Method and device for processing information in game and terminal equipment

Abstract

The invention provides an information processing method and device in a game, terminal equipment and a computer readable storage medium. The method comprises the following steps: determining a first virtual character; obtaining role information of the first virtual role and an interaction scene corresponding to the first virtual role, and determining social relation features and interaction scene features of the first virtual role according to the role information and the interaction scene; inputting the social relation features and the interaction scene features into a pre-trained large language model, and determining interaction range parameters for the first virtual character; and adjusting the interaction range of the first virtual character in the game scene according to the interaction range parameter so as to trigger a first interaction behavior corresponding to the first virtual character when the second virtual character enters the adjusted interaction range. The method solves the technical problems that in the prior art, the labor cost is high, the game development efficiency is low, the interaction flexibility of player roles and non-player roles is poor, and then negative game experience feeling is brought to players.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an information processing method, apparatus, terminal device, and computer-readable storage medium in a game. Background Art

[0002] In some games, the interaction between player characters and non-player characters (NPCs) is an important part of the player's gaming experience. High-quality interactive experience can enhance the player's immersion and sense of involvement. Among them, the interaction range of non-player characters is a key factor affecting the player's interactive experience.

[0003] At present, the interaction range of non-player characters depends on the manual configuration of developers. For various interaction situations, different interaction ranges are manually configured so that during the game, the corresponding interaction range can be applied according to the interaction situation that occurs. However, manual configuration of the interaction range not only increases the workload of developers and is time-consuming and labor-intensive, but also because it is impossible to exhaust all possible interaction situations manually, the interaction range applied during the game operation lacks flexibility, cannot meet the psychological expectations of players, and affects the players' gaming experience.

[0004] Therefore, the prior art has technical problems such as high labor costs and low game development efficiency due to the manual configuration of the interaction range of non-player characters, and poor flexibility in the interaction between player characters and non-player characters, which in turn brings a negative gaming experience to players. Summary of the invention

[0005] The present application provides an information processing method, apparatus, terminal device and computer-readable storage medium in a game to solve the technical problems existing in the prior art that the interaction range of non-player characters is manually configured, resulting in high labor costs, low game development efficiency, and poor flexibility in interaction between player characters and non-player characters, thereby bringing a negative gaming experience to players.

[0006] In a first aspect, an embodiment of the present application provides an information processing method in a game, wherein a game scene of the game includes multiple virtual characters, and the method includes: determining a first virtual character from the multiple virtual characters; obtaining character information of the first virtual character and an interaction scene corresponding to the first virtual character, and determining social relationship characteristics and interaction scene characteristics of the first virtual character based on the character information and the interaction scene; inputting the social relationship characteristics and the interaction scene characteristics into a pre-trained large language model, and determining an interaction range parameter for the first virtual character through the large language model; adjusting the interaction range of the first virtual character in the game scene based on the interaction range parameters, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

[0007] In a second aspect, an embodiment of the present application provides an information processing device in a game, wherein a game scene of the game includes multiple virtual characters, and the device includes: a role determination unit, a feature determination unit, an interaction parameter determination unit, and an interaction range adjustment unit; the role determination unit is used to determine a first virtual character from the multiple virtual characters; the feature determination unit is used to obtain character information of the first virtual character and an interaction scene corresponding to the first virtual character, and determine the social relationship characteristics and interaction scene characteristics of the first virtual character according to the character information and the interaction scene; the interaction parameter determination unit is used to input the social relationship characteristics and the interaction scene characteristics into a pre-trained large language model, and determine the interaction range parameters for the first virtual character through the large language model; the interaction range adjustment unit is used to adjust the interaction range of the first virtual character in the game scene according to the interaction range parameters, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

[0008] In a third aspect, an embodiment of the present application provides a terminal device, comprising: a memory, a processor, and a display; the memory is used to store one or more computer instructions; the processor is used to execute the one or more computer instructions to implement the above method; the display is used to display a graphical user interface.

[0009] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having one or more computer instructions stored thereon, which, when executed by a processor, executes the above method.

[0010] Compared with the prior art, the information processing method in the game provided by the present application includes: determining a first virtual character from multiple virtual characters; obtaining the character information of the first virtual character and the interaction scene corresponding to the first virtual character, and determining the social relationship characteristics and interaction scene characteristics of the first virtual character according to the character information and the interaction scene; inputting the social relationship characteristics and the interaction scene characteristics into a pre-trained large language model, and determining the interaction range parameters for the first virtual character through the large language model; adjusting the interaction range of the first virtual character in the game scene according to the interaction range parameters, so that when the second virtual character among the multiple virtual characters enters the adjusted interaction range, the first interaction behavior corresponding to the first virtual character is triggered. First, the interaction range parameters in the method are output by the pre-trained large language model, and do not require manual configuration, which reduces the manpower cost of game development and improves the efficiency of game development. Second, the interaction range parameters in the method are determined by the pre-trained large language model based on the character information of the first virtual character and the interaction scene corresponding to the first virtual character, that is, when the first virtual character and / or the interaction scene are different, the interaction range parameters output by the pre-trained large language model are different, and the interaction range adjusted based on the interaction range parameters are also different, which increases the flexibility of interaction between virtual characters during the game. Third, the interaction range parameter in the method is determined based on the social relationship characteristics and interaction scene characteristics of the first virtual character, that is, the interaction range of the first virtual character in the game scene adjusted based on the interaction range parameter is in line with the real psychological expectations of interpersonal communication, which can bring a real interpersonal atmosphere to the players and enhance the players' immersion and sense of substitution in the game. In summary, the information processing method in the game provided by the present application solves the technical problems existing in the prior art that the interaction range of non-player characters is manually configured, resulting in high labor costs, low game development efficiency, and poor interaction flexibility between player characters and non-player characters, which in turn brings a negative game experience to players. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a schematic diagram of an interaction range of a non-player character provided in an embodiment of the present application;

[0012] Figure 2 is a schematic diagram of another interaction range of a non-player character provided in an embodiment of the present application;

[0013] Figure 3 This is an application system diagram of an information processing method in a game provided by an embodiment of the present application;

[0014] Figure 4 is a flowchart of the information processing method in the game provided by the first embodiment of the present application;

[0015] FIG5( a ) is a first schematic diagram of an information processing method in a game provided by the first embodiment of the present application;

[0016] FIG5( b ) is a second schematic diagram of the information processing method in the game provided by the first embodiment of the present application;

[0017] FIG5( c ) is a third schematic diagram of the information processing method in the game provided by the first embodiment of the present application;

[0018] FIG5( d ) is a fourth schematic diagram of the information processing method in the game provided by the first embodiment of the present application;

[0019] FIG5( e ) is a fifth schematic diagram of the information processing method in the game provided by the first embodiment of the present application;

[0020] Figure 6 is a schematic diagram of the structure of an information processing device in a game provided by the second embodiment of the present application;

[0021] Figure 7 It is a structural diagram of a terminal device provided in the third embodiment of the present application. DETAILED DESCRIPTION

[0022] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0023] With the rapid advancement of computer technology and game development technology, the game industry has entered an unprecedented stage of development. The focus of game developers has gradually shifted from improving game functions to improving player experience in order to extend the life cycle of the game.

[0024] In some games, the interaction between player characters and non-player characters (NPCs) is an important part of the player's gaming experience. High-quality interactive experience can enhance the player's immersion and sense of involvement. Among them, the interaction range of non-player characters is a key factor affecting the player's interactive experience.

