Social game recommendation method and system based on artificial intelligence

Abstract

The invention discloses a social game recommendation method and system based on artificial intelligence, relates to the technical field of game recommendation, and identifies low social pressure sensitivity users, obtains virtual image features, skill features, emotion tags and text chat keywords of multi-game platforms of the low social pressure sensitivity users, and provides a social game recommendation method and system. The social contact type of the user with low social contact pressure sensitivity is output and obtained; obtaining an initial recommendation value of each to-be-recommended game in combination with the core recommendation parameters of the to-be-recommended games and the social types of the low social pressure sensitivity users; performing recommendation sorting on the games according to the initial recommendation values; and based on the in-game data and emotion data output monitored in real time, obtaining an adjustment optimization value of each game, and further generating a comprehensive recommendation value of each game. The problems that due to the fact that a traditional game recommendation system lacks accurate recognition and dynamic adaptation on user social contact pressure sensitivity, recommendation content is not matched with user requirements, and the user loses the social contact pressure are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of game recommendation, and particularly to a social game recommendation method and system based on artificial intelligence. Background Art

[0002] With the rapid development of mobile Internet and artificial intelligence technologies, the scale of the social game market has been continuously expanding, and users' demands for social games have become increasingly diverse and personalized. At present, the background technologies of social game recommendation mainly include content-based recommendation technology, collaborative filtering-based recommendation technology, and social-relationship-based recommendation technology. However, when facing the complex and changeable social needs and game scenarios of users, these traditional recommendation technologies gradually expose some limitations.

[0003] Existing recommendation technologies often only focus on users' game behaviors and preferences, and lack in-depth analysis of users' social characteristics (such as social stress sensitivity, social types, etc.); traditional recommendation technologies perform static recommendations based on users' historical data and cannot respond in a timely manner to the real-time feedback and changing needs of users during the game process; moreover, when selecting recommendation parameters, existing recommendation technologies often only consider some features of the game, such as game type or popularity, lack of comprehensive consideration of the core recommendation parameters of the game, and fail to fully integrate users' social behaviors and game characteristics, resulting in the recommended games being difficult to meet users' expectations in terms of social interaction.

[0004] Therefore, aiming at the above problems, there is an urgent need for a social game recommendation method and system based on artificial intelligence. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a social game recommendation method and system based on artificial intelligence, which solves the problem that the recommended content does not match the user's needs and users lose due to social stress caused by the lack of accurate identification and dynamic adaptation of the user's social stress sensitivity in the traditional game recommendation system.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions: A social game recommendation method based on artificial intelligence, comprising the following steps: S1, identifying users with low social stress sensitivity based on the social avoidance index in the user's historical behavior data; S2, obtaining the virtual image features and skill features of users with low social stress sensitivity on multiple game platforms, and identifying the emotional labels and text chat keywords in the multiple game platforms of users with low social stress sensitivity, and then outputting the social types of users with low social stress sensitivity; S3, extracting the core recommendation parameters of the games to be recommended, and then combining the social types of users with low social stress sensitivity to obtain the initial recommendation values of each game to be recommended; S4, performing recommendation ranking on the games according to the magnitude of the initial recommendation values, and simultaneously monitoring the in-game data and user emotional data during the game process of users with low social stress sensitivity in real time; S5, outputting the adjustment and optimization values of each game based on the in-game data and emotional data, combining the initial recommendation values and the adjustment and optimization values to generate the comprehensive recommendation values of each game, and re-performing recommendation ranking on the games according to the comprehensive recommendation values, continuously updating the recommended social games for users.

[0007] Further, the specific steps of S1 are as follows: Extracting the social avoidance index, where the social avoidance index includes the friend number increase rate, social interaction frequency, chat silence ratio, and team refusal times; Using the social avoidance index as the input, training with a support vector machine to construct a user social stress sensitivity model, and outputting the social stress sensitivity value of the user; Comparing the social stress sensitivity value with the sensitivity threshold, when the social stress sensitivity value is greater than or equal to the sensitivity threshold, marking the user as a user with high social stress sensitivity, entering the single-player task game recommendation branch, and recommending single-player task-dominated games; When the social stress sensitivity value is less than the sensitivity threshold, marking the user as a user with low social stress sensitivity, triggering the social multi-player task game recommendation requirement.

[0008] Further, the specific steps of outputting the social types of users with low social stress sensitivity are as follows: Obtaining the virtual image features, skill features, emotional labels, and text chat keywords of known social types as training samples, and then using the training samples to train the decision tree algorithm to construct a decision forest model composed of multiple decision trees; Using the virtual image features, skill features, emotional labels, and text chat keywords of users with low social stress sensitivity as the input of the decision forest model, and outputting the social types of users with low social stress sensitivity, where the social types include leadership type, auxiliary type, observation type, and follower type.

[0009] Furthermore, the virtual image feature acquisition and analysis are as follows: acquiring multi-source heterogeneous data of the user's virtual image, constructing a virtual image feature vector space matrix, and numerically representing the virtual image through feature encoding and attribute weighting; the skill feature acquisition and analysis are as follows: acquiring skill structured data, extracting skill text features, constructing a skill knowledge graph network, and realizing multi-dimensional quantification of skill features through level mapping, frequency normalization, and attribute weighting; the emotion feature acquisition and analysis are as follows: acquiring the user's multi-platform voice chat records based on user authorization, performing hierarchical emotion annotation on the voice chat records, and identifying the comprehensive emotion index based on the label frequency and polarity weight; the text chat keyword acquisition and analysis are as follows: screening keywords through the TF-IDF algorithm combined with a domain dictionary, and dynamically adjusting the weight coefficient according to the scenario type to achieve semantic quantification.

[0010] Furthermore, the specific steps of S3 are as follows: collecting the core recommendation parameter positive review data of multi-user games corresponding to each social type, where the core recommendation parameter positive review data is specifically the total number of positive reviews of multi-users for a certain core recommendation parameter; by comparing the total number of positive reviews of each core recommendation parameter of users of each social type, using the preference ranking organization method to assign proportionality coefficients to each core recommendation parameter respectively, and determining the proportionality coefficients of each core recommendation parameter of users of each social type; identifying the proportionality coefficients of each core recommendation parameter of the social type to which the low social stress sensitivity users belong, and at the same time combining the core recommendation parameters of the games to be recommended, generating the initial recommendation values of each game to be recommended respectively.

[0011] Furthermore, the core recommendation parameters specifically include social interaction complexity parameters, social role positioning parameters, game atmosphere style parameters, and player matching mechanism parameters.

[0012] Furthermore, the specific steps of S5 are as follows: during the game process of low social stress sensitivity users, real-time acquiring in-game data, where the in-game data includes the participation popularity of activities released in the game and the player interaction frequency in the community; using emotion recognition technology to real-time acquire user emotion data, where the user emotion data includes user pleasure, user concentration, and user stress level; using the in-game data and user emotion data to construct an adjustment and optimization model, and then outputting the adjustment and optimization values of each game; acquiring a fusion coefficient, and combining the initial recommendation value, the adjustment and optimization value, and the fusion coefficient to generate a comprehensive recommendation value, thereby realizing the re-ranking of game recommendations.

[0013] A social game recommendation system based on artificial intelligence, applying the above-mentioned social game recommendation method based on artificial intelligence, includes: a user type recognition module for identifying users with low social stress sensitivity based on social avoidance indicators in user historical behavior data; a social type recognition module for obtaining virtual image features and skill features of users with low social stress sensitivity on multiple game platforms, and identifying emotional tags and text chat keywords in multiple game platforms of users with low social stress sensitivity, and then outputting the social types of users with low social stress sensitivity; an initial recommendation analysis module for extracting core recommendation parameters of games to be recommended, and then obtaining initial recommendation values of each game to be recommended in combination with the social types of users with low social stress sensitivity; a real-time monitoring module for sorting and recommending games according to the magnitude of the initial recommendation values, and simultaneously real-time monitoring in-game data and user emotion data during the game process of users with low social stress sensitivity; an optimization and adjustment module for outputting adjustment and optimization values of each game based on in-game data and emotion data, generating comprehensive recommendation values of each game in combination with the initial recommendation values and the adjustment and optimization values, re-recommending and sorting the games according to the comprehensive recommendation values, and continuously updating the recommended social games for users.

