Game information recommendation method, device, equipment and storage medium

Abstract

This application belongs to the field of computer technology and discloses a method, apparatus, device, and storage medium for recommending game information. The method comprises: obtaining user behavior data and game reference data for each game; analyzing the game reference data for each game to obtain a game feature vector for each game; analyzing the user behavior data to obtain a user preference vector; matching the user preference vector with each game feature vector to obtain a plurality of recommended games to be selected; reordering each recommended game to be selected based on the game reference data to obtain a recommended game list; and generating a game recommendation interface for the user based on the recommended game list and interface recommendation areas of different sizes. This application can improve the accuracy of determining user preference vectors and game feature vectors and the accuracy of the recommended games to be selected.

Description

Technical Field

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

[0002] Current game recommendation algorithms usually match similar gameplay based on user operation interaction data and make recommendations. In the process of similarity matching, they often rely on tags manually uploaded by game developers. However, these tags are often set when the game is launched. Some games are constantly updated and their gameplay is upgraded, and their tags often cannot summarize the comprehensive nature of the game. In addition, some features that users are more concerned about cannot be summarized by tags, such as the quality of the game plot, the game modeling style, the intensity of the game's sociality, etc. Therefore, the technical problem solved by this application is that the existing tag-based game recommendation algorithm is limited by the untimely update of tags and description limitations, and the recommendation accuracy is poor and the quality is uneven. Summary of the Invention

[0003] The present application provides a game information recommendation method, apparatus, device, and storage medium, which can improve the accuracy of determining user preference vectors and game feature vectors and the accuracy of recommended games.

[0004] In a first aspect, an embodiment of the present application provides a method for recommending game information, comprising:

[0005] Obtain user behavior data and game reference data of each game;

[0006] Among them, user behavior data includes users' subjective evaluation data, login and play data, consumption data, and game social data; game reference data includes game application evaluation data, game tags, and social comment data;

[0007] Analyze the game reference data of each game to obtain the game feature vector of each game;

[0008] Analyze user behavior data to obtain user preference vectors;

[0009] Match the user preference vector with each game feature vector to obtain multiple recommended games;

[0010] Reorder each recommended game to be selected according to the game reference data to obtain a recommended game list;

[0011] Generate a game recommendation interface for users based on the recommended game list and interface recommendation areas of different sizes.

[0012] Furthermore, subjective evaluation data includes game evaluation data published by users on application download platforms; game social data includes media data with game-related tags published by users on social platforms and interaction data published in game chat channels.

[0013] Furthermore, the application evaluation data includes the evaluation text, number of ratings and comprehensive ratings of the game on various application download platforms; the social comment data includes media data released by the game-related tags of the game on at least one social platform within a preset time period.

[0014] Furthermore, the game reference data of each game is analyzed to obtain the game feature vector of each game, including:

[0015] Perform text analysis on app evaluation data and social comment data to obtain multiple keyword vectors;

[0016] Cluster each keyword vector to obtain multiple feature clusters and the core vector of each feature cluster;

[0017] Perform sentiment analysis on each keyword vector to obtain the corresponding sentiment score;

[0018] Calculate the sentiment score of each feature cluster based on the sentiment score of each keyword vector;

[0019] The core vector of the feature cluster with a sentiment score greater than the preset positive threshold is used as the target vector;

[0020] Combine the target vector and the game label to get the game feature vector.

[0021] Furthermore, the sentiment analysis of each keyword vector is performed to obtain the corresponding sentiment direction, including:

[0022] Obtain user evaluation data and user rating data from the application download platform;

[0023] The TextCNN model is trained based on user evaluation data and user rating data to obtain a sentiment labeling model;

[0024] Get a general sentiment dictionary and a game-specific dictionary;

[0025] Score the keyword vectors according to the general sentiment dictionary and the game-specific dictionary to obtain the dictionary score;

[0026] Input the keyword vector into the sentiment tagging model to obtain the model score;

[0027] The dictionary score and model score are weighted to obtain the sentiment score of the keyword vector.

[0028] Furthermore, the above analysis of user behavior data yields a user preference vector, including:

[0029] The subjective preference vector is extracted based on the subjective evaluation data and social behavior data;

[0030] Determine behavioral preference vectors based on login, play, and consumption data;

[0031] The subjective preference vector and the behavioral preference vector are combined to obtain the user preference vector.