[0025] Currently, in games, the interaction range of non-player characters is mainly implemented in two ways: one is the fixed shape trigger area method, which uses a fixed area such as a circle or rectangle as the interaction range of the non-player character. Figure 1As shown, the circular area 102 with the non-player character 101 as the center and R as the radius is the interaction range of the non-player character 101. When the player character 103 enters the circular area 102, the interaction between the non-player character 101 and the player character 103 will be triggered. In this implementation, developers need to consider different interaction situations (such as the social relationship between the non-player character and the player character, character characteristics, etc.), and manually configure different interaction ranges for different interaction situations to increase the player's interactive experience to a certain extent. The second is the ray detection trigger method, which determines whether the player character has entered the interaction range of the non-player character through intersection detection between the ray and the collision box or skeleton of the non-player character. As shown in FIG. Figure 2 As shown, a ray 202 (the ray has a preset length) is emitted from the player character 201, and as the player character 201 and the non-player character 203 approach, when the ray 202 intersects with the collision box 204 of the non-player character 203, it is judged that the player character 201 has entered the interaction range of the non-player character 203, and the interaction between the player character 201 and the non-player character 203 will be triggered. In this implementation, although the interaction range of the non-player character 203 is no longer a fixed-shape area, the developer still needs to consider different interaction situations and manually configure different ray parameters for different interaction situations to increase the player's interactive experience to a certain extent. In summary, manually configuring the interaction range not only increases the workload of the developer, is time-consuming and labor-intensive, and affects the efficiency of game development. It is also impossible to exhaust all possible interaction situations manually, so that the applicable interaction range during the game operation lacks flexibility, cannot meet the psychological expectations of the player, and affects the player's interactive experience.

[0026] In view of this, the present application provides an information processing method in a game. First, the interaction range parameter in the method is output by a pre-trained large language model, and does not require manual configuration, which reduces the manpower cost of game development and improves the efficiency of game development. Second, the interaction range parameter in the method is determined by a pre-trained large language model based on the role information of the first virtual character and the interaction scene corresponding to the first virtual character, that is, when the first virtual character and / or the interaction scene are different, the interaction range parameter output by the pre-trained large language model is different, and the interaction range adjusted based on the interaction range parameter is also different, which increases the flexibility of interaction between virtual characters during the game. Third, the interaction range parameter in the method is determined based on the social relationship characteristics and interaction scene characteristics of the first virtual character, that is, the interaction range of the first virtual character in the game scene adjusted based on the interaction range parameter is in line with the real psychological expectations of interpersonal communication, which can bring a real interpersonal atmosphere to the player and enhance the player's immersion and sense of substitution in the game. .

[0027] The information processing method, apparatus, terminal device and computer-readable storage medium in the game described in this application are further described in detail below in conjunction with specific embodiments and drawings.

[0028] Figure 3 1 is an application system diagram of an information processing method in a game provided by an embodiment of the present application. Figure 3 As shown, the system includes a user terminal 301 and a server terminal 302. The user terminal 301 can be any device such as a smart phone, a tablet computer, a laptop computer, a desktop computer, a personal digital assistant (PDA), etc. The server terminal 302 can be a processing module located inside the user terminal 301, or a processing device electrically connected to the user terminal 301, or a server connected to multiple user terminals 301 in communication. The server terminal 302 is deployed with the information processing method in the game provided by the present application to determine the interaction range parameters of the virtual character according to the character information and interaction scene of the virtual character during the game process.

[0029] The first embodiment of the present application provides a method for processing information in a game.

[0030] The game scene of the game includes multiple virtual characters, which can be player characters or non-player characters, without specific restrictions. The player character refers to a virtual character directly controlled by a real player, and the behavior of the player character depends on the player's operation and technical level. The non-player character refers to a virtual character controlled by a pre-written script or artificial intelligence algorithm, and the behavior of the non-player character is usually preset or based on conditional response.

[0031] In addition to multiple virtual characters, the game scene may also have other virtual objects, such as virtual pets, virtual props to be picked up, virtual weapons, etc. The specifics vary depending on different game settings and are not limited here.

[0032] Figure 4 Flowchart of the information processing method in the game provided by this embodiment. Figure 4 The information processing method in the game provided by this embodiment is described in detail. The embodiments described below are used to explain the technical solution of this application and are not intended to be used as limitations for actual use.

[0033] like Figure 4 As shown, the information processing method in the game provided by this embodiment includes the following steps S410 to S440.

[0034] Step S410: determining a first virtual character from a plurality of virtual characters.

[0035] The first virtual character may be a non-player character among the multiple virtual characters, or a player character among the multiple virtual characters with which an interaction is about to occur. The first virtual character may be one of the multiple virtual characters, or may be several of the multiple virtual characters.

[0036] In an optional implementation, the plurality of virtual characters include at least one non-player character, and the at least one non-player character is determined as the first virtual character.

[0037] In another optional implementation, at least two player characters among the plurality of virtual characters are about to interact, and any one of the at least two player characters is determined as the first virtual character.

[0038] Step S420, obtaining the role information of the first virtual role and the interaction scene corresponding to the first virtual role, and determining the social relationship characteristics and interaction scene characteristics of the first virtual role according to the role information and the interaction scene.

[0039] The obtaining of the character information of the first virtual character and the interaction scene corresponding to the first virtual character can be understood as obtaining from game data, for example, obtaining the character code of the first virtual character from the game plot text, obtaining the character attributes of the first virtual character, etc., for example, obtaining the current emotional state of the first virtual character from the game log, obtaining the current scene atmosphere of the interaction scene, etc.

[0040] The character information can be understood as information that can describe at least some of the characteristics of the virtual character, such as the role type of the virtual character (gender, status, etc.), the personality traits of the virtual character (such as introversion, extroversion, etc.), the interpersonal relationships of the virtual character (such as being friends with a certain virtual character, being enemies with a certain virtual character, etc.), the emotional state of the virtual character (such as joy, sadness), etc.

[0041] The interaction scene can be understood as the situational state when the virtual characters interact in the game scene, for example, the task type (main task, side task, etc.), scene atmosphere (serious, relaxed, etc.) when the first virtual character interacts in the game scene.

[0042] In an optional implementation, when the first virtual character is a non-player character, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained, including: in response to monitoring a first game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the first game event at least includes a game process change event of the game.

[0043] The change in game progress may be caused by a change in the game stage, such as the player unlocks the second game stage and the game progress enters the second game stage from the first game stage. It may also be caused by a change in the game version, such as the game version is updated from version 1.0 to version 2.0, and the game progress changes in version 2.0 of the game.

[0044] Changes in the game process often cause the evolution of non-player characters controlled by the game system, such as non-player characters evolving from level 50 to level 80, which leads to changes in the character information of non-player characters, such as changes in the status relationship between non-player characters and other virtual characters, and changes in the personality traits of non-player characters from modesty to arrogance. Changes in the game process will also cause changes in the settings of the game plot, etc., which will lead to changes in the interactive scene, such as the atmosphere of the interactive scene changing from relaxed to tense, and the task context of the interactive scene changing from a side quest to a main quest. Therefore, when the game process is monitored to have changed, it is necessary to obtain the character information of the changed non-player characters and the changed interactive scenes, so that the interactive range corresponding to the non-player characters can change adaptively with the changes of the non-player characters and the interactive scenes.

[0045] In another optional implementation, when the first virtual character is a player character, obtaining character information of the first virtual character and an interaction scene corresponding to the first virtual character includes: in response to monitoring a second game event, obtaining character information of the first virtual character and an interaction scene corresponding to the first virtual character; wherein the second game event includes at least a character state change event of the player character.

[0046] The character status change may be caused by the player unlocking new achievements, new levels, new maps, new world levels, new reputations, new game factions, etc., which enables the player character to obtain new permissions.

[0047] The change of character status will lead to the change of the character information of the player character, such as the change of the status relationship between the player character and other virtual characters, the change of the interpersonal communication preference of the player character, etc. The change of character status often leads to the change of interaction scene, such as the player character corresponds to a new interaction scene in the new map, the task context of the interaction scene changes from a side task to a main task, etc. Therefore, when the player character's character status is monitored to change, it is necessary to obtain the character information and interaction scene of the changed player character, so that the interaction range corresponding to the player character can change adaptively with the change of the player character and the interaction scene.