[0014] The present invention has the following beneficial effects:

[0015] The disclosed social game recommendation method and system based on artificial intelligence classify users into two categories of high and low social stress sensitivity by identifying the user's social stress sensitivity. For users with low social stress sensitivity, their virtual image characteristics, skill characteristics, emotion tags, and text chat keywords are further analyzed to determine the user's social type. This multi-dimensional and in-depth user portrait construction enables the recommendation system to accurately capture the personalized needs of users. Compared with traditional recommendation methods, it greatly improves the accuracy and relevance of recommendations, providing users with games that better match their own interests and social preferences. For users with low social stress sensitivity, games are recommended based on their social type and game core recommendation parameters, and in-game data and user emotion data are monitored in real time during the game process to dynamically adjust the recommendation results. It can promptly respond to changes in the user's experience during the game. For example, when the user shows dissatisfaction with the current game atmosphere or low enthusiasm for participating in activities during the game, the system updates the recommendation by adjusting the optimization value to ensure that the user always obtains a good game experience, enhancing the user's satisfaction and loyalty to the recommendation system and the game platform. When recommending games, core recommendation parameters such as the social interaction complexity, social role positioning, game atmosphere style, and player matching mechanism of the game are comprehensively considered, and recommendations are made in combination with the user's social type, which helps the game platform reasonably allocate resources, accurately recommend games to the appropriate user group, improve the exposure rate and user participation of the games, and at the same time reduce the waste of time and energy of users on inappropriate games, achieving an efficient match between game resources and user needs. During the recommendation process, the user's social behaviors and preferences are fully considered. For example, the user type is identified through the social avoidance index, and the social type is determined based on the user's social interaction data in the game. The recommended games can better meet the user's social needs, promote social interaction within the game, enhance the social attributes of the game, form a good game social ecosystem, and improve the attractiveness and user stickiness of the game.

[0016] Of course, not necessarily all the advantages described above need to be achieved simultaneously by any product implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a social game recommendation method based on artificial intelligence according to the present invention.

[0018] Figure 2 It is a structural diagram of a social game recommendation system based on artificial intelligence according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In the embodiments of the present application, a social game recommendation method and system based on artificial intelligence are adopted to achieve accurate game recommendations with low social stress for users through multi-modal data fusion and real-time dynamic optimization, significantly improving user satisfaction, retention rate, and platform business value.

[0020] The overall idea of the embodiments of this application is as follows:

[0021] Based on the social avoidance indicators in the user's historical behavior data, use the support vector machine model to identify users with low social stress sensitivity. Then, for such users, obtain their virtual image features and skill features on multiple game platforms, and at the same time identify emotion tags and text chat keywords. Output the user's social type through the decision forest model to complete the in-depth feature extraction and classification of users, laying a foundation for subsequent accurate recommendations; extract the core recommendation parameters of the games to be recommended, including social interaction complexity, social role positioning, game atmosphere style, and player matching mechanism. Combine the social types of users with low social stress sensitivity, and use the preference ranking organization method to determine the proportional coefficients of each core recommendation parameter, and then calculate the initial recommendation values of each game to be recommended, realizing the initial matching recommendation of games and user social types; after sorting and recommending games according to the initial recommendation values, real-time monitor the in-game data and user emotion data during the user's game process. Based on the real-time data, construct an adjustment and optimization model, calculate the adjustment and optimization values of each game, and fuse them with the initial recommendation values to generate a comprehensive recommendation value. Re-sort the games according to the comprehensive recommendation value, continuously update the recommendations, and realize dynamic and accurate personalized social game recommendations.

[0022] Please refer to Figure 1 , an embodiment of the present invention provides a technical solution: A social game recommendation method based on artificial intelligence, including the following steps: S1, identify users with low social stress sensitivity based on the social avoidance indicators in the user's historical behavior data; S2, obtain the virtual image features and skill features of users with low social stress sensitivity on multiple game platforms, and identify the emotion tags and text chat keywords in multiple game platforms of users with low social stress sensitivity, and then output the social type of users with low social stress sensitivity; S3, extract the core recommendation parameters of the games to be recommended, and then combine the social types of users with low social stress sensitivity to obtain the initial recommendation values of each game to be recommended; S4, recommend and sort the games according to the size of the initial recommendation values, and at the same time real-time monitor the in-game data and user emotion data during the game process of users with low social stress sensitivity; S5, output the adjustment and optimization values of each game based on the in-game data and emotion data, combine the initial recommendation values and the adjustment and optimization values to generate the comprehensive recommendation values of each game, and re-recommend and sort the games according to the comprehensive recommendation values, and continuously update the recommended social games for users.

[0023] Specifically, the specific steps of S1 are as follows: Extract social avoidance indicators, which include the growth rate of the number of friends, the frequency of social interaction, the proportion of chat silence, and the number of times of rejecting team invitations; Use the social avoidance indicators as inputs and train with a support vector machine to construct a user social stress sensitivity model, and output the social stress sensitivity value of the user; Compare the social stress sensitivity value with the sensitivity threshold. When the social stress sensitivity value is greater than or equal to the sensitivity threshold, mark the user as a user with high social stress sensitivity and enter the single-player task game recommendation branch to recommend single-player task-dominated games; When the social stress sensitivity value is less than the sensitivity threshold, mark the user as a user with low social stress sensitivity and trigger the social multi-player task game recommendation requirement.

[0024] In this implementation plan, the growth rate of the number of friends refers to the ratio of the number of newly added friends to the original number of friends of the user within a certain time period, reflecting the willingness and behavior intensity of the user to actively expand social relationships. The higher this ratio, the more actively the user participates in social expansion. On the contrary, it means that the user is relatively passive in establishing social relationships. From the user database of the social game platform, extract the friend list data of the user within a specific time interval (such as the past week, month), count the number of newly added friends and the original number of friends, and obtain it through the ratio of the number of newly added friends to the original number of friends.

[0025] The frequency of social interaction is used to measure the frequency of a user's social interaction with other players in the game, such as the number of occurrences of behaviors such as chatting, teaming up to play games, and participating in social activities, reflecting the user's activity level in the game social scenario. With the help of the game platform's log recording system, collect various social interaction behavior data of the user during the game process, and count the number of social interactions according to the set time unit (such as per hour, per day), which is used as the frequency of social interaction.

[0026] The proportion of chat silence indicates the proportion of the duration when the user is in a silent state in the total duration of the user's participation in game social chat, reflecting the user's participation and enthusiasm in social chat. Record the start and end times of each chat of the user through the game chat system, as well as the time stamps of the user's sent messages during the chat, calculate the total chat duration and the silent duration of the user, and obtain the value of this indicator using the ratio of the silent duration to the total chat duration.

[0027] The number of times of rejecting team invitations indicates the number of times the user chooses to reject after receiving a team invitation from other players in the game, intuitively reflecting the degree of resistance of the user to social teaming behavior. Set a counter in the game teaming system. Whenever the user rejects a team invitation once, the counter automatically increments by 1, and regularly count the number of times of rejecting team invitations within a certain period of time.