[0032] Furthermore, the subjective preference vector extracted from the subjective evaluation data and social behavior data includes:

[0033] Use the BERT model to extract at least one keyword vector from the subjective evaluation data;

[0034] Determine whether the keyword vector includes the preset visual vocabulary;

[0035] If so, ResNet50 is used to extract the game style features corresponding to the subjective evaluation data;

[0036] The painting style features and keyword vectors are used as subjective preference vectors.

[0037] Furthermore, the login and play data includes the number of times the user logs in to each downloaded game within a preset time period and the game duration of each login; the consumption data includes the cumulative recharge amount of the user for each downloaded game within a preset time period.

[0038] Furthermore, the above-mentioned determination of the behavior preference vector based on login, play, and consumption data includes:

[0039] Update login and play data according to the preset time decay factor;

[0040] The updated login and play data, consumption data, and the corresponding game feature vector are input into the Transformer model to obtain the behavioral preference score for each feature data in the game feature vector;

[0041] Sum the behavioral preference scores of the same feature data in each game feature vector to obtain the feature preference score;

[0042] Sort each feature data according to the feature preference score;

[0043] A preset number of feature data in the first column of the sort is used as the behavior preference vector.

[0044] Furthermore, the method further comprises:

[0045] Get the latest update log of the game downloaded by the user;

[0046] Determine whether there is a limited-time activity within the preset time period based on the latest update log;

[0047] If yes, then obtain the average game revenue data and the second revenue data within a preset time period;

[0048] The activity attenuation factor is calculated based on the average flow data and the second flow data;

[0049] Update consumption data based on activity decay factors.

[0050] Furthermore, the method further comprises:

[0051] Collect the games downloaded by users on the game recommendation interface according to a preset period;

[0052] Obtain the user's cumulative playing time of the downloaded game within a preset period;

[0053] Compare the download time and cumulative playing time of users to get the conversion rate;

[0054] Determine downloaded games with conversion rates higher than a preset conversion threshold and corresponding target game feature vectors;

[0055] Update the user preference vector based on the target game feature vector.

[0056] Furthermore, the method further comprises:

[0057] Real-time monitoring of the log update date of each game;

[0058] Update the game reference data of the game within a preset time after the game log is updated;

[0059] Update the game feature vector of the game according to the updated game reference data.

[0060] Furthermore, the login and play data also includes the historical IP address of each login; the method also includes:

[0061] Get the current IP address and filter out abnormal login records whose historical IP addresses are not equal to the current IP address;

[0062] The number of logins and the game duration of each login are filtered and eliminated based on abnormal login records.

[0063] Furthermore, the aforementioned recommended games are reordered according to the game reference data to obtain a recommended game list, including:

[0064] Obtain the number of ratings and comprehensive ratings of the recommended games on any app download platform;

[0065] Calculate the credibility rating of the recommended games based on the number of ratings and the comprehensive rating;

[0066] Determine the corresponding trust weight based on the average daily active users of each app download platform;

[0067] Weight the trustworthiness ratings of the recommended games on various app download platforms based on the trustworthiness weights;

[0068] The recommended games are reordered according to the weighted total credibility score to obtain a recommended game list.

[0069] In a second aspect, an embodiment of the present application provides a game information recommendation device, comprising:

[0070] The acquisition module is used to obtain the user behavior data of users and the game reference data of each game;

[0071] Among them, user behavior data includes users' subjective evaluation data, login and play data, consumption data, and game social data; game reference data includes game application evaluation data, game tags, and social comment data;

[0072] A game analysis module is used to analyze the game reference data of each game and obtain the game feature vector of each game;

[0073] User analysis module, used to analyze user behavior data and obtain user preference vectors;

[0074] A matching module is used to match the user preference vector with each game feature vector to obtain multiple recommended games;

[0075] A reordering module is used to reorder each recommended game to be selected according to the game reference data to obtain a recommended game list;

[0076] The recommendation module is used to generate a game recommendation interface for users based on the recommended game list and interface recommendation areas of different sizes.