[0048] In this embodiment, after the character information of the first virtual character and the interaction scene corresponding to the first virtual character are obtained, the social relationship characteristics and the interaction scene characteristics of the first virtual character are determined according to the character information and the interaction scene.

[0049] The social relationship feature can be understood as feature information that can describe the social relationship of the first virtual character in the game. Generally, the social relationship of the first virtual character can be described from multiple dimensions, that is, the social relationship feature is a collection of sub-feature information under multiple social feature dimensions.

[0050] The interaction scene feature can be understood as feature information that can describe the interaction scene corresponding to the first virtual character. Generally, the interaction scene corresponding to the first virtual character can be described from multiple dimensions, that is, the interaction scene feature is a collection of sub-feature information under multiple scene feature dimensions.

[0051] Optionally, the social relationship features and the interactive scene features are represented in the form of vectors. Specifically, the social relationship feature vector and the interactive scene feature vector corresponding to the first virtual character are generated according to the role information and the interactive scene. Among them, the social relationship feature vector can be understood as a vector composed of feature values ​​corresponding to multiple social feature dimensions that can represent the social relationship of the first virtual character. Exemplarily, the social relationship of the first virtual character is described by feature 1, feature 2, and feature 3, then the social relationship feature vector corresponding to the first virtual character can be expressed as [feature value of feature 1, feature value of feature 2, feature value of feature 3]. Specifically, the feature value corresponding to the social feature dimension can be a specific numerical value, such as using 1 to represent that the character of the virtual character is extroverted, or it can be a discrete classification label, such as using the "extrovert" label to represent that the character of the virtual character is extroverted. The scene feature vector can be understood as a vector composed of feature values ​​corresponding to multiple scene feature dimensions that can represent the interactive scene. Exemplarily, the interactive scene is described by feature a, feature b, and feature c, then the interactive scene feature vector corresponding to the first virtual character can be expressed as [feature value of feature a, feature value of feature b, feature value of feature c]. Specifically, the feature values ​​corresponding to the scene feature dimensions are mainly represented by discrete classification labels. For example, the "main line" label is used to indicate that the task type corresponding to the interaction scene is the main line task.

[0052] In an optional implementation, this step may specifically include the following steps S421 to S425:

[0053] Step S421, based on the role information and the interaction scene, extract the feature values ​​corresponding to multiple preset social feature dimensions and the feature values ​​corresponding to multiple preset scene feature dimensions, wherein the social feature dimensions and the scene feature dimensions are associated with the training samples when training the large language model.

[0054] Step S422: determining a feature label corresponding to each social feature dimension according to a mapping relationship between a feature value and a feature label preset for each social feature dimension.

[0055] Step S423: combining the feature labels corresponding to each social feature dimension into social relationship features according to a preset structure.

[0056] Step S424, determining the feature label corresponding to each scene feature dimension according to the mapping relationship between the feature value and the feature label preset for each scene feature dimension.

[0057] Step S425, combining the feature labels corresponding to each scene feature dimension into interactive scene features according to a preset structure.

[0058] In order to make the final predicted interaction range parameters more consistent with the role characteristics of the first virtual character and the scene characteristics of the interaction scene, in this embodiment, the social relationship of the first virtual character is described with social characteristics of multiple dimensions, and the interaction scene corresponding to the first virtual character is described with scene characteristics of multiple dimensions.

[0059] In a specific implementation, the dimensions of the social features describing the social relationship of the first virtual character are preset. Exemplarily, in some games, the social features of multiple dimensions include the first virtual character's favorability towards other virtual characters, the difference in status between the first virtual character and other virtual characters, etc. In other games, the social features of multiple dimensions include the first virtual character's cultural background, personality traits, etc. Optionally, since the social features need to be input as input data into a pre-trained large language model in the later stage, the dimensions of the social features are actually associated with the training data when training the large language model. Specifically, it is consistent with the dimensions of the social features included in the training data. Exemplarily, the social features in the training data include three dimensions of favorability, status difference, and cultural background. Then, in this step, the corresponding feature values ​​will be extracted for the three dimensions of favorability, status difference, and cultural background.

[0060] Extracting the corresponding feature value for each social feature dimension is performed based on the role information of the first virtual character. Exemplarily, the favorability value corresponding to the favorability of the first virtual character to the second virtual character extracted from the role information is 70.

[0061] In the method provided in this embodiment, a mapping table between feature values ​​and feature labels is pre-set for each social feature dimension to establish a mapping relationship between each feature value and feature label under the social feature dimension. For example, the mapping table set for the social feature dimension of "favorability" is shown in Table 1:

[0062] Table 1 “Favorability” social feature dimension mapping table

[0063] Eigenvalue Feature Tags 80-99 lover 60-79 Best Friend 20-59 rusty 0-19 stranger

[0064] Based on the mapping table shown in Table 1, since the extracted favorability value corresponding to the first virtual character's favorability towards the second virtual character is 70, the feature label corresponding to the first virtual character's favorability towards the second virtual character is "close friend".

[0065] By combining the feature labels corresponding to each social feature dimension according to a preset structure, a social relationship feature can be formed, specifically, a social relationship feature vector. Exemplarily, multiple social feature dimensions include favorability, status difference, and cultural background, and the preset structure is [favorability, cultural background, status difference]. Then, the social relationship feature vector can be expressed as [close friend, close contact culture, companion]. In a specific implementation, the preset structure is associated with the training data when training the large language model, specifically, it is consistent with the combined structure of multiple social feature dimensions included in the training data.

[0066] In a specific implementation, the dimensions of the scene features describing the interactive scene are preset. Exemplarily, in some games, the scene features of multiple dimensions include task type, task objectives, etc., and in other games, the scene features of multiple dimensions include character emotional state, scene atmosphere state, etc. Optionally, since the scene features need to be input as input data into a pre-trained large language model in the later stage, the dimensions of the scene features are actually associated with the training data when training the large language model, and specifically, are consistent with the dimensions of the scene features included in the training data. Exemplarily, the scene features in the training data include three dimensions: task type, emotional state of the virtual character, and scene atmosphere of the interactive scene. Then, in this step, corresponding feature values ​​will be extracted for the three dimensions of task type, emotional state of the first virtual character, and scene atmosphere of the interactive scene.

[0067] Extracting the corresponding feature value for each scene feature dimension is performed according to the interaction scene corresponding to the first virtual character. Exemplarily, the atmosphere value corresponding to the scene atmosphere extracted from the interaction scene is 90.

[0068] In the method provided in this embodiment, a mapping table between feature values ​​and feature labels is pre-set for each scene feature dimension to establish a mapping relationship between each feature value and feature label under the scene feature dimension. For example, the mapping table set for the "scene atmosphere" scene feature dimension is shown in Table 2:

[0069] Table 2 "Scene atmosphere" scene feature dimension mapping table

[0070] Eigenvalue Feature Tags 80-99 nervous 60-79 serious 30-59 romantic 0-29 easy

[0071] Based on the mapping table shown in Table 2, since the atmosphere value corresponding to the extracted scene atmosphere is 90, the feature label corresponding to the scene atmosphere of the interactive scene is “tense”.

[0072] By combining the feature labels corresponding to each scene feature dimension according to a preset structure, an interactive scene feature can be formed, specifically, an interactive scene feature vector. Exemplarily, the multiple scene feature dimensions include task type, emotional state of the virtual character, and scene atmosphere of the interactive scene. The preset structure is [scene atmosphere of the interactive scene, emotional state of the virtual character, task type]. Then, the scene feature vector can be expressed as [tension, sadness, small talk and chatting]. In a specific implementation, the preset structure is associated with the training data when training the large language model, specifically, it is consistent with the combined structure of multiple scene feature dimensions included in the training data.

[0073] Step S430: input the social relationship features and the interaction scene features into a pre-trained large language model, and determine the interaction range parameters for the first virtual character through the large language model.