[0028] The specific steps to build a user social stress sensitivity model using a support vector machine are as follows: Collect historical user behavior data containing social avoidance indicators such as the growth rate of the number of friends, the frequency of social interaction, the proportion of chat silence, and the number of team rejection times from the user behavior database of the social game platform; Clean the collected data, remove missing values and outliers, and use the normalization method to map indicator data of different magnitudes to the same numerical interval to eliminate the influence of data dimension differences; Use the preprocessed social avoidance indicator data as feature vectors, and label each feature vector with the corresponding user social stress sensitivity category label (high social stress sensitivity is marked as 1, low social stress sensitivity is marked as 0) to form a training dataset; Select an appropriate support vector machine kernel function (such as the radial basis function RBF, polynomial kernel function, etc.), and set hyperparameters such as the penalty factor C; Use the training dataset to train the support vector machine, and optimize the weight vector and bias term of the support vector machine by minimizing the structural risk to build a user social stress sensitivity model; Use the reserved validation dataset to test the trained user social stress sensitivity model, and calculate evaluation indicators such as the classification accuracy, recall rate, and F1 value of the model; According to the evaluation results, adjust the kernel function type and hyperparameter values of the support vector machine, and repeat the training and validation process until the model meets the predetermined performance index requirements. The support vector machine model expression is: Assume the training set is , where is the social avoidance indicator feature vector of the th sample, , corresponding to the growth rate of the number of friends, the frequency of social interaction, the proportion of chat silence, and the number of team rejection times respectively; is the sample label, , 0 represents a user with low social stress sensitivity, and 1 represents a user with high social stress sensitivity. The goal of the sVM is to find a hyperplane such that samples of different classes can be separated by the maximum margin. Its optimization objective function is: ; ; where, is the weight vector, is the bias term, is the slack variable, which is used to handle outliers or linearly inseparable cases in the training set, is the penalty factor, which is used to control the penalty degree for misclassified samples, The larger it is, the heavier the penalty for misclassification, and the more the model tends to reduce classification errors.

[0029] The sensitivity value is calculated as follows: For the feature vector of a new user, its social stress sensitivity value is calculated through the decision function of the SVM model: ; where, sgn is the sign function, when When sgn , it represents users with high social stress sensitivity; when When , it represents users with low social stress sensitivity.

[0030] The method for obtaining the sensitivity threshold is as follows: Based on the experience of game industry experts and the user data analysis accumulated by the game platform operation team for a long time, combined with the research results of the behavior of users of similar social games, set an initial sensitivity threshold as a reference benchmark; it is also possible to collect the social stress sensitivity values of a large number of users and the feedback data of actual game behaviors, analyze indicators such as user recommendation satisfaction and game participation under different threshold settings, and use optimization algorithms such as grid search and random search to perform a traversal search within the threshold range to identify the sensitivity threshold that makes the comprehensive evaluation index reach the optimal; or establish a real-time monitoring mechanism to continuously track the behavior changes of users during the game process and the recommendation effect data. When it is found that the behavior pattern of the user group has changed significantly (such as a large number of users having poor feedback on the recommended games) or the business requirements of the game platform are adjusted, automatically trigger the threshold adjustment program, and recalculate and update the sensitivity threshold based on the current data.

[0031] The specific recommendation steps for single-player mission-dominated games are as follows: Screen out the game set with single-player missions as the core gameplay from the game resource library, and extract the key feature parameters of each game in this set, including but not limited to game mission types (such as adventure, puzzle-solving, cultivation), game difficulty levels, plot richness, mission completion duration, etc.; analyze the historical game behavior data of users with high social stress sensitivity, and extract the preference characteristics of users for aspects such as game mission types, difficulty, and plot; use similarity calculation algorithms such as cosine similarity and Euclidean distance to calculate the matching degree scores of each single-player mission-dominated game and the user preference characteristics; sort the single-player mission-dominated games in descending order according to the matching degree scores, select several games with the highest rankings to form a recommendation list. If there are multiple games with the same matching degree scores, further compare additional indicators such as the user ratings and popularity of the games to determine the final recommendation order, and push the recommendation list to the user client to complete the recommendation process of single-player mission-dominated games.

[0032] By quantifying multi-dimensional social avoidance indicators such as the growth rate of the number of friends and the frequency of social interactions, and combining with the powerful classification ability of the support vector machine, users can be scientifically and accurately classified into two categories: high and low social stress sensitivity. Compared with the traditional user classification method that only relies on subjective judgment or a single indicator, the accuracy and reliability of classification are improved; based on the accurate user classification, differential recommendation strategies are formulated for users with different social stress sensitivities. Single-player task-dominated games are recommended for users with high social stress sensitivity, and social multi-player task game recommendations are triggered for users with low social stress sensitivity to meet the personalized game needs of users and improve the user game experience and recommendation satisfaction; after clarifying the social stress sensitivity of users, the game platform pushes games that meet the user's needs in a targeted manner, avoiding resource waste, improving the game promotion efficiency and user participation, achieving an efficient match between game resources and user needs, and promoting the healthy development of the game platform ecosystem.

[0033] Specifically, the specific steps to output the social types of users with low social stress sensitivity are as follows: Obtain the virtual image features, skill features, emotion labels, and text chat keywords of known social types as training samples, and then use the training samples to train the decision tree algorithm to construct a decision forest model composed of multiple decision trees; Use the virtual image features, skill features, emotion labels, and text chat keywords of users with low social stress sensitivity as the input of the decision forest model, and output the social types of users with low social stress sensitivity. The social types include leader type, auxiliary type, observer type, and follower type.

[0034] The acquisition and analysis of virtual image features are as follows: Obtain the multi-source heterogeneous data of the user's virtual image, construct a virtual image feature vector space matrix, and numerically represent the virtual image by feature coding and attribute weighting. The quantization of virtual image features is as follows: Let the virtual image feature vector be , where is the quantization value of the th feature. For appearance features, such as the hairstyle is encoded as , and the hairstyle type is converted into a numerical value through the coding table. For clothing and dressing, let the quantization value of the clothing type be , and the weight of the matching complexity is , then the quantization value of this part is . And so on, each feature is encoded and weighted, and finally the virtual image feature vector is obtained; The acquisition and analysis of skill features are as follows: Obtain the structured data of skills, extract the skill text features, construct a skill knowledge graph network, and realize the multi-dimensional quantization of skill features through level mapping, frequency normalization, and attribute weighting. The quantization of skill features is as follows: Let the skill feature vector be , where is the th dimension of the Quantization values of each dimension, skill level are mapped to , and normalization processing is performed; the skill usage frequency after normalization is ; for the key attribute words in the skill effect description, let the word quantization value be , and the weight be , then the quantization value of this part is . Combine the quantization values of each dimension to obtain the skill feature vector , and then comprehensively process all the skill feature vectors to obtain the skill feature representation of the user; the acquisition and analysis of emotional features are as follows: obtain the multi-platform voice chat records of the user based on user authorization, perform hierarchical emotional annotation based on the voice chat records, and identify the comprehensive emotional index based on the label frequency and polarity weight. The emotional label quantization is as follows: let the emotional label set be , and the corresponding quantization value be , such as the positive emotion quantization value , the negative emotion quantization value etc. For the voice chat records, count the frequency of occurrence of each emotional label , then the comprehensive emotional index ; the acquisition and analysis of text chat keywords are as follows: screen keywords through the TF-IDF algorithm combined with the domain dictionary, and realize semantic quantization by dynamically adjusting the weight coefficient according to the scene type. The quantization of text chat keywords is as follows: for the keyword in the text chat record, its TF-IDF value is , where is the word frequency of the keyword in the text , is the total number of texts, is the number of texts containing the keyword . Let the weight of the scene type be , then the final quantization value of the keyword is .

[0035] The decision forest consists of . For the input feature vector (including virtual image features, skill features, emotional index, quantization values of text chat keywords), each decision tree classifies according to the splitting rules of the internal nodes, and outputs the prediction result Leadership type, auxiliary type, observational type, follower type , finally, the social type of the user is determined through a voting mechanism, that is, the type with the same prediction result of the majority decision tree is the final social type. If there is a tie, the selection can be made according to the weight of the decision tree or other rules.