[0077] Furthermore, the game analysis module includes:

[0078] A text analysis unit, used to perform text analysis on application evaluation data and social comment data to obtain multiple keyword vectors;

[0079] The clustering unit is used to cluster each keyword vector to obtain multiple feature clusters and the core vector of each feature cluster;

[0080] Sentiment analysis unit, used to perform sentiment analysis on each keyword vector and obtain the corresponding sentiment score;

[0081] The score unit is used to calculate the sentiment score of each feature cluster based on the sentiment score of each keyword vector;

[0082] A target unit is used to take the core vector of the feature cluster whose sentiment score is greater than a preset positive threshold as the target vector;

[0083] The combining unit is used to combine the target vector and the game label to obtain the game feature vector.

[0084] Furthermore, the sentiment analysis unit is used to:

[0085] Obtain user evaluation data and user rating data from the application download platform;

[0086] The TextCNN model is trained based on user evaluation data and user rating data to obtain a sentiment labeling model;

[0087] Get a general sentiment dictionary and a game-specific dictionary;

[0088] Score the keyword vectors according to the general sentiment dictionary and the game-specific dictionary to obtain the dictionary score;

[0089] Input the keyword vector into the sentiment tagging model to obtain the model score;

[0090] The dictionary score and model score are weighted to obtain the sentiment score of the keyword vector.

[0091] Furthermore, the user analysis module includes:

[0092] A subjective preference unit is used to extract a subjective preference vector based on subjective evaluation data and social behavior data;

[0093] A behavior preference unit, configured to determine a behavior preference vector based on login and play data and consumption data;

[0094] The user combination unit is used to combine the subjective preference vector and the behavioral preference vector to obtain the user preference vector.

[0095] In a third aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the processor executes the steps of a method for recommending game information as in any of the above embodiments.

[0096] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a game information recommendation method as in any of the above embodiments.

[0097] In summary, compared with the prior art, the technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0098] The embodiment of the present application provides a game information recommendation method. First, in the data collection stage, multi-source data such as social data and evaluation data are added to determine user preference vectors and game feature vectors, thereby improving the accuracy of determining user preference vectors and game feature vectors and the accuracy of the recommended games. At the same time, the application evaluation data and social comment data in the game reference data can also reflect the quality of the games, thereby re-ranking the recommended games to prevent low-quality games from affecting the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 A flowchart of a method for recommending game information is provided as an exemplary embodiment of the present application.

[0100] Figure 2 A flowchart of the game feature vector extraction steps provided for an exemplary embodiment of the present application.

[0101] Figure 3 A flowchart of the keyword vector sentiment analysis steps provided for an exemplary embodiment of the present application.

[0102] Figure 4 A structural diagram of a game information recommendation method and device provided as an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0103] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0104] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0105] See Figure 1 , an embodiment of the present application provides a game information recommendation method, comprising:

[0106] Step S1: Obtain user behavior data of users and game reference data of various games.

[0107] Among them, user behavior data includes users' subjective evaluation data, login and play data, consumption data and game social data; game reference data includes game application evaluation data, game tags and social comment data.

[0108] Furthermore, the subjective evaluation data includes game evaluation data published by users on the application download platform.

[0109] App review data includes the game’s review text, number of reviewers, and overall rating on various app download platforms.

[0110] Generally speaking, platforms that provide APP and application software download and installation will also open comment and rating functions to every user who downloads the application software. In the game field, many players will express their opinions and views on the game through this channel. These evaluations can not only reflect the current user's focus, but also reflect the quality of the game.

[0111] Game social data includes media data posted by users on social platforms with game-related tags and interaction data posted in game chat channels. Social commentary data includes media data posted within a preset time period using game-related tags for at least one game on a social platform. It is understandable that most current social platforms have a tagging function that allows users to search for media data related to a specific topic. Similarly, gamers often add game-related tags when posting game-related content, such as praise for the plot, demonstrations of the exquisiteness of costumes or skins, and game walkthroughs. These tags are generally the game name, followed by a nickname or abbreviation. Other gamers can then search for these tags to access this game content and engage in discussion and communication.