[0074] The large language model refers to an artificial intelligence system with a large number of parameters obtained through machine learning training, which is specifically used to process natural language tasks. The model contains hundreds of millions to trillions of parameters, which enables it to perform well in processing complex language structures, can generate or understand language according to the context, and provide a more human-friendly understanding and generation capability. It can perform new tasks without additional training (zero-sample learning), or only a small number of examples are needed to quickly learn new tasks (few-sample learning). In this embodiment, the pre-trained large language model can be a mature large language model, or it can be obtained by training a general large language model. In a specific implementation, in order to make the output effect of the model better, before application, the general large language model will be trained with pre-built training samples to obtain a pre-trained large language model that can be applied to the game process.

[0075] Based on this, the method provided in this embodiment also includes: training a large language model. In an optional implementation, the training of the large language model can be divided into two steps. Step one is pre-training, the main purpose of which is to enable the large language model to learn game knowledge and improve the large language model's ability to understand and construct game knowledge; step two is targeted training, the main purpose of which is to allow the large language model that already has game knowledge to master the distribution law of the interaction range under different social relationships and interaction scenarios. The following is a detailed description of the training of the large language model:

[0076] In an optional implementation, the pre-training step may include the following steps S11 to S13:

[0077] Step S11, obtaining a second training sample set, the second training sample set includes a pre-constructed game knowledge graph and a variety of game text data; wherein, the game knowledge graph includes multiple abstract concepts, multiple specific objects, and associations between multiple abstract concepts, between multiple specific objects, and between multiple abstract concepts and multiple specific objects in the game, and the multiple game text data includes at least game plot text and game log text.

[0078] Step S12: Use the second training sample set to train the large language model so that the large language model learns game knowledge.

[0079] Step S13, generating a pre-trained large language model.

[0080] The game knowledge graph is a game knowledge network formed by developers extracting structured concepts, entities, relationships and other knowledge elements based on the background settings, character relationships, cultural customs, character personality and other settings of the game. The basic elements of the game knowledge graph include: concepts, entities, and relationships. Among them, concepts refer to abstract concepts that appear in the game, such as race, profession, power, skills, etc., which are important components of the game settings and reflect the basic rules and characteristics of the game world; entities refer to specific objects that appear in the game, such as characters, places, items, etc., which are the core elements of game narrative and player interaction; relationships refer to semantic connections between concepts, entities, and concepts and entities, such as affiliation, location, ownership, use, influence, etc. By depicting the relationship between concepts and entities, the internal logic of the game settings can be further revealed. The game knowledge graph is the main source of knowledge for training large language models.

[0081] For example, in a specific game, concepts include "wizard", "demon hunter", "mage", "empire", "alliance", etc., which depict the game's occupations and power settings; entities include important characters such as "Xiao Ming" and "Xiao Hong", places such as "such and such building" and "such and such grassland", iconic items such as "such and such equipment" and "such and such weapon"; relationships include "Xiao Ming belongs to the wizard" (affiliation relationship from concept to entity), "Xiao Hong is located in such and such grassland" (location relationship from entity to entity), "Xiao Ming is equipped with such and such weapon" (ownership relationship from entity to entity), etc. In addition, the game knowledge graph can also depict the character's personality traits, such as "Xiao Ming has a cold and rational personality" and "Xiao Hong has a strong personality", etc., to enrich the character image.

[0082] Game text data covers all text-related parts of the game, such as game plot text (such as dialogues, narrations, text in cutscenes, etc.), game rule text (such as text explanations of game mechanics, operating instructions, mission objectives, and how to win the game or achieve specific achievements), in-game character setting text (such as the character's background story, personality description, skill description, etc.), game log text (such as the character's action records in the game, player's behavior records, system logs, etc.), etc.

[0083] Optionally, the game knowledge graph and game text data of the pre-trained large language model are not for a specific game. The large language model is pre-trained by self-supervised learning and other methods using massive game text data and game knowledge graphs, so that the large language model can initially learn common sense and narrative style in the game field, and become a pre-trained large language model in the game field. Of course, it is also possible to further fine-tune the pre-trained large language model based on the pre-trained large language model using game text data for a specific game, so as to further enhance the large language model's ability to understand specific games.

[0084] It should be noted that through the pre-training in this step, the large language model has a certain basic understanding ability in the field of games, but has not yet reached the generation task of the interaction range for virtual characters.

[0085] In another optional implementation, the targeted training may include the following steps S21 to S23:

[0086] Step S21, obtaining a first training sample set, the first training sample set includes at least one training sample, the training sample includes social relationship features, interaction scene features, and interaction range parameters annotated for a third virtual character in the game, the third virtual character is any one of a plurality of virtual characters preset in the game.

[0087] Step S22, using the first training sample set to train the large language model, so that the large language model learns to infer the interaction range parameters according to the social relationship characteristics and the interaction scene characteristics; specifically, using the first training sample set to train the pre-trained large language model.

[0088] Step S23, generating a pre-trained large language model.

[0089] In a specific implementation, a multi-dimensional annotation system of social features and scene features is pre-built for training samples. The purpose of this annotation system is to provide structured and standardized annotation data for training large language models. By fine-grained annotation of key factors affecting the scope of virtual character interaction, the large language model can efficiently and accurately learn the correspondence between different social relationships and interaction scene combinations and the scope of virtual character interaction. Optionally, feature annotation is completed by developers or based on preset rules, without specific restrictions.

[0090] In this embodiment, the training sample is composed of social relationship features, interaction scene features, and interaction range parameters. Optionally, it is composed of social relationship feature vectors, interaction scene feature vectors, and interaction range parameters. The three training data in the training sample are all labeled data.

[0091] In an optional implementation, the social relationship feature includes a first feature label annotated for at least one of the following social feature dimensions: favorability, status difference, cultural background, and personality traits; wherein favorability is used to characterize the degree of emotional intimacy between the third virtual character and the fourth virtual character, status difference is used to characterize the relative status of the third virtual character and the fourth virtual character in the game, cultural background is used to characterize the interpersonal communication preference of the third virtual character in the interactive scenario, and personality traits are used to characterize the tendencies of the third virtual character in multiple personality dimensions.

[0092] The fourth virtual character refers to any one of the multiple virtual characters preset in the game except the third virtual character. Optionally, the third virtual character can be randomly selected or representative, such as a virtual character with a significant difference in status from the first virtual character, without specific limitation. It should be noted that targeted training does not require too many training samples, and one or a few training samples can be used to fine-tune the pre-trained large language model, so that it can infer a reasonable virtual character interaction range based on social relationship characteristics and interaction scene characteristics. Specifically, multiple groups of <social relationship feature vectors, interaction scene feature vectors, interaction range parameters> training samples are used to fine-tune the pre-trained large language model.

[0093] In a specific implementation, it is necessary to annotate social feature dimensions such as favorability, status difference, and cultural background to obtain feature labels corresponding to the social feature dimensions such as favorability, status difference, and cultural background. In this embodiment, the feature labels annotated for the social feature dimensions are defined as the first feature labels.

[0094] In a specific implementation, the first feature label is a specific numerical value or classification label for the social feature dimension. For example, a specific numerical value of 0-100 is used to label the favorability, or a discrete classification label such as "lover", "close friend", "acquaintance" is used to label the favorability. The specific labeling method to be used is determined according to the specific game and the developer's settings, and is not limited here.

[0095] In an optional implementation, the interaction scene feature includes a second feature label annotated for at least one of the following scene feature dimensions: task context, emotional state, and scene atmosphere; wherein the task context is used to characterize the task goal of the second interaction behavior corresponding to the third virtual character; the emotional state is used to characterize the emotions of the third virtual character in the interaction scene, and the scene atmosphere is used to characterize the overall atmosphere of the interaction scene.