[0036] In this implementation plan, the specific steps for training the decision tree algorithm with training samples to construct a decision forest model are as follows: Clean the training sample data such as the virtual image features, skill features, emotion labels, and text chat keywords of the known social types obtained, and remove duplicate, incorrect, or incomplete data records; For numerical data such as virtual image features and skill features, use normalization or standardization methods to map them to a specific numerical interval to eliminate the difference in data dimensions; For text-type emotion labels and text chat keywords, perform operations such as word segmentation and stop word removal, and convert them into numerical vector forms through techniques such as word vector embedding (such as Word2Vec, BERT-Embedding) to form a standardized training sample data set; Randomly select a certain proportion of data from the standardized training sample data set as the training subset, and the remaining data as the validation subset; For each training subset, select an appropriate decision tree splitting criterion (such as information gain, information gain ratio, Gini index), start from the root node, and recursively divide the data set according to the feature attributes of the training samples to construct a single decision tree; During the construction process, set hyperparameters such as the maximum depth, minimum sample split number, and minimum sample leaf node number of the decision tree to prevent overfitting of the decision tree; Use the validation subset to evaluate the constructed decision tree, calculate evaluation indicators such as accuracy, recall rate, and F1 value, and adjust the hyperparameters according to the evaluation results to optimize the decision tree model; Repeat the above decision tree training steps to generate multiple different decision trees; Combine these decision trees into a decision forest model. During the combination process, the Bagging (bootstrap aggregating) strategy can be used, that is, randomly sample the training samples with replacement to generate multiple different training subsets for training each decision tree to increase the difference between decision trees; The random subspace method can also be used. When training each decision tree, randomly select some feature attributes for splitting to further improve the generalization ability and classification performance of the decision forest; The finally formed decision forest model can comprehensively classify and predict the input low social stress sensitivity user feature data and output the social type of the user.

[0037] The multi-source heterogeneous data of the user avatar refers to data from different channels with different data structures and presentation forms, specifically including: visual appearance data, action and pose data, and personalized setting data. Visual appearance data includes the facial features, clothing features, weapons, and tool features of the avatar. Action and pose data includes various action information of the avatar in the game, such as the pose parameters of actions like standing, walking, running, combat attacking, and skill releasing, as well as dynamic attribute data such as the smoothness and speed of the actions. Personalized setting data includes the personalized customization parameters set by the user for the avatar, such as unique facial expressions, personalized voice packs, and exclusive action special effects data. The multi-source heterogeneous data is connected to the official interface of the game platform, and data request instructions are sent according to the interface specifications to obtain the data related to the user avatar stored on the game platform, including static appearance data and dynamic action and pose data, etc.; or, on the premise of user authorization, the hardware devices such as the camera and sensors of the user terminal device (such as mobile phones and computers) are used to collect the real-time action data of the avatar during the game process; it is also possible to obtain the avatar data published or shared by the user from third-party platforms related to the game (such as game communities and avatar design platforms), such as the avatar design works uploaded by the user and the avatar screenshots displayed in the community, to enrich the data sources of the avatar.

[0038] The steps for constructing the avatar feature vector space matrix are as follows: perform feature analysis on the obtained multi-source heterogeneous data of the user avatar. For visual appearance data, extract features such as facial proportion features, hairstyle coding features, and clothing style features; for action and pose data, extract dynamic features such as action speed, angle, and duration; from the personalized setting data, extract features such as expression types and voice pack keywords; summarize these extracted features to form the original feature set of the avatar; use methods such as one-hot encoding and ordinal encoding to perform encoding conversion on the non-numerical features in the original feature set and map them to numerical features. For example, encode different hairstyle styles into different numerical vectors. For numerical features, perform normalization or standardization processing to unify their numerical ranges to obtain the encoded feature vector set; determine the dimensions and elements of the matrix according to the encoded feature vector set; use the number of feature vectors as the number of columns of the matrix and the dimension of each feature vector as the number of rows of the matrix, and arrange each feature vector in turn as the column vector of the matrix to construct the avatar feature vector space matrix; this matrix can comprehensively and systematically represent the various feature information of the user avatar and provide the basic data structure for subsequent numerical representation and analysis.

[0039] The steps for constructing the skill knowledge graph network are as follows: Collect the structured data of users' skills on multiple game platforms from the skill database of the game platform, including information such as skill names, skill levels, skill descriptions, skill effect parameters (such as damage values, healing amounts, durations), skill prerequisite conditions, and skills associated with other skills; Parse the collected skill data, extract key attribute and relationship information, and convert the unstructured skill description text into a structured data format; Use the skill name, skill level, skill effect parameters, etc. as nodes of the knowledge graph, and each node represents a specific skill-related entity; Define the edges between nodes according to the association relationships between skills (such as the relationship between prerequisite skills and subsequent skills, the complementary relationship of skill effects, the combined use relationship of skills, etc.), and the type and weight of the edges are used to represent the nature and strength of the skill relationships; Use a graph database (such as Neo4j) or a graph computing framework (such as GraphX) to store and organize the defined nodes and edges according to a certain topological structure to construct the skill knowledge graph network; During the construction process, perform attribute annotation on the nodes and edges to supplement the detailed information and relationship descriptions of the skills; Through querying, traversing, and analyzing the skill knowledge graph network, the complex relationships between skills can be intuitively displayed, providing a structured knowledge basis for skill feature quantification and recommendation analysis.

[0040] The steps for multi-dimensional quantification of skill features through level mapping, frequency normalization, and attribute weighting are as follows: For skill level data, establish a level mapping function to uniformly map the different skill level systems in different games to a standard numerical interval; For example, set the standard numerical interval to [0, 100]. If the highest level of a skill in a certain game is 10 levels and the current skill level is 5 levels, then the mapped level value is calculated as 50 through the mapping function "mapped value = (current level / highest level) × 100"; Map the levels of all skills in this way to achieve a quantitative representation of skill levels on the same scale; Count the number of times each skill is used by the user within a certain time period as the original data of skill usage frequency; Use a normalization method (such as min-max normalization) to map the original data of skill usage frequency to the [0, 1] interval, and the calculation formula is "normalized frequency = (original frequency - minimum frequency) / (maximum frequency - minimum frequency)"; Through normalization processing, eliminate the influence brought by the magnitude difference of different skill usage frequencies, making the frequency data comparable and facilitating subsequent analysis; According to the different attributes of skills (such as offensive, defensive, auxiliary, etc.) and their importance in the game, assign corresponding weight coefficients to the level mapping value and frequency normalization value of each skill; For example, assign a higher weight to the core offensive skill and a lower weight to the auxiliary secondary skill; Calculate the comprehensive quantification value of each skill through weighted summation. The comprehensive quantification value reflects the characteristics of the skill in multiple dimensions such as level and usage frequency, and realizes the multi-dimensional quantification of skill features.

[0041] The steps for hierarchical emotion annotation based on voice chat records are as follows: Using speech recognition technology (such as an end-to-end speech recognition model based on deep learning), perform real-time or offline conversion on the obtained multi-platform voice chat records of users, converting the voice signals into text content; during the conversion process, perform preprocessing such as noise reduction and enhancement on the voice to improve the accuracy of speech recognition; segment the converted text content into individual sentences according to punctuation marks or semantics; adopt sentiment analysis algorithms (such as a sentiment classification model based on BERT), perform sentiment tendency analysis on each sentence, judge the sentiment category expressed by the sentence (such as positive, negative, neutral, angry, happy, sad, etc.), and label each sentence with the corresponding sentiment label; based on the sentiment labels at the sentence level, perform higher-level emotion annotation; group the sentences according to the conversation topic or time period, and divide the relevant sentences into a conversation unit; perform statistical analysis on the sentiment labels of the sentences within each conversation unit, and use the majority voting method or weighted average method (assign different weights according to the importance of the sentences) to determine the sentiment label of the conversation unit; perform summary analysis on the sentiment labels of multiple conversation units to determine the emotion annotation results at different levels of the entire voice chat record, and form a hierarchical emotion annotation system.