[0112] It should be noted that the discussion data of game-related content posted on social platforms will be far greater than the evaluation of game application software by different users on application download platforms. This is because the number of words in social platforms' comments, the content that can be evaluated and posted, and even the restrictions on blocked words are more relaxed than those on application download platforms. At the same time, social platforms are oriented towards a wide range of user groups and have a strong downward trend. Therefore, the media data on social platforms can reflect the discussions and attitudes of a large number of players towards the different features of a certain game (such as plot, style, fashion, and gameplay, etc.). This makes these media data play a great reference role both in determining user preferences and in the process of extracting game features, greatly improving the accuracy of determining game features and user preferences.

[0113] Step S2: Analyze the game reference data of each game to obtain the game feature vector of each game.

[0114] Step S3: Analyze the user behavior data to obtain the user preference vector.

[0115] Step S4: Match the user preference vector with each game feature vector to obtain multiple recommended games.

[0116] Specifically, by calculating the similarity between the user preference vector and each game feature vector, the games corresponding to the game feature vectors whose similarity is greater than a preset similarity threshold can be selected as recommended games.

[0117] The similarity calculation may adopt methods such as cosine similarity calculation and Euclidean distance calculation.

[0118] Step S5: reorder the recommended games to be selected according to the game reference data to obtain a recommended game list.

[0119] The above-mentioned process of reordering the recommended games to be selected based on the game reference data to obtain a recommended game list includes:

[0120] Step S51: Obtain the number of reviewers and the comprehensive review scores of the recommended game on any application download platform.

[0121] Among them, the number of raters refers to the number of people who have rated the recommended game on a certain application download platform. When counting, users with the same ID need to be merged to ensure the authenticity of the number of raters and avoid cheating.

[0122] The comprehensive rating is the game rating obtained by a certain application download platform based on the ratings of various users.

[0123] The above two data can be obtained through the application evaluation data in the game reference data.

[0124] Furthermore, the comprehensive score can filter out user ratings from a period of time (eg, 1 year or 3 years ago), and use the rating calculation algorithm of the application download platform for the remaining user ratings to obtain the latest comprehensive score.

[0125] It is understandable that changes in the game development team or internal changes in the game company may lead to changes in the development focus of a certain game or even the player's reputation; for example, the plot copy at the time of the game's launch was criticized for being simple and meaningless. After subsequent updates by the copywriting team, the plot has been getting better and better, which has led to higher corresponding ratings from players who like the plot; therefore, this application can obtain the latest period of rating data when obtaining the comprehensive rating to ensure the credibility and reliability of the rating.

[0126] Step S52: Calculate the credibility score of the recommended game based on the number of scorers and the comprehensive score.

[0127] Specifically, the calculation formula for the trust score is as follows:

[0128]

[0129] in, is the number of raters; It is a comprehensive rating; is the minimum credible sample threshold, the default value is 100; C is the average comprehensive score of each game application in this application download platform.

[0130] Step S53: Determine the corresponding trust weight according to the average daily active users of each application download platform.

[0131] Specifically, the average daily active users of a single app download platform divided by the sum of the average daily active users of all app download platforms is the trust weight of the app download platform.

[0132] Step S54: weighting the credibility scores of the recommended games on various application download platforms according to the credibility weights.

[0133] It should be noted that this application does not directly refer to the comprehensive score of the application download platform, but calculates the trustworthy score and trustworthy weight separately. This is because the scores given by users of the application download platform are not reliable in many cases. For example, some game developers will buy accounts and brush scores to attract players in order to improve the comprehensive score. Some games with a large number of active users may cause players to be dissatisfied with the low scores due to individual decision-making errors, but this does not affect the high playability of the game. Therefore, the re-ranking algorithm of this application can more accurately block the impact of brushing scores through the calculation of trustworthy scores and trustworthy weights, and extract high-quality recommended games for selection.

[0134] In step S55, the recommended games are reordered according to the weighted total credibility scores to obtain a recommended game list, wherein the reordering is to sort the recommended games from largest to smallest according to the weighted credibility scores.

[0135] Step S6: Generate a game recommendation interface for the user based on the recommended game list and interface recommendation areas of different sizes.