[0096] In a specific implementation, it is necessary to annotate scene feature dimensions such as task context, emotional state, and scene atmosphere to obtain feature labels corresponding to scene feature dimensions such as task context, emotional state, and scene atmosphere. In this embodiment, the feature labels annotated for the scene feature dimensions are defined as second feature labels.

[0097] In a specific implementation, the second feature label is a specific numerical value or classification label for the scene feature dimension. For example, a specific numerical value of 0-10 is used to label the emotional state, or a discrete classification label such as "joy", "sadness", "anger" is used to label the emotional state. The specific labeling method used is determined according to the specific game and the developer's settings, and is not limited here.

[0098] Exemplarily, the social relationship features and interaction scene features obtained based on the multi-dimensional social feature and scene feature annotation system are as follows:

[0099] First, social relationship characteristics include at least:

[0100] 1. Favorability label: This is optional. If there is a specific value for favorability in the game (such as 0-100 points), the specific value can be used as the label directly. Otherwise, discrete classification labels such as "lover", "close friend", "acquaintance", "unfamiliar", "stranger", etc. are used to directly describe the emotional intimacy between two virtual characters.

[0101] 2. Status difference label: This is optional. If there is a specific numerical value of the strategy progress in the game (such as 0-100%), the specific numerical value can be used directly as the label. Otherwise, discrete classification labels such as "master and servant", "companion", "superior and subordinate", "no status difference" and so on are used to directly describe the relative status difference between the two virtual characters in the game social structure or game organization.

[0102] 3. Cultural background label: optional. Use discrete classification labels, such as "close contact culture" and "courtesy alienation culture", to directly describe the interpersonal communication space preferences of the game culture circle in which the virtual character is located.

[0103] 4. Personality trait label, optional, uses continuous values ​​[-1,1] to characterize the virtual character's tendency in various personality dimensions, such as introversion / extroversion (-1 to 1), rationality / emotionality (-1 to 1), humility / arrogance (-1 to 1), etc.

[0104] Second, the scene feature vector includes at least:

[0105] 1. Task context labels, optional, use discrete classification labels, such as "main task", "side task", "chat dialogue", "emergency", etc., to directly describe the task objectives of the interactive behaviors corresponding to the virtual characters.

[0106] 2. Emotional state label, optionally, using discrete classification labels such as "joy", "sadness", "anger", "calm", etc., to directly describe the current emotional state of the virtual character in the interactive scene.

[0107] 3. Scene atmosphere label: Optional, use discrete classification labels such as "serious", "relaxed", "romantic", "tense", etc. to directly describe the overall atmosphere of the interaction scene.

[0108] In an optional implementation provided in this embodiment, the large language model may be trained using language texts including training samples, which may specifically include the following steps S31 to S32:

[0109] Step S31, based on a preset basic example, convert each training sample in the first training sample set into a language text that conforms to the basic example.

[0110] Step S32: input the language text into the large language model so that the large language model can learn from the language text to infer the interaction range parameters according to the social relationship characteristics and the interaction scene characteristics.

[0111] The basic example can be understood as the framework of the language text. By filling the social relationship features, interaction scene features, and interaction range parameters in the training sample into the basic example, the language text corresponding to the training sample can be obtained.

[0112] Exemplarily, the language text converted from the training samples may be: <"Xiao Ming"["lover","protagonist","Western culture","rational and cold"],"Xiao Hong"["lover","supporting role","Western culture","rational and strong"],"private conversation","joyful","romantic",1.2>, which means that in a pleasant and romantic private conversation scenario between two virtual characters, considering that the two virtual characters are lovers but rational Westerners, a natural conversation distance should be 1.2 meters.

[0113] It should be noted that the pre-trained large language model obtained through the pre-training step can be applied to game tasks that infer a reasonable virtual character interaction range based on social relationship characteristics and interaction scene characteristics after fine-tuning the training samples as above <social relationship characteristics, interaction scene characteristics, interaction range parameters>. Similarly, after fine-tuning with different training data, the large language model can also be applied to other game tasks, such as virtual character dialogue generation tasks, virtual character behavior decision-making tasks, etc. Exemplarily, the training samples required for training a large language model that can infer reasonable virtual character dialogues based on social relationship characteristics and interaction scene characteristics can be <social relationship characteristics, interaction scene characteristics, dialogue text>.

[0114] Based on the above steps, after training a large language model that can infer a reasonable virtual character interaction range based on social relationship characteristics and interaction scene characteristics, the large language model can be deployed in a game application so that during the game, based on the real-time obtained social relationship characteristics and interaction scene characteristics, the virtual character interaction range parameters that meet the character characteristics of the first virtual character and the scene characteristics of the current interaction scene can be inferred.

[0115] In an optional implementation, inputting the social relationship features and the interaction scene features into a pre-trained large language model may specifically include the following steps S431 to S432:

[0116] Step S431, based on a preset basic example, encapsulates the social relationship features and the interaction scene features into a language text that conforms to the basic example.

[0117] Step S432: input the language text into a pre-trained large language model.

[0118] Since the advantage of the large language model is that it understands natural language, just like the training samples when training the large language model, based on basic examples, the social relationship features and interaction scene features obtained in real time during the game can be converted into language text, and the language text can be input into the pre-trained large language model.

[0119] In a specific implementation, the preset basic example must follow the following principles:

[0120] First, the description of the interaction range in the basic example is consistent with the representation parameters of the interaction range in the game. For example, if the game uses a radius to represent the interaction range, the basic example should ask for a reasonable interaction trigger radius; another example is that if the game uses a field of view angle to represent the interaction range, the basic example should ask for the reasonable value range of the interaction trigger field of view angle.

[0121] Second, the inquiry about the interaction range in the basic example should be as specific as possible, refined to specific numerical values and units, to reduce the ambiguity of the output of the large language model and obtain directly usable parameter values.

[0122] Third, if there is a value range for a certain interaction range parameter, the range limit can be stated in the basic example to constrain the output of the large language model.

[0123] Exemplarily, the basic example can be: Currently, the social relationship characteristics between the first virtual character and the second virtual character are <relationship label, identity label, cultural label, personality label>, the first virtual character is <NPC name, [relationship label, identity label, cultural label, personality label]>, and the current interaction scenario is <task type, emotional state, scene atmosphere>. Please infer a reasonable and natural <interaction range parameter name> (unit: <unit name>) based on the above context.

[0124] It should be noted that when there are multiple forms of interaction areas in the game, a corresponding basic example can be created for each form. During the game operation, according to the upcoming interaction type, the corresponding basic example can be selected to generate language text and input it into the pre-trained large language model.

[0125] In a specific implementation, the interaction range output by the large language model based on the language text is also represented in the form of language text.

[0126] Exemplarily, the input language text is: Currently, the social relationship characteristics between the first virtual character and the second virtual character are <"lovers", "protagonist", "Western culture", "rational and cold">, the first virtual character is <"Xiaohong", ["lovers", "supporting role", "Western culture", "rational and strong"]>, and the current interaction scenario is <"chatting dialogue", "relaxed", "pleasant">. Please infer a reasonable interaction trigger radius (unit: meter) based on the above context. Input this language text into the pre-trained large language model, and the language text output by the large language model can be: Considering that the first virtual character and the second virtual character are in a lover relationship, having a private chatting dialogue in a relaxed and pleasant atmosphere, but both are Westerners with rational personalities, a reasonable interaction trigger radius is 1.2 meters.

[0127] For example, the input language text is: the current social relationship characteristics between the first virtual character and the second virtual character are <"monarch and minister", "protagonist", "Oriental culture", "loyal and humble">, the first virtual character is <"Zhang San", ["teacher", "supporting role", "Oriental culture", "wise and kind"]>, the current interaction scene is <"important main task", "serious", "tense">, please infer a reasonable and natural interaction triggering field of view angle range (unit: degree) based on the above context. The language text is input into the pre-trained large language model, and the language text output by the large language model can be: considering that the first virtual character and the second virtual character are in a monarch-minister and teacher-student relationship, in a major task interaction, the atmosphere is serious and tense, and both parties are oriental people who value etiquette, a reasonable interaction triggering field of view angle is within the range of 40 to 60 degrees.

[0128] Step S440, adjusting the interaction range of the first virtual character in the game scene according to the interaction range parameter, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

[0129] In an optional implementation, the interaction range parameter is a specific interaction range value, and the interaction range corresponding to the first virtual character can be obtained based on the interaction range parameter. For example, the interaction range parameter is "interaction trigger radius is 1.2 meters", then the interaction range for the first virtual character can be obtained as "a circle with a radius of 1.2 meters".

[0130] In another optional implementation, the interaction range parameter is expressed in the form of a scaling factor and / or offset distance relative to the basic interaction range, and using the interaction range parameter to adjust the interaction range of the first virtual character in the game scene is to use the interaction range parameter to adjust the basic interaction range to obtain the adjusted interaction range. Exemplarily, the interaction range parameter is "reducing the radius of the basic interaction range by 20%", assuming that the basic interaction range is "a circle with a radius of 1.0 meters", then, using the interaction range parameter to adjust the basic interaction range can obtain the adjusted interaction range "a circle with a radius of 0.8 meters".

[0131] After adjusting the interaction range of the first virtual character in the game scene based on the interaction range parameter, when the second virtual character enters the adjusted interaction range, the interaction behavior corresponding to the first virtual character can be triggered. In this embodiment, the interaction behavior corresponding to the first virtual character is defined as the first interaction behavior. The second virtual character can be any one of the multiple virtual characters except the first virtual character, specifically, the player character that is about to interact with the first virtual character.

[0132] In an optional implementation, the first interactive behavior may be an attack behavior of the first virtual character attacking the second virtual character, or a communication behavior of the first virtual character talking to the second virtual character, or a transfer behavior of the first virtual character transferring a virtual item to the second virtual character, without specific limitation. Optionally, before the first interactive behavior corresponding to the first virtual character is triggered, an interactive control for controlling the first interactive behavior of the first virtual character is displayed on the graphical user interface, so that the player can independently control the interaction between the first virtual character and the second virtual character.

[0133] The following is an example of the information processing method in the game provided by this embodiment from the perspective of the player. Figure 5 is a schematic diagram of the information processing method in the game provided by this embodiment.

[0134] As shown in FIG5(a), a graphical user interface is provided by the terminal device of the current player, and a portion of a game scene 510 and a mobile control 520 are displayed on the graphical user interface. The game scene 510 includes a plurality of virtual characters, wherein the virtual character 511 is a virtual character controlled by the current player, and the virtual character 512 is a non-player character in the game.

[0135] As shown in FIG. 5( b ), in response to the player's rightward dragging operation on the movement control 520 , the virtual character 511 is controlled to move toward the virtual character 512 .

[0136] As shown in Fig. 5(c), in response to the virtual character 511 moving to five meters (preset distance) from the virtual character 512, the interaction range parameter of the virtual character 512 is determined based on the method provided in this embodiment. Assume that the interaction range parameter for the virtual character 512 determined by the pre-trained large language model is "interaction trigger radius is 1.2 meters".

[0137] As shown in FIG5(d), according to the social range parameter "interaction trigger radius is 1.2 meters", the basic social range 531 (interaction trigger radius is 1.5 meters) is adjusted to the interaction range 532 (interaction trigger radius is 1.2 meters) of the virtual character 512 in the game scene 510. In an actual game, the interaction range may not be displayed in the graphical user interface. For ease of understanding, the basic social range 531 and the updated interaction range 532 are displayed in this diagram.

[0138] As shown in FIG5(e), in response to the virtual character 511 moving into the interaction range 532, the interaction behavior of the virtual character 512 is triggered. In this diagram, the triggered interaction behavior is a dialogue behavior, and a dialogue text 540 of the virtual character 512, "Item A is at ××", is displayed on the graphical user interface.

[0139] The first embodiment described above provides an information processing method in a game. The method is based on a pre-trained large language model, and according to the character information of the first virtual character and the interaction scene corresponding to the first virtual character, determines the interaction range parameters that meet the social relationship characteristics of the first virtual character and the interaction scene characteristics, so that the first virtual character and the second virtual character can trigger interaction within a reasonable interaction range. First, the interaction range parameters in the method are output by a pre-trained large language model, and do not require manual configuration, which reduces the manpower cost of game development and improves game development efficiency. Second, the interaction range parameters in the method are determined by a pre-trained large language model based on the character information of the first virtual character and the interaction scene corresponding to the first virtual character, that is, when the first virtual character and / or the interaction scene are different, the interaction range parameters output by the pre-trained large language model are different, and the interaction range adjusted based on the interaction range parameters is also different, which increases the flexibility of interaction between virtual characters during the game. Third, the interaction range parameters in the method are determined based on the social relationship characteristics and interaction scene characteristics of the first virtual character, that is, the interaction range of the first virtual character in the game scene adjusted based on the interaction range parameters is in line with the real psychological expectations of interpersonal communication, can bring a real interpersonal atmosphere to the players, and enhance the players' immersion and sense of substitution in the game.

[0140] A second embodiment of the present application provides an information processing device in a game, wherein the game scene of the game includes a plurality of virtual characters. Figure 6 It is a schematic diagram of the structure of the information processing device in the game provided by this embodiment.

[0141] like Figure 6 As shown, the information processing device in the game provided by this embodiment includes: a role determination unit 601, a feature determination unit 602, an interaction parameter determination unit 603, and an interaction range adjustment unit 604.

[0142] The role determination unit 601 is used to determine a first virtual role from the multiple virtual roles.

[0143] The feature determination unit 602 is used to obtain the role information of the first virtual role and the interaction scene corresponding to the first virtual role, and determine the social relationship features and interaction scene features of the first virtual role according to the role information and the interaction scene.

[0144] Optionally, the first virtual character is a non-player character;

[0145] The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes:

[0146] In response to monitoring a first game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the first game event at least includes a game process change event of the game.

[0147] Optionally, the first virtual character is a player character;

[0148] The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes:

[0149] In response to monitoring a second game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the second game event at least includes a character state change event of the player character.

[0150] Optionally, determining the social relationship characteristics and interaction scene characteristics of the first virtual character according to the character information and the interaction scene includes:

[0151] Based on the role information and the interaction scenario, extracting feature values ​​corresponding to a plurality of preset social feature dimensions and feature values ​​corresponding to a plurality of preset scene feature dimensions, wherein the social feature dimensions and the scene feature dimensions are associated with training samples when training the large language model;

[0152] Determine a feature label corresponding to each social feature dimension according to a mapping relationship between a feature value and a feature label preset for each social feature dimension;

[0153] Combining the feature labels corresponding to each of the social feature dimensions into the social relationship feature according to a preset structure;

[0154] Determine a feature label corresponding to each of the scene feature dimensions according to a mapping relationship between a feature value and a feature label preset for each of the scene feature dimensions;

[0155] The feature labels corresponding to each of the scene feature dimensions are combined into the interactive scene feature according to a preset structure.

[0156] The interaction parameter determination unit 603 is used to input the social relationship characteristics and the interaction scene characteristics into a pre-trained large language model, and determine the interaction range parameters for the first virtual character through the large language model.

[0157] Optionally, inputting the social relationship features and the interaction scene features into a pre-trained large language model includes:

[0158] Based on a preset basic example, the social relationship feature and the interaction scenario feature are encapsulated into a language text that conforms to the basic example;

[0159] The language text is input into the pre-trained large language model.

[0160] The interaction range adjustment unit 604 is used to adjust the interaction range of the first virtual character in the game scene according to the interaction range parameter, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

[0161] Optionally, the device further includes a model training unit; the model training unit is used to:

[0162] Acquire a first training sample set, the first training sample set including at least one training sample, the training sample including social relationship features, interaction scene features, and interaction range parameters annotated for a third virtual character in the game, the third virtual character being any one of a plurality of virtual characters preset in the game;

[0163] Using the first training sample set to train the large language model, so that the large language model learns to infer interaction range parameters according to social relationship features and interaction scene features;

[0164] Generate the pre-trained large language model.