[0042] The steps for identifying the comprehensive emotion index based on label frequency and polarity weight are as follows: For the voice chat records after hierarchical emotion annotation, count the frequencies of each sentiment label (such as positive, negative, neutral, etc.) at different levels (sentence, conversation unit, overall) to form a sentiment label frequency distribution table; for example, in the overall voice chat record, the positive label appears 15 times, the negative label appears 8 times, and the neutral label appears 20 times; assign corresponding polarity weights to different sentiment labels to represent the positive and negative tendencies and intensities of the emotions. For example, set the weight of the positive label to +1, the weight of the negative label to -1, and the weight of the neutral label to 0. For more detailed sentiment labels (such as angry, happy, etc.), further refine the weights according to their emotional intensities, such as the weight of the angry label is -0.8 and the weight of the happy label is +0.7, etc.; according to the sentiment label frequency and polarity weight, calculate the comprehensive emotion index of the voice chat record. This comprehensive emotion index can quantitatively reflect the overall emotional tendency and intensity of the user during the voice chat process, providing a quantitative basis for the subsequent recommendation analysis in terms of emotion dimension.

[0043] The steps for keyword screening by combining the TF-IDF algorithm with a domain dictionary are as follows: Clean the user's multi-platform text chat records to remove irrelevant content such as special characters, punctuation marks, HTML tags, etc.; Convert the text to a unified lowercase letter format for subsequent processing; Use a word segmentation tool (such as Jieba segmentation, NLTK segmentation) to segment the text and split the sentences into individual words; Collect professional terms, common vocabulary, game-specific vocabulary, etc. related to social games to build a domain dictionary; Classify and label the words in the dictionary, such as game character names, game item names, game play terms, etc.; Continuously update and improve the domain dictionary to ensure it covers the latest game-related vocabulary; For each word after segmentation, calculate its term frequency in the current text and the inverse document frequency of the word, multiply the term frequency and the inverse document frequency to obtain the TF-IDF value of the word, and the TF-IDF value reflects the importance of the word in the current text and its distinctiveness in the entire document set; According to the calculated TF-IDF values, combine with the domain dictionary for keyword screening; Set a TF-IDF threshold and use the words with TF-IDF values higher than the threshold as candidate keywords; From the candidate keywords, screen out the words that exist in the domain dictionary as the final text chat keywords; For words with similar TF-IDF values, preferentially select the words that are more important or more representative in the classification of the domain dictionary to ensure that the selected keywords can accurately reflect the core information related to the game in the text chat content.

[0044] The steps to dynamically adjust the weight coefficient according to the scene type to achieve semantic quantization are as follows: Define multiple social game scene types, such as game battle scenes, social interaction scenes, task completion scenes, game store scenes, etc.; Establish a scene type recognition model. By analyzing keywords, context, game event trigger information, etc. in the text chat record, determine the scene type to which the text chat record belongs. For example, when keywords such as "team battle" and "kill" appear in the text, it is recognized as a game battle scene; Preset an initial weight coefficient for the text chat keywords under each scene type. According to the importance and semantic contribution degree of the keywords in different scenes, assign different weight values. For example, in the game battle scene, the initial weight of the keyword "skill release" is set to 0.8, while in the social interaction scene, the weight of this keyword is set to 0.3; Real-time monitor the user's behavior data, changes in text chat content, and scene conversion information in the game; When it is detected that the scene type has changed, according to the characteristics and requirements of the new scene, dynamically adjust the weight coefficient of the keywords. For example, when switching from a game battle scene to a social interaction scene, increase the weight of social-related keywords (such as "chat" and "friend"), and decrease the weight of battle-related keywords. At the same time, according to factors such as the occurrence frequency and context relevance of the keywords in the current scene, further fine-tune the weight coefficient so that the weight coefficient can more accurately reflect the semantic importance of the keywords in a specific scene; According to the adjusted weight coefficient, perform semantic quantization on the text chat keywords; For each keyword, multiply its TF-IDF value by the corresponding weight coefficient to obtain the semantic quantization value of the keyword; Aggregate the semantic quantization values of all keywords, and identify the entire text chat record through weighted summation or other appropriate aggregation methods.

[0045] Through the in-depth mining and analysis of multi-dimensional and multi-source data (avatar features, skill features, emotion tags, and text chat keywords), it is possible to comprehensively and meticulously depict the social behavior patterns and personality preferences of users with low social stress sensitivity. Compared with the analysis methods based on a single dimension or a small amount of data, the accuracy of user portraits is improved, laying a solid foundation for subsequent personalized recommendations. The decision forest model is used to classify and predict the social types of users. The decision forest model integrates the advantages of multiple decision trees, has strong generalization ability and robustness, can effectively handle the noise and complex relationships in the data, thereby improving the accuracy and stability of classification. Compared with a single decision tree or other simple classification models, it can more reliably output the social types of users. Based on the accurate identification of users' social types, the recommendation system can push highly matching social games according to the characteristics and needs of users with different social types, improving users' acceptance and satisfaction of the recommended games, enhancing user stickiness, and at the same time improving the recommendation efficiency and resource utilization efficiency of the game platform. Define multiple social types, covering different behavior tendencies that users may exhibit in social games, which can meet the diverse social needs of users, making the recommended social games not only meet users' preferences in terms of entertainment, but also fit with users' behavior habits in terms of social interaction patterns, promoting the establishment of good social relationships among users in the game.

[0046] Specifically, the specific steps of S3 are as follows: collect the positive review data of the core recommendation parameters of multi-user games corresponding to each social type. The positive review data of the core recommendation parameters is specifically the total number of positive reviews of a certain core recommendation parameter by multiple users. By comparing the total number of positive reviews of each core recommendation parameter of users of each social type, use the preference ranking organization method (PROMETHEE) to assign proportionality coefficients to each core recommendation parameter respectively, and determine the proportionality coefficients of each core recommendation parameter of users of each social type. Identify the proportionality coefficients of each core recommendation parameter of the social type to which the low social stress sensitivity users belong, and at the same time combine the core recommendation parameters of the games to be recommended to generate the initial recommendation values of each game to be recommended.

[0047] The core recommendation parameters specifically include the social interaction complexity parameter, the social role positioning parameter, the game atmosphere style parameter, and the player matching mechanism parameter.

[0048] In this implementation plan, the method for obtaining positive evaluation data of the core recommended parameters is as follows: Set up a dedicated evaluation entry within the game platform to encourage users to evaluate the core recommended parameters of the game. For example, set up an evaluation button on the game details page, community forum, or within the game. Users can score (e.g., from 1 to 5) or provide text evaluations for parameters such as the social interaction complexity and social role positioning of the game. The platform background collects and organizes these evaluation data in real time; it is also possible to regularly or irregularly send out questionnaires about the core recommended parameters of the game to users. The content of the questionnaire can include questions such as users' satisfaction evaluations and importance rankings of the core recommended parameters for different social types of games. Invite users to participate in the survey via email, in-game message push, etc., and conduct statistical analysis on the recovered questionnaire data to obtain positive evaluation data; it is also possible to use natural language processing and sentiment analysis technologies to monitor and analyze the user discussion content in the game community (such as official forums, social media groups, player communities), extract topics and comments related to the core recommended parameters, judge the sentiment tendency (positive, negative, neutral) of users towards each parameter, and count the number of positive evaluations as positive evaluation data; or analyze the in-game behavior data of users, such as game duration, frequency of participating in social activities, number of times of changing game characters, etc., to indirectly infer users' preference levels for the core recommended parameters of the game. For example, if a user frequently participates in high-difficulty social interaction activities and has a long game duration, it can be considered that they have a high evaluation of the social interaction complexity parameter, and positive evaluation data can be statistically obtained based on this.