[0136] Specifically, in the game simulator interface or game application download platform that the user logs in to, there are often areas of different sizes used to display recommended game information. For example, the upper middle part of the entire page will serve as the main display area, cyclically displaying the information of each recommended game to be selected in the front row of the recommended game list; and the lower middle part of the page is divided into multiple smaller areas, which fixedly display the information of each recommended game to be selected remaining in the recommended game list.

[0137] Furthermore, before re-ranking, the recommended games whose weighted total credibility score is lower than the preset recommendation threshold can be eliminated to prevent games of extremely poor quality from entering the recommendation position.

[0138] The above embodiment provides a game information recommendation method. First, in the data collection stage, multi-source data such as social data and evaluation data are added to determine user preference vectors and game feature vectors, thereby improving the accuracy of determining user preference vectors and game feature vectors and the accuracy of the recommended games to be selected. At the same time, the application evaluation data and social comment data in the game reference data can also reflect the quality of the games, thereby re-ranking the recommended games to be selected and preventing low-quality games from affecting the user experience.

[0139] See Figure 2 In some embodiments, the above analysis of the game reference data of each game to obtain the game feature vector of each game may specifically include:

[0140] Step S21 : Perform text analysis on the application evaluation data and social comment data to obtain multiple keyword vectors.

[0141] Specifically, during the text analysis process, a preset key dictionary and semantic understanding algorithm will be used to screen descriptions and comments related to game features mentioned in application evaluation data and social comment data, such as "masterpiece", "graphics, exciting", "style, beautiful", "graphic quality, lag", "character design, brainless", etc., and use them as keyword vectors.

[0142] Step S22: clustering each keyword vector to obtain multiple feature clusters and a core vector of each feature cluster.

[0143] Among them, the clustering algorithm can adopt the K-means clustering algorithm to put keyword vectors with the same or similar features (such as keyword vectors mentioning pictures and keyword vectors mentioning painting styles) into a feature cluster, and at the same time determine the keyword vector that best reflects the feature as the core vector of this feature cluster during the clustering process.

[0144] Step S23: Perform sentiment analysis on each keyword vector to obtain a corresponding sentiment score.

[0145] Specifically, see Figure 3 ,The above sentiment analysis is performed on each keyword vector to obtain the corresponding sentiment direction, including:

[0146] Step S231: Obtain user evaluation data and user rating data from the application download platform.

[0147] Step S232: Train the TextCNN model based on the user evaluation data and the user rating data to obtain a sentiment labeling model.

[0148] Specifically, the level of user ratings can directly reflect whether users are satisfied with certain features of the game. Therefore, this application directly collects user evaluation data and associated rating data to train the TextCNN model, enabling the sentiment tagging model to learn whether the sentiment corresponding to different texts is positive (satisfied) or negative (dissatisfied).

[0149] For example, if a user severely criticizes the imbalance of the game's gameplay mechanism in the evaluation data and gives the game a low score, the TextCNN model will learn the way the user describes the gameplay in the evaluation data and identify this description as having a negative sentiment, outputting a negative sentiment score (i.e., a negative value).

[0150] Step S233: Obtain a general sentiment dictionary and a game-specific dictionary.

[0151] Among them, the general sentiment dictionary consists of common descriptive words expressing sentiment and their corresponding sentiment scores. For example, positive words include excellent, extremely beautiful, smooth, exciting, fun, etc., and negative words include brainless, garbage, laggy, boring, money-grabbing, unfinished, etc. These words have different sentiment scores according to the degree of their expression. For example, "fun" corresponds to 1 point, "extremely beautiful" corresponds to 3 points, "brainless" corresponds to -1 point, and "garbage" corresponds to -2 points.

[0152] In addition, since this application focuses on sentiment analysis in the game field, it is also necessary to pay attention to the unique expressions in the game field. The meanings of these expressive words are different in daily life and in the game field. For example, negative words such as "grind" and "pay to win" and positive words such as "casual play". This application pre-collects these unique expressive words and presets the corresponding sentiment scores as the game-specific dictionary (only needs to be constructed once and can be applied to each user's game recommendation).

[0153] Step S234: Score the keyword vector according to the general sentiment dictionary and the game-specific dictionary to obtain a dictionary score.

[0154] Specifically, match the keyword vector with the two dictionaries and use the sentiment score of the matching words as the dictionary score.