[0165] Optionally, the social relationship feature includes a first feature label annotated for at least one of the following social feature dimensions:

[0166] Favorability, status differences, cultural background, personality traits;

[0167] The favorability is used to characterize the emotional intimacy between the third virtual character and the fourth virtual character, the status difference is used to characterize the relative status of the third virtual character and the fourth virtual character in the game, the cultural background is used to characterize the interpersonal communication preference of the interactive scene in which the third virtual character is located, and the personality trait is used to characterize the tendency of the third virtual character in multiple personality dimensions;

[0168] The fourth virtual character is any one of the multiple virtual characters preset in the game except the third virtual character.

[0169] Optionally, the interaction scene feature includes a second feature label annotated for at least one of the following scene feature dimensions:

[0170] Task context, emotional state, and scene atmosphere;

[0171] Among them, the task context is used to characterize the task goal of the second interactive behavior corresponding to the third virtual character; the emotional state is used to characterize the emotion of the third virtual character in the interactive scene, and the scene atmosphere is used to characterize the overall atmosphere of the interactive scene.

[0172] Optionally, the using the first training sample set to train the large language model includes:

[0173] Based on a preset basic example, converting each of the training samples in the first training sample set into a language text that conforms to the basic example;

[0174] The language text is input into the large language model so that the large language model learns from the language text to infer the interaction range parameter according to the social relationship characteristics and the interaction scene characteristics.

[0175] Optionally, before the step of using the first training sample set to train the large language model, the method is further configured to:

[0176] Obtaining a second training sample set, the second training sample set comprising a pre-constructed game knowledge graph and a plurality of game text data; wherein the game knowledge graph comprises a plurality of abstract concepts, a plurality of specific objects, and associations between the plurality of abstract concepts, between the plurality of specific objects, and between the plurality of abstract concepts and the plurality of specific objects in the game, and the plurality of game text data comprises at least a game plot text and a game log text;

[0177] Using the second training sample set to train a large language model so that the large language model learns game knowledge;

[0178] Generate a pre-trained large language model;

[0179] The using the first training sample set to train the large language model comprises:

[0180] The pre-trained large language model is trained using the first training sample set.

[0181] A third embodiment of the present application provides a terminal device, Figure 7 It is a schematic diagram of the structure of the terminal device provided in this embodiment.

[0182] like Figure 7 As shown, the terminal device provided in this embodiment includes: a memory 701, a processor 702, and a display 703;

[0183] The memory 701 is used to store computer instructions for executing the information processing method in the game;

[0184] The processor 702 is configured to execute computer instructions stored in the memory 701 to perform the following operations:

[0185] determining a first virtual character from among the plurality of virtual characters;

[0186] Acquire role information of the first virtual role and an interaction scene corresponding to the first virtual role, and determine social relationship characteristics and interaction scene characteristics of the first virtual role according to the role information and the interaction scene;

[0187] Inputting the social relationship feature and the interaction scene feature into a pre-trained large language model, and determining an interaction range parameter for the first virtual character through the large language model;

[0188] The interaction range of the first virtual character in the game scene is adjusted according to the interaction range parameter, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

[0189] Optionally, the first virtual character is a non-player character;

[0190] The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes:

[0191] In response to monitoring a first game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the first game event at least includes a game process change event of the game.

[0192] Optionally, the first virtual character is a player character;

[0193] The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes:

[0194] In response to monitoring a second game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the second game event at least includes a character state change event of the player character.

[0195] Optionally, also execute:

[0196] Acquire a first training sample set, the first training sample set including at least one training sample, the training sample including social relationship features, interaction scene features, and interaction range parameters annotated for a third virtual character in the game, the third virtual character being any one of a plurality of virtual characters preset in the game;

[0197] Using the first training sample set to train the large language model, so that the large language model learns to infer interaction range parameters according to social relationship features and interaction scene features;

[0198] Generate the pre-trained large language model.

[0199] Optionally, the social relationship feature includes a first feature label annotated for at least one of the following social feature dimensions:

[0200] Favorability, status differences, cultural background, personality traits;

[0201] The favorability is used to characterize the emotional intimacy between the third virtual character and the fourth virtual character, the status difference is used to characterize the relative status of the third virtual character and the fourth virtual character in the game, the cultural background is used to characterize the interpersonal communication preference of the interactive scene in which the third virtual character is located, and the personality trait is used to characterize the tendency of the third virtual character in multiple personality dimensions;

[0202] The fourth virtual character is any one of the multiple virtual characters preset in the game except the third virtual character.

[0203] Optionally, the interaction scene feature includes a second feature label annotated for at least one of the following scene feature dimensions:

[0204] Task context, emotional state, and scene atmosphere;

[0205] Among them, the task context is used to characterize the task goal of the second interactive behavior corresponding to the third virtual character; the emotional state is used to characterize the emotion of the third virtual character in the interactive scene, and the scene atmosphere is used to characterize the overall atmosphere of the interactive scene.

[0206] Optionally, the using the first training sample set to train the large language model includes:

[0207] Based on a preset basic example, converting each of the training samples in the first training sample set into a language text that conforms to the basic example;

[0208] The language text is input into the large language model so that the large language model learns from the language text to infer the interaction range parameter according to the social relationship characteristics and the interaction scene characteristics.

[0209] Optionally, before the step of using the first training sample set to train the large language model, the following steps are further performed:

[0210] Obtaining a second training sample set, the second training sample set comprising a pre-constructed game knowledge graph and a plurality of game text data; wherein the game knowledge graph comprises a plurality of abstract concepts, a plurality of specific objects, and associations between the plurality of abstract concepts, between the plurality of specific objects, and between the plurality of abstract concepts and the plurality of specific objects in the game, and the plurality of game text data comprises at least a game plot text and a game log text;

[0211] Using the second training sample set to train a large language model so that the large language model learns game knowledge;

[0212] Generate a pre-trained large language model;

[0213] The using the first training sample set to train the large language model comprises:

[0214] The pre-trained large language model is trained using the first training sample set.

[0215] Optionally, determining the social relationship characteristics and interaction scene characteristics of the first virtual character according to the character information and the interaction scene includes:

[0216] Based on the role information and the interaction scenario, extracting feature values ​​corresponding to a plurality of preset social feature dimensions and feature values ​​corresponding to a plurality of preset scene feature dimensions, wherein the social feature dimensions and the scene feature dimensions are associated with training samples when training the large language model;

[0217] Determine a feature label corresponding to each social feature dimension according to a mapping relationship between a feature value and a feature label preset for each social feature dimension;

[0218] Combining the feature labels corresponding to each of the social feature dimensions into the social relationship feature according to a preset structure;

[0219] Determine a feature label corresponding to each of the scene feature dimensions according to a mapping relationship between a feature value and a feature label preset for each of the scene feature dimensions;

[0220] The feature labels corresponding to each of the scene feature dimensions are combined into the interactive scene feature according to a preset structure.

[0221] Optionally, inputting the social relationship features and the interaction scene features into a pre-trained large language model includes:

[0222] Based on a preset basic example, the social relationship feature and the interaction scenario feature are encapsulated into a language text that conforms to the basic example;

[0223] The language text is input into the pre-trained large language model.

[0224] The display 703 is used to display the game scene.

[0225] A fourth embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium includes computer instructions. When the computer instructions are executed by a processor, they are used to implement the methods described in the embodiments of the present application.

[0226] It should be noted that the relational terms such as "first" and "second" in this article are only used to distinguish one entity or operation from another entity or operation, and do not require or imply any actual relationship or order between these entities or operations. In addition, the words "include", "have", "include" and "includes" and other similar forms are the same in meaning, and the end of any one or more items after any of the above words is open-ended, and any of the above nouns does not mean that the one or more items have been exhaustively listed or are limited to the one or more items listed.