[0049] The social interaction complexity parameter represents the richness, complexity, and depth of social interactions within the game, including the diversity of social gameplay (such as the quantity and types of gameplay like team dungeons, guild wars, player trading, etc.), the complexity of interaction rules (such as the process of task collaboration, management rules of social relationships), and the real-time and dynamic nature of interactions (such as the smoothness of multiplayer real-time battles, instant chat interactions). Obtain the social gameplay information designed for the game from game development documents and official introduction materials, record the types, frequencies, and durations of social activities actually participated in by players in the game through in-game monitoring tools, and analyze the event data related to social interactions in the game logs, such as the number of team requests and the number of participants in social activities; adopt a hierarchical scoring method to divide the social interaction complexity into multiple levels (such as from 1 to 5 levels). Level 1 indicates simple social interactions, only including basic chat functions, and level 5 indicates very complex social interactions with multiple in-depth social gameplay and complex rules; it is also possible to assign corresponding level scores to each game according to the actual social gameplay and interaction situation of the game; it can also be quantified by calculating the weighted sum of indicators such as the number of social gameplay and the frequency of interaction events, and the weights are determined according to the importance and complexity of the gameplay.

[0050] Social role positioning parameters refer to the different social roles set for players in the game, their functions and responsibilities, which clarify the players' positions and roles in social interactions. For example, in a team game, there are roles such as commanders, attackers, defenders, healers, etc. Each role undertakes specific tasks in the team, which affects team collaboration and social relationships. By referring to the game's character setting documents, understanding the preset character types and function descriptions in the game, observing the actual behaviors and task assignments of players in the game, analyzing the roles played by players, collecting players' feedback and evaluations on the roles, and understanding the rationality and popularity of role positioning; constructing a role positioning evaluation index system, evaluating from dimensions such as role function integrity (such as whether the role skills are rich and the responsibilities are clear), role balance (whether the importance and capabilities of different roles in the team are balanced), and role substitutability (whether it is easily replaced by other roles). Each dimension uses a 0-10 point scoring system, and the quantitative value of the social role positioning parameter is obtained by synthesizing the scores of each dimension; it is also possible to perform weighted calculations based on data such as the usage rate of the role in the game and the positive review rate of players for the role.

[0051] The game atmosphere style parameter reflects the overall emotional atmosphere and style characteristics created by the game, including the game's graphic style (such as realistic, cartoon, magic, ancient style, etc.), music and sound effect style (such as exciting, soothing, tense, cheerful), plot atmosphere (such as relaxed and humorous, suspenseful and tense, touching and inspiring), and social atmosphere (such as friendly and harmonious, highly competitive, mutual assistance and cooperation), etc., which affects players' emotional experiences and social willingness. Analyze the game's graphic style through game screenshots and video materials; collect game music and sound effect files, and let professionals or judge their styles through user evaluations. You can also read the game's plot text, watch the plot animation, evaluate the plot atmosphere, or observe the interaction behaviors and community atmosphere of players in the game to understand the social atmosphere; adopt a method combining classification coding and scoring. First, classify and code the game atmosphere style. For example, the realistic style is coded as 1, the cartoon style is coded as 2, etc.; for each style type, score from multiple sub-dimensions (such as graphic beauty, music suitability, plot attractiveness, social activity) on a scale of 0-10. The weighted sum of the coded value and the scores of each sub-dimension is used as the quantitative value of the game atmosphere style parameter, and the weights are determined according to the importance of each sub-dimension.

[0052] Player matching mechanism parameters refer to the rules and algorithms used by the game to match players for social interaction or battles, including matching criteria (such as level, skill level, playing time, hobbies, etc.), matching speed (the time from initiating a matching request to successful matching), matching fairness (whether the strength of the opponents or teammates matched is balanced), etc. These directly affect the player's gaming experience and social effects. Obtain the technical documentation of the matching mechanism from the game development team to understand the principles and rules of the matching algorithm. Conduct multiple matching experiments in the game, record the matching time, information of the opponents or teammates matched, collect the feedback from players on the matching results, and analyze the fairness and satisfaction of the matching. Establish evaluation indicators for the matching mechanism, and evaluate from dimensions such as the rationality of the matching criteria (such as whether the matching criteria can effectively guarantee the gaming experience), matching speed (quantified by the average matching time), and matching fairness (calculate the fairness index by comparing the strength data of both sides of the match, such as level difference, win rate difference, etc.). Each dimension uses a standardized score (such as 0 - 1), and the quantitative value of the player matching mechanism parameters is obtained by integrating the scores of each dimension.

[0053] Let the set of core recommendation parameters be , corresponding to the social interaction complexity parameter, social role positioning parameter, game atmosphere style parameter, and player matching mechanism parameter respectively, and construct the preference graph matrix indicating that when parameter is compared with parameter , the importance score of relative to . If is more important than , then , otherwise , and . Calculate the total score of each parameter, then the proportionality coefficient of parameter

[0054] Initial recommendation value calculation: Let the core recommendation parameter vector of the game to be recommended be , corresponding to the parameter values of social interaction complexity, social role positioning, game atmosphere style, and player matching mechanism respectively. The proportionality coefficient vector of the core recommendation parameters of the social type to which low - social - stress - sensitivity users belong is , then the initial recommendation value of game .

[0055] By collecting the positive review data of the core recommended parameters corresponding to each social type in the game, it is possible to understand the preferences of users of different social types for the core elements of the game. For example, leader-type users may place more emphasis on the complexity of social interaction and the positioning of social roles. The proportionality coefficient determined based on the positive review data enables the recommendation system to accurately match the user's needs, improving the fit and user satisfaction of the recommendation. Using the analytic hierarchy process to determine the proportionality coefficient of the core recommended parameters can systematically quantify the importance of each parameter to users of different social types, helping the recommendation system to prioritize the parameters that are more important to users when recommending games, reducing the interference of irrelevant parameters, thereby improving the quality and efficiency of the recommendation and making the recommendation results more targeted. Providing game recommendations that suit the social type for users with low social stress sensitivity can meet the personalized needs of users in social games. When users frequently receive game recommendations that match their preferences, it increases their trust and dependence on the recommendation system and the game platform, thereby enhancing user stickiness and promoting the long-term development of the game platform. Based on the preferences of users of each social type for the core recommended parameters, the game platform can optimize the game resource allocation accordingly; to meet the needs of auxiliary-type users, it focuses on optimizing the game atmosphere style and the player matching mechanism to achieve efficient use of resources.

[0056] Specifically, the specific steps of S5 are as follows: During the game process of users with low social stress sensitivity, game data in the game is obtained in real time. The game data includes the participation popularity of activities published in the game and the interaction frequency of players in the community. The user emotion data is obtained in real time using emotion recognition technology. The user emotion data includes user pleasure, user concentration, and user stress level. An adjustment and optimization model is constructed using the game data and the user emotion data, and then the adjustment and optimization values of each game are output. A fusion coefficient is obtained, and a comprehensive recommendation value is generated by combining the initial recommendation value, the adjustment and optimization value, and the fusion coefficient, thereby realizing the reordering of game recommendations.

[0057] In this implementation plan, the participation popularity of activities published in the game represents the degree of attention and participation of players in the activities published in the game, reflecting the popularity and influence of the activities among the player group, and is one of the important indicators to measure the activity and attractiveness of the game. By recording data such as the number of participants, the number of participation times, and the completion progress of each activity in the game server background, the discussion popularity related to the activities can also be monitored, such as the number of posts, comments, and likes about the activities in the game community, or the change in the online duration and login frequency of players during the activity period, indirectly reflecting the participation popularity of the activities. The comprehensive scoring method is adopted, setting a basic participation popularity score (such as taking the number of participants as the main basis, and counting 1 point for every 100 people participating), and then adding weighted scores according to factors such as the activity completion progress and community discussion popularity.

[0058] The interaction frequency of players in the community refers to the degree of frequent communication and interaction among players in the game community (such as official forums, in-game chat channels, player groups, etc.), which reflects the activity of the game community and the social atmosphere among players. By using the log recording system of the game community, data such as the number of messages sent by players within a certain period of time, the number of discussions initiated, the number of responses to others' messages, and the frequency of participation in group activities are statistically analyzed; the number and frequency of private messages exchanged among players in the community are monitored; the various interaction behaviors of players within a unit of time (such as 1 hour, 1 day) are weighted and summed up. For example, sending an ordinary message is counted as 1 point, initiating a discussion is counted as 3 points, and responding to others' messages is counted as 2 points. By setting different weights, a quantitative value of the player interaction frequency is calculated.