[0155] Step S235: Input the keyword vector into the sentiment tagging model to obtain a model score.

[0156] Step S236: Weight the dictionary score and the model score to obtain the sentiment score of the keyword vector.

[0157] Among them, the weighting can be to take the average of the dictionary score and the model score to obtain the sentiment score.

[0158] Specifically, this application performs sentiment analysis on keyword vectors from two different paths (dictionary and model), and finally obtains the results by weighting, avoiding the errors that may occur in one-sided analysis and improving the accuracy of sentiment scoring.

[0159] Step S24: Calculate the sentiment score of each feature cluster based on the sentiment score of each keyword vector.

[0160] Specifically, the sentiment scores of each keyword vector (including the core vector) in the feature cluster are added together, and the sum of the addition is used as the sentiment score of the feature cluster, thereby obtaining the sentiment score of each feature cluster.

[0161] Step S25: The core vector of the feature cluster whose sentiment score is greater than a preset positive threshold is used as the target vector.

[0162] Specifically, if the sentiment score is greater than the preset positive threshold, it means that there is a positive sentiment towards a certain feature of the game (such as the style, plot, etc.), that is, there are a large number of satisfied players and the evaluation is very high, so this feature can be included in the game feature vector.

[0163] It is worth noting that there can be multiple target vectors determined in this step.

[0164] Step S26: Combine the target vector and the game tag to obtain a game feature vector.

[0165] The method of extracting game feature vectors in the above embodiment, especially the process of determining the target vector based on game reference data, can accurately locate the features of a game that are highly satisfactory to players but are unlikely to be reflected in the game tag, greatly increasing the description dimension of the game feature vector, thereby improving the accuracy of locating the recommended games to be selected.

[0166] In some embodiments, the above analysis of user behavior data to obtain a user preference vector includes:

[0167] Step S31: extracting a subjective preference vector based on the subjective evaluation data and the social behavior data.

[0168] Specifically, the subjective preference vector extracted from the subjective evaluation data and social behavior data includes:

[0169] Step S311: Use the BERT model to extract at least one keyword vector from the subjective evaluation data.

[0170] Among them, BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model. Its core idea is to capture contextual information in the text through a bidirectional Transformer encoder (the bidirectional encoder part), thereby generating richer language representations.

[0171] It is worth noting that since the amount of subjective evaluation data of a single user is relatively small, this application uses a large model to extract keyword vectors. The keyword vectors extracted here are combined with contextual information and are more accurate.

[0172] The above-mentioned method of extracting keyword vectors through text analysis can quickly extract language representations, but lacks the analysis process of contextual information, so its accuracy is slightly lower than that of the BERT model. It is suitable for analyzing large amounts of text data such as social comment data.

[0173] Step S312: Determine whether the keyword vector includes a preset visual vocabulary.

[0174] Among them, the preset visual vocabulary may include descriptions related to the aesthetics of the game, such as painting style, modeling, portraits, skins, and fashions.

[0175] Step S313: If yes, ResNet50 is used to extract the game style features corresponding to the subjective evaluation data.

[0176] Step S314: Use the painting style features and the keyword vector as a subjective preference vector.

[0177] Specifically, if the user mentions descriptions related to the game graphics in the subjective evaluation data, it means that the user will consider visual features such as the style or fineness of the game graphics when choosing a game. Therefore, the ResNet50 model is used to extract the style features of the games that the user has played and evaluated and put them into the subjective preference vector.

[0178] Step S32: determining a behavior preference vector based on the login and play data and the consumption data.

[0179] Among them, the login and play data includes the number of times the user logs in to each downloaded game within a preset time period and the game duration of each login; the consumption data includes the cumulative recharge amount of the user for each downloaded game within a preset time period.

[0180] The preset time period can be within a week or a month.

[0181] The above-mentioned behavior preference vector is determined based on login and game data and consumption data, including:

[0182] Step S321: Update the login and play data according to a preset time decay factor.

[0183] Specifically, multiply the login and play data of different days by the corresponding preset time attenuation factor. For example, the preset time attenuation factor of the current day = 1, the preset time attenuation factor of the previous day = 0.85, the preset time attