marked for the collaborative behavior feature preference labels, and the operation behavior feature preference labels and collaborative behavior feature preference labels of the players to be predicted are summarized, so as to facilitate the intelligent qualitative evaluation research in the form of model fitting in step S400.

[0101] Combine the following Figure 3 The explainable strategy game multi-player style evaluation method of the present invention is described, and the player combat style model is trained by the following steps:

[0102] A100. Obtain sample operation behavior characteristics and sample collaborative behavior characteristics, and perform dimensionality reduction processing.

[0103] Different players in strategy games have different combat styles. When collaborating, the appropriate style combination can greatly improve the collaborative effect and allow each player to play to their strengths and avoid their weaknesses. Combat style is different from combat capability and there is no specific indicator to identify it. Therefore, this method is based on player behavior characteristics and uses cluster analysis in unsupervised learning methods to train the model and divide the player combat styles.

[0104] A200, taking the sample operation behavior characteristics and the sample collaborative behavior characteristics after dimensionality reduction processing as units, and according to relevant clustering algorithms, clustering and dividing the sample operation behavior characteristics and the sample collaborative behavior characteristics after dimensionality reduction processing.

[0105] The basis for clustering is the behavioral features of various types extracted by mining the frequent items of single-operator action features and mining the collaborative action features of multiple operators. All features are normalized by the two-norm. In order to visualize the above data, the method uses the PCA algorithm to reduce the high-dimensional data to three dimensions. Considering the diversity of data distribution, the method uses the spectral clustering method with strong adaptability to data distribution, in which the number of cluster centers is set to 3, and finally clusters multiple types of player combat styles for the red and blue sides respectively.

[0106] A300, using the sample operation behavior characteristics and sample collaborative behavior characteristics after clustering as training data, and using machine learning to obtain a player combat style model for generating an estimated combat style of the player to be predicted. The loss function of the player combat style model is determined based on the vector similarity between the sample operation behavior characteristics and the target style, and the vector similarity between the sample collaborative behavior characteristics and the target style.

[0107] For example, for the red side {A,C,F,I} players, from the perspective of action support, tank squads and infantry squads account for a relatively high proportion, and tend to seize the primary and secondary control points. Combined with the coordinated actions, fire support and defense support account for a relatively large proportion. It can be judged that this type of player is good at using infantry and tanks to coordinate attacks and seize control, and can be identified as an offensive and aggressive style.

[0108] For the red side {B,D,E} players, from the perspective of action support, infantry squads, unmanned combat vehicles, and cruise missiles account for a balanced proportion. This type of arms is light and flexible, can effectively harass the enemy, and there are fewer control-grabbing behaviors. The proportion of coordinated actions is basically balanced, with no obvious bias. It can be judged that this type of player flexibly arranges light arms, avoids the risk of control-grabbing, and fights in the form of harassment. This can be identified as a stable and flexible style.

[0109] The player combat style model is trained by unsupervised clustering. Therefore, the player combat style model is used to classify and divide the combat styles of multiple players by unsupervised clustering analysis method to complete the player style evaluation.

[0110] At the same time, there is no linear correlation between player style and win rate. This is because the outcome of a single game is more determined by the player's tactical decision-making ability. The differences in tactical arrangement and execution shown by players of a specific style in the game will cause a huge difference in win rate. For example, when player A is on the red side, the win rate is 0.933, while when player I is on the red side, the win rate is 0.533. In addition, the performance of different styles of tactics in the game is affected by the characteristics of the map assumptions. When on the red side, the offensive and aggressive types are more conducive to winning, while when on the blue side, the support and flexible types are more advantageous.

[0111] The explainable strategy game multi-player style assessment method provided by the present invention can deeply explore the characteristics of the players while avoiding the assessment bias caused by the winning rate. The characteristics of clustered players can be labeled through expert knowledge. For unknown players, the player combat style model can be used to calculate the category to which they belong, thereby quickly completing the style assessment.

[0112] The interpretable strategy game multi-player style assessment device provided by the present invention is described below. The interpretable strategy game multi-player style assessment device described below and the interpretable strategy game multi-player style assessment method described above can refer to each other.

[0113] Combine the following Figure 4 The present invention describes an interpretable strategy game multi-player style assessment device, which includes:

[0114] The feature extraction module 100 is used to obtain the game data of the player to be predicted and extract the operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is the operator.

[0115] The first acquisition module 200 is used to acquire the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order.

[0116] The second acquisition module 300 is used to acquire, in a time sequence relationship, at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted, so as to obtain the collaborative behavior features of the player to be predicted.

[0117] The style evaluation module 400 is used to input the operation behavior characteristics and the collaborative behavior characteristics into the player combat style model to obtain the estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on the target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0118] Combine the following Figure 5 The interpretable strategy game multi-player style evaluation device of the present invention is described, in which the first acquisition module 200 specifically includes:

[0119] The encoding unit 210 is used to encode the operator action features in a manner of arranging them from front to back in action time sequence to form 0-order time sequence action data.

[0120] An acquisition unit is used to mine N-1 order operator action feature frequent items and corresponding support in N-1 order time series action data based on a preset support by using an association rule mining algorithm; wherein N is a positive integer, and the N-1 order time series action data contains N-1 order operator action feature frequent items that are not less than the preset support.

[0121] The accumulation unit 230 is used to accumulate the support corresponding to the frequent items of the operator action features of each operator at each order from 0 to N, obtain the accumulated support, and obtain the operation behavior characteristics of the player to be predicted based on the accumulated support.

[0122] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the interpretable strategy game multi-player style evaluation method, which includes the following steps:

[0123] S100, obtaining game data of a player to be predicted, and extracting operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is an operator;

[0124] S200, obtaining the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order;

[0125] S300, obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to the time sequence relationship, and obtaining the collaborative behavior features of the player to be predicted;

[0126] S400, inputting the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model to obtain an estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on a target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0127] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0128] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the interpretable strategy game multi-player style evaluation method provided by the above methods, the method comprising the following steps:

[0129] S100, obtaining game data of a player to be predicted, and extracting operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is an operator;

[0130] S200, obtaining the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order;

[0131] S300, obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to the time sequence relationship, and obtaining the collaborative behavior features of the player to be predicted;

[0132] S400, inputting the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model to obtain an estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on a target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0133] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the interpretable strategy game multi-player style evaluation method provided by the above methods, the method comprising the following steps:

[0134] S100, obtaining game data of a player to be predicted, and extracting operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is an operator;

[0135] S200, obtaining the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order;

[0136] S300, obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to the time sequence relationship, and obtaining the collaborative behavior features of the player to be predicted;

[0137] S400, inputting the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model to obtain an estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on a target style, sample operation behavior characteristics and sample collaborative behavior characteristics.

[0138] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An explainable strategy game multi-player style assessment method, characterized in that: The following steps are involved: Obtaining game data of the player to be predicted, and extracting operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is the operator; Based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action features of each order, the operation behavior features of the player to be predicted are obtained; According to the time sequence relationship, at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted are obtained to obtain the collaborative behavior features of the player to be predicted; Inputting the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model to obtain an estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is trained based on the target style, sample operation behavior characteristics and sample collaborative behavior characteristics; The step of obtaining the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of the operator action characteristics of each order specifically includes the following steps: Encoding the operator action features in a manner of arranging them from front to back in action time sequence to form 0-order time sequence action data; Based on the preset support, an association rule mining algorithm is used to mine N-1 order operator action feature frequent items and corresponding support in N-1 order time series action data; wherein N is a positive integer, and the N-1 order time series action data contains N-1 order operator action feature frequent items that are not less than the preset support; The support corresponding to the frequent items of the operator action features of each operator at orders 0 to N is accumulated to obtain the accumulated support, and based on the accumulated support, the operation behavior features of the player to be predicted are obtained.

2. The explainable strategy game multi-player style assessment method according to claim 1, characterized in that: The encoding of the operator action features is arranged from front to back in the action time sequence to form 0-order time sequence action data, specifically including: The encoding is based on the form of who-type-how, wherein who represents the execution operator type of the action feature, type represents the player action feature type, and how represents the attributes or parameters of each action feature.

3. The explainable strategy game multi-player style assessment method according to claim 1, characterized in that: The method of mining the N-order operator action feature frequent items and the corresponding support in the N-1-order temporal action data based on the preset support using the association rule mining algorithm specifically includes: In the process of mining association rule mining algorithm, frequent items of operator action features of each order less than or equal to N-2 are pruned.

4. The explainable strategy game multi-player style assessment method according to any one of claims 1 to 3, characterized in that: The method of obtaining at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted according to the time sequence relationship to obtain the collaborative behavior features of the player to be predicted specifically includes the following steps: Based on the preset time, at least one type of other operator action features that appear continuously after a certain type of operator action features are obtained in the 0th order time series action data according to the time series relationship, so as to obtain the collaborative behavior features.

5. The explainable strategy game multi-player style assessment method according to claim 1, characterized in that: The player combat style model is trained by the following steps: Obtaining the sample operation behavior characteristics and the sample collaborative behavior characteristics, and performing dimensionality reduction processing; Taking the sample operation behavior characteristics and the sample collaborative behavior characteristics after the dimensionality reduction processing as units, clustering and dividing the sample operation behavior characteristics and the sample collaborative behavior characteristics after the dimensionality reduction processing according to relevant clustering algorithms; The sample operation behavior characteristics and the sample collaborative behavior characteristics after clustering are used as training data, and machine learning is used to obtain the player combat style model for generating the estimated combat style of the player to be predicted; wherein the loss function of the player combat style model is determined based on the vector similarity between the sample operation behavior characteristics and the target style, and the vector similarity between the sample collaborative behavior characteristics and the target style.

6. An interpretable strategy game multi-player style assessment device, characterized in that: include: A feature extraction module, used to obtain game data of the player to be predicted, and extract operator action features of the player to be predicted; wherein the manipulable object that the player to be predicted can act on in the game is an operator; The first acquisition module is used to acquire the operation behavior characteristics of the player to be predicted based on the cumulative support of each operator of the player to be predicted in the frequent items of operator action characteristics of each order; The second acquisition module is used to acquire, according to the time sequence relationship, at least one type of other operator action features that appear continuously after a certain type of operator action features of the player to be predicted, so as to obtain the collaborative behavior features of the player to be predicted; A style evaluation module, used for inputting the operation behavior characteristics and the collaborative behavior characteristics into a player combat style model, and obtaining the estimated combat style of the player to be predicted output by the player combat style model; wherein the player combat style model is obtained by training based on the target style, sample operation behavior characteristics and sample collaborative behavior characteristics; The first acquisition module is specifically used for: Encoding the operator action features in a manner of arranging them from front to back in action time sequence to form 0-order time sequence action data; Based on the preset support, an association rule mining algorithm is used to mine N-1 order operator action feature frequent items and corresponding support in N-1 order time series action data; wherein N is a positive integer, and the N-1 order time series action data contains N-1 order operator action feature frequent items that are not less than the preset support; The support corresponding to the frequent items of the operator action features of each operator at orders 0 to N is accumulated to obtain the accumulated support, and based on the accumulated support, the operation behavior features of the player to be predicted are obtained.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the interpretable strategy game multi-player style evaluation method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the interpretable strategy game multi-player style evaluation method according to any one of claims 1 to 5 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the interpretable strategy game multi-player style evaluation method according to any one of claims 1 to 5 are implemented.

Images