[0059] User pleasure is used to measure the degree of positive emotions such as pleasure and happiness that users feel during the game process, and it is a key indicator for evaluating user satisfaction with the game experience. Physiological data of users (such as heart rate, skin conductance response, facial expressions) are collected by using wearable devices (such as smart bracelets, smart glasses), and the user's pleasure is inferred through physiological signal analysis algorithms. Real-time feedback buttons can also be set in the game client to allow users to actively rate their degree of pleasure, or analyze the voice intonation and chat content of users in the game, and judge their pleasure emotions through sentiment analysis techniques; the pleasure is divided into multiple levels (such as 0 - 10 points), where 0 points means completely unhappy and 10 points means extremely happy. According to the analysis results of physiological data, the results of users' active ratings or sentiment analysis, they are mapped to the corresponding level scores to achieve quantification.

[0060] User concentration reflects the degree of concentration of users' attention during the game process, which reflects the degree of investment and interest of users in the game content. By monitoring the operation behaviors of users' devices (such as mouse click frequency, keyboard input frequency, screen switching times), the operation rhythm and coherence of users are analyzed to judge the concentration. Eye tracking technology (such as mobile phone cameras or external eye trackers) can also be used to record the user's line of sight focus and fixation duration, evaluate the user's attention to the game screen, or analyze the user's stay time and task completion efficiency in the game to indirectly reflect the concentration; a concentration evaluation model is constructed, and indicators such as operation behaviors, eye movement data, and stay time are weighted and calculated. For example, the weights of mouse click frequency and keyboard input frequency are 0.3, the weight of eye movement fixation duration is 0.4, and the weight of stay time is 0.3. According to the quantitative values and weights of each indicator, a quantitative score of the user concentration is calculated.

[0061] The user stress level represents the magnitude of the psychological stress borne by the user during the game, which is caused by factors such as game difficulty, competition pressure, and task urgency, and affects the user's gaming experience and willingness to continue participating. By using wearable devices to collect the user's physiological indicators (such as heart rate variability, respiratory rate), the user's stress level is evaluated through a physiological stress detection algorithm. It is also possible to analyze the user's voice emotions (such as rapid tone, anxious vocabulary) and negative emotion expressions in the chat content during the game, or set up a stress self-assessment questionnaire in the game and regularly invite the user to subjectively evaluate their own stress level. The stress level is divided into different grades (such as 0 - 10 points), where 0 points indicate no stress and 10 points indicate extremely high stress. Based on the analysis results of physiological indicators, voice emotion analysis, and user self-assessment results, the corresponding quantitative scores can be determined.

[0062] The steps to construct an adjustment and optimization model are as follows: Clean the in-game data (activity participation popularity, player interaction frequency) and user emotion data (pleasure level, concentration level, stress level) obtained in real-time, removing outliers and missing values. Using normalization or standardization methods, map data of different types and magnitudes to the same numerical range to ensure data consistency and comparability, forming a preprocessed dataset. Extract key features from the preprocessed dataset. For example, for activity participation popularity data, extract features such as activity type, growth rate of the number of participants, etc.; for user emotion data, extract features such as emotion change trend, emotion fluctuation range, etc. Through feature selection algorithms (such as chi-square test, mutual information method), select features that have a significant impact on game recommendation adjustment and construct a feature vector. Select a machine learning model (such as neural network, random forest, support vector regression) as the basic framework of the adjustment and optimization model. Use the feature vector and the corresponding game recommendation adjustment target (such as adjusting the recommendation order, increasing or decreasing the recommendation weight) as training data, and use the training data to train the model. Adjust the parameters of the model through optimization algorithms (such as stochastic gradient descent, Adam optimization algorithm) so that the model can learn the relationship between in-game data, user emotion data, and game recommendation adjustment. Use the reserved validation dataset to evaluate the trained adjustment and optimization model. Use evaluation metrics such as mean squared error (MSE), mean absolute error (MAE), etc. to measure the prediction accuracy of the model. According to the evaluation results, adjust the hyperparameters of the model (such as the number of layers and nodes of the neural network, the number of trees in the random forest) or reselect features, and repeat the training and evaluation process until the model reaches satisfactory performance indicators, and finally complete the construction of the adjustment and optimization model. The expression of the adjustment and optimization model is: Let the in-game data vector be , corresponding to activity participation popularity and community player interaction frequency respectively; the user emotion data vector be , corresponding to user pleasure level, user concentration level, and user stress level respectively. The adjustment and optimization model is: , where is a weight coefficient, which is set through training or experience. The comprehensive recommendation value is calculated as: for the game 's comprehensive recommendation value , where is a fusion coefficient, which is used to balance the importance of the initial recommendation value and the adjusted and optimized value. .

[0063] The way to obtain the fusion coefficient is as follows: According to the experience of the game industry and the historical operation data of the recommendation system, a domain expert or developer manually sets an initial fusion coefficient as a reference value for the initial operation of the recommendation system. For example, at the initial stage of system development, the fusion coefficient is set to 0.6, indicating that the initial recommendation value accounts for 60% in the comprehensive recommendation value, and the adjusted and optimized value accounts for 40%; A real-time monitoring mechanism can also be established to track the feedback data of users on the recommended games (such as game download volume, game duration, user evaluation) and the performance indicators of the recommendation system (such as recommendation accuracy, recall rate). According to the changes in the feedback and performance indicators, an adaptive algorithm (such as reinforcement learning algorithm, genetic algorithm) is used to dynamically adjust the fusion coefficient. For example, when it is found that the recommendation accuracy decreases, the weight of the adjusted and optimized value is appropriately increased, and the weight of the initial recommendation value is reduced. Through continuous trial and error and learning, the optimal fusion coefficient is found; A custom option for the fusion coefficient can also be provided in the user settings interface of the recommendation system, allowing users to manually adjust the proportion of the initial recommendation value and the adjusted and optimized value in the comprehensive recommendation value according to their own preferences and needs. For example, users can set the fusion coefficient to 0.8, being more inclined to the initial recommendation value, or set it to 0.2, attaching more importance to the adjusted and optimized value based on real-time data.

[0064] By obtaining in-game data and user sentiment data in real time, it is possible to timely capture the behavioral changes and emotional feedback of users during the game process. Based on this dynamic information, an adjustment and optimization model is constructed to correct the initial recommendation value, realizing the dynamic update of game recommendations, making the recommendation results more in line with the current needs and experiences of users, and improving the accuracy of recommendations; Pay attention to the emotional state of users (pleasure, concentration, stress level). When users have negative emotions or their interest in the current game decreases during the game, the system can adjust the recommendation in a timely manner according to the adjusted and optimized value, providing more suitable games for users, avoiding the loss of users due to bad experiences, and effectively improving the overall experience of users during the game process; Based on data such as the participation heat of in-game activities and the interaction frequency of community players, the recommendation system can understand the real-time heat and activity of the game, tilt more resources towards popular and suitable games for users, improve the utilization efficiency of game resources, and at the same time help game developers and platforms understand user needs and optimize game operation strategies.

[0065] Please refer to Figure 2, an artificial intelligence-based social game recommendation system that applies the above artificial intelligence-based social game recommendation method, including: a user type recognition module for identifying users with low social stress sensitivity based on the social avoidance index in the user's historical behavior data; a social type recognition module for obtaining the virtual image features and skill features of users with low social stress sensitivity on multiple game platforms, and identifying the emotional tags and text chat keywords in the multiple game platforms of users with low social stress sensitivity, and then outputting the social type of users with low social stress sensitivity; an initial recommendation analysis module for extracting the core recommendation parameters of the games to be recommended, and then obtaining the initial recommendation values of each game to be recommended in combination with the social type of users with low social stress sensitivity; a real-time monitoring module for sorting the recommended games according to the magnitude of the initial recommendation values, and simultaneously monitoring the in-game data and user emotion data during the game process of users with low social stress sensitivity in real time; an optimization and adjustment module for outputting the adjustment and optimization values of each game based on the in-game data and emotion data, generating the comprehensive recommendation values of each game by combining the initial recommendation values and the adjustment and optimization values, re-sorting the recommended games according to the comprehensive recommendation values, and continuously updating the recommended social games for the user.

[0066] In summary, the present application has at least the following effects: accurately identifying users with low social stress sensitivity through the social avoidance index in the user's historical behavior, avoiding misjudgment of highly sensitive users, and reducing ineffective recommendations; comprehensively depicting the user's social preferences by combining virtual image features, skill features, emotional tags, and text chat keywords, and outputting a more accurate social type; extracting the core recommendation parameters of the games to be recommended, and calculating the initial recommendation values in combination with the user's social type to ensure a high degree of matching between the recommended games and the user's preferences; dynamically correcting the recommendation values by calculating the adjustment and optimization values in real time through the in-game data and user emotion data to avoid user churn due to changes in the game experience; avoiding anxiety and frustration caused by forced social interaction for users by accurately identifying users with low social stress sensitivity and recommending single-player tasks or mildly social games, and improving the game experience; forming a technical barrier by combining multi-modal data such as virtual images, skills, emotions, and texts, and through real-time monitoring and dynamic adjustment, being different from traditional recommendation systems based on single behavioral data, and improving user satisfaction and retention rate.

[0067] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods and systems. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0068] The present invention is described with reference to the flowcharts and block diagrams of methods and systems according to embodiments of the present invention. It should be understood that each flow and combination of blocks in the flowcharts and block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and block diagrams Figure 1 or multiple blocks.

[0069] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and block diagrams Figure 1 or multiple blocks.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and block diagrams Figure 1 or multiple blocks.

[0071] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0072] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A social game recommendation method based on artificial intelligence, characterized in that, It includes the following steps: S1. Identify users with low social stress sensitivity based on the social avoidance indicators in the user's historical behavior data; S2. Obtain the virtual image features and skill features of users with low social stress sensitivity on multiple game platforms, identify the emotion tags and text chat keywords in the multiple game platforms of users with low social stress sensitivity, and then output the social types of users with low social stress sensitivity; S3. Extract the core recommendation parameters of the games to be recommended, and then combine the social types of users with low social stress sensitivity to obtain the initial recommendation values of each game to be recommended; S4. Sort the games for recommendation according to the magnitude of the initial recommendation values, and simultaneously monitor the in-game data and user emotion data during the game process of users with low social stress sensitivity in real time; S5. Output the adjustment and optimization values of each game based on the in-game data and emotion data, combine the initial recommendation values and the adjustment and optimization values to generate the comprehensive recommendation values of each game, re-sort the games for recommendation according to the comprehensive recommendation values, and continuously update the recommended social games for users.

2. The social game recommendation method based on artificial intelligence according to claim 1, wherein The specific steps of S1 are as follows: Extract the social avoidance indicators, and the social avoidance indicators include the friend number increase rate, social interaction frequency, chat silence ratio, and team rejection times; Take the social avoidance indicators as the input, use the support vector machine for training, construct a user social stress sensitivity model, and output the social stress sensitivity value of the user; Compare the social stress sensitivity value with the sensitivity threshold. When the social stress sensitivity value is greater than or equal to the sensitivity threshold, mark the user as a user with high social stress sensitivity, enter the single-player task game recommendation branch, and recommend single-player task-dominated games; When the social stress sensitivity value is less than the sensitivity threshold, mark the user as a user with low social stress sensitivity, and trigger the social multi-player task game recommendation requirement.

3. The social game recommendation method based on artificial intelligence according to claim 1, characterized in that, The specific steps of the outputting the social types of users with low social stress sensitivity are as follows: Obtain the virtual image features, skill features, emotion tags, and text chat keywords of known social types as training samples, and then use the training samples to train the decision tree algorithm to construct a decision forest model composed of multiple decision trees; Take the virtual image features, skill features, emotion tags, and text chat keywords of users with low social stress sensitivity as the input of the decision forest model, and output the social types of users with low social stress sensitivity. The social types include the leadership type, the auxiliary type, the observation type, and the following type.

4. The social game recommendation method based on artificial intelligence according to claim 1, characterized in that The acquisition and analysis of the virtual image features are as follows: Obtain the multi-source heterogeneous data of the user's virtual image, construct a virtual image feature vector space matrix, and numerically represent the virtual image by feature coding and attribute weighting; The acquisition and analysis of the skill features are as follows: Obtain the skill structured data, extract the skill text features, construct a skill knowledge graph network, and realize the multi-dimensional quantization of the skill features through level mapping, frequency normalization, and attribute weighting; The acquisition and analysis of the emotion features are as follows: Obtain the user's multi-platform voice chat records based on user authorization, perform hierarchical emotion annotation based on the voice chat records, and identify the comprehensive emotion index based on the label frequency and polarity weight; The analysis of obtaining text chat keywords is as follows: keyword screening is performed through the TF-IDF algorithm combined with a domain dictionary, and semantic quantification is achieved by dynamically adjusting the weight coefficient according to the scenario type.

5. A method for recommending social games based on artificial intelligence according to claim 1, characterized in that, The specific steps of S3 are: collect the core recommendation parameter positive review data corresponding to each social type for multi-user games, and the core recommendation parameter positive review data is specifically the total number of positive reviews of multi-users for a certain core recommendation parameter; By comparing the total number of positive reviews of each core recommendation parameter of users of each social type, the proportionality coefficient of each core recommendation parameter is assigned respectively using the preference ranking organization method (PROMETHEE) to determine the proportionality coefficient of each core recommendation parameter of users of each social type; Identify the proportionality coefficient of each core recommendation parameter of the social type to which the low social stress sensitivity users belong, and at the same time, combine the core recommendation parameters of the game to be recommended to generate the initial recommendation value of each game to be recommended respectively.

6. The social game recommendation method based on artificial intelligence according to claim 5, wherein The core recommendation parameters specifically include social interaction complexity parameters, social role positioning parameters, game atmosphere style parameters, and player matching mechanism parameters.

7. A social game recommendation method based on artificial intelligence according to claim 1, characterized in that The specific steps of S5 are: during the game process of low social stress sensitivity users, obtain in-game data in real time, and the in-game data includes the participation popularity of activities released in the game and the interaction frequency of players in the community; Use emotion recognition technology to obtain user emotion data in real time, and the user emotion data includes user pleasure, user concentration, and user stress level; Construct an adjustment and optimization model using in-game data and user emotion data, and then output the adjustment and optimization value of each game; Obtain a fusion coefficient, combine the initial recommendation value, the adjustment and optimization value, and the fusion coefficient to generate a comprehensive recommendation value, and then realize the re-ranking of game recommendations.

8. A social game recommendation system based on artificial intelligence, which applies a social game recommendation method based on artificial intelligence according to any one of claims 1-7, characterized in that Including: A user type recognition module for identifying low social stress sensitivity users based on the social avoidance index in the user's historical behavior data; A social type recognition module for obtaining the virtual image characteristics and skill characteristics of low social stress sensitivity users on multiple game platforms, and identifying the emotion tags and text chat keywords on multiple game platforms of low social stress sensitivity users, and then outputting the social type of low social stress sensitivity users; An initial recommendation analysis module for extracting the core recommendation parameters of the game to be recommended, and then obtaining the initial recommendation value of each game to be recommended in combination with the social type of low social stress sensitivity users; A real-time monitoring module for recommending and ranking games according to the size of the initial recommendation value, and at the same time, real-time monitoring the in-game data and user emotion data during the game process of low social stress sensitivity users; An optimization and adjustment module for outputting the adjustment and optimization value of each game based on the in-game data and emotion data, combining the initial recommendation value and the adjustment and optimization value to generate the comprehensive recommendation value of each game, re-recommending and ranking the games according to the comprehensive recommendation value, and continuously updating the recommended social games for users.

Images