[0227] As used herein, unless expressly stated otherwise, the term "or" includes all possible combinations, except those that are not feasible. For example, if it is expressed that a database may include A or B, then unless otherwise specifically stated or not feasible, it may include database A, or B, or A and B. As a second example, if it is expressed that a database may include A, B, or C, then unless otherwise specifically stated or not feasible, the database may include database A, or B, or C, or A and B, or A and C, or B and C, or A, B, and C.

[0228] It is worth noting that the above embodiments can be implemented by hardware or software (program code), or a combination of hardware and software. If implemented by software, it can be stored in the above-mentioned computer-readable medium. When the software is executed by a processor, it can execute the above-mentioned disclosed method. The computing unit and other functional units described in the present disclosure can be implemented by hardware or software, or a combination of hardware and software. Those of ordinary skill in the art will also understand that the above-mentioned multiple modules / units can be combined into one module / unit, and each of the above-mentioned modules / units can be further divided into multiple sub-modules / sub-units.

[0229] In the above detailed description, the embodiments have been described with reference to many specific details, which may vary from implementation to implementation. Certain adaptations and modifications may be made to the embodiments. For those skilled in the art, other embodiments may be readily apparent from the specific embodiments disclosed in the present application. This specification and examples are for exemplary purposes only, and the true scope and essence of the present application are described by the claims. The sequence of steps shown in the diagrams is also for the purpose of explanation only and is not intended to be limited to any particular steps or sequence. Therefore, those skilled in the art will appreciate that these steps may be performed in different orders when implementing the same method.

[0230] In the drawings and detailed description of the present application, exemplary embodiments are disclosed. However, many variations and modifications may be made to these embodiments. Accordingly, although specific terms are used, these terms are only general and descriptive and not for the purpose of limitation.

Claims

1. A method for processing information in a game, characterized in that: The game scene of the game includes a plurality of virtual characters, and the method includes: determining a first virtual character from among the plurality of virtual characters; Acquire role information of the first virtual role and an interaction scene corresponding to the first virtual role, and determine social relationship characteristics and interaction scene characteristics of the first virtual role according to the role information and the interaction scene; Inputting the social relationship feature and the interaction scene feature into a pre-trained large language model, and determining an interaction range parameter for the first virtual character through the large language model; The interaction range of the first virtual character in the game scene is adjusted according to the interaction range parameter, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

2. The method according to claim 1, characterized in that The first virtual character is a non-player character; The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes: In response to monitoring a first game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the first game event at least includes a game process change event of the game.

3. The method according to claim 1, characterized in that The first virtual character is a player character; The acquiring the character information of the first virtual character and the interaction scene corresponding to the first virtual character includes: In response to monitoring a second game event, character information of the first virtual character and an interaction scene corresponding to the first virtual character are obtained; wherein the second game event at least includes a character state change event of the player character.

4. The method according to claim 1, characterized in that: The method further comprises: Acquire a first training sample set, the first training sample set including at least one training sample, the training sample including social relationship features, interaction scene features, and interaction range parameters annotated for a third virtual character in the game, the third virtual character being any one of a plurality of virtual characters preset in the game; Using the first training sample set to train the large language model, so that the large language model learns to infer interaction range parameters according to social relationship features and interaction scene features; Generate the pre-trained large language model.

5. The method according to claim 4, characterized in that The social relationship feature includes a first feature tag annotated for at least one of the following social feature dimensions: Favorability, status differences, cultural background, personality traits; The favorability is used to characterize the emotional intimacy between the third virtual character and the fourth virtual character, the status difference is used to characterize the relative status of the third virtual character and the fourth virtual character in the game, the cultural background is used to characterize the interpersonal communication preference of the interactive scene in which the third virtual character is located, and the personality trait is used to characterize the tendency of the third virtual character in multiple personality dimensions; The fourth virtual character is any one of the multiple virtual characters preset in the game except the third virtual character.

6. The method according to claim 4, characterized in that The interaction scene feature includes a second feature tag annotated for at least one of the following scene feature dimensions: Task context, emotional state, and scene atmosphere; Among them, the task context is used to characterize the task goal of the second interactive behavior corresponding to the third virtual character; the emotional state is used to characterize the emotion of the third virtual character in the interactive scene, and the scene atmosphere is used to characterize the overall atmosphere of the interactive scene.

7. The method according to claim 4, characterized in that The using the first training sample set to train the large language model comprises: Based on a preset basic example, converting each of the training samples in the first training sample set into a language text that conforms to the basic example; The language text is input into the large language model so that the large language model learns from the language text to infer the interaction range parameter according to the social relationship characteristics and the interaction scene characteristics.

8. The method according to claim 4, characterized in that Before the step of using the first training sample set to train the large language model, the method further includes: Obtaining a second training sample set, the second training sample set comprising a pre-constructed game knowledge graph and a plurality of game text data; wherein the game knowledge graph comprises a plurality of abstract concepts, a plurality of specific objects, and associations between the plurality of abstract concepts, between the plurality of specific objects, and between the plurality of abstract concepts and the plurality of specific objects in the game, and the plurality of game text data comprises at least a game plot text and a game log text; Using the second training sample set to train a large language model so that the large language model learns game knowledge; Generate a pre-trained large language model; The using the first training sample set to train the large language model comprises: The pre-trained large language model is trained using the first training sample set.

9. The method according to claim 1, characterized in that: The determining of the social relationship characteristics and the interaction scene characteristics of the first virtual character according to the character information and the interaction scene includes: Based on the role information and the interaction scenario, extracting feature values ​​corresponding to a plurality of preset social feature dimensions and feature values ​​corresponding to a plurality of preset scene feature dimensions, wherein the social feature dimensions and the scene feature dimensions are associated with training samples when training the large language model; Determine a feature label corresponding to each social feature dimension according to a mapping relationship between a feature value and a feature label preset for each social feature dimension; Combining the feature labels corresponding to each of the social feature dimensions into the social relationship feature according to a preset structure; Determine a feature label corresponding to each of the scene feature dimensions according to a mapping relationship between a feature value and a feature label preset for each of the scene feature dimensions; The feature labels corresponding to each of the scene feature dimensions are combined into the interactive scene feature according to a preset structure.

10. The method according to claim 1, characterized in that The step of inputting the social relationship features and the interaction scene features into a pre-trained large language model includes: Based on a preset basic example, the social relationship feature and the interaction scenario feature are encapsulated into a language text that conforms to the basic example; The language text is input into the pre-trained large language model.

11. An information processing device in a game, characterized in that: The game scene of the game includes a plurality of virtual characters, and the device includes: a character determination unit, a feature determination unit, an interaction parameter determination unit, and an interaction range adjustment unit; The role determination unit is used to determine a first virtual role from the multiple virtual roles; The feature determination unit is configured to obtain the role information of the first virtual role and the interaction scene corresponding to the first virtual role, and determine the social relationship feature and the interaction scene feature of the first virtual role according to the role information and the interaction scene; The interaction parameter determination unit is used to input the social relationship feature and the interaction scene feature into a pre-trained large language model, and determine the interaction range parameter for the first virtual character through the large language model; The interaction range adjustment unit is used to adjust the interaction range of the first virtual character in the game scene according to the interaction range parameter, so that when a second virtual character among the multiple virtual characters enters the adjusted interaction range, a first interaction behavior corresponding to the first virtual character is triggered.

12. A terminal device, characterized in that: include: Memory, processor, and display; The memory is used to store one or more computer instructions; The processor is configured to execute the one or more computer instructions to implement the method according to any one of claims 1 to 10; The display is used to display the game scene.

13. A computer-readable storage medium having one or more computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the method according to any one of claims 1 to 10 is performed.

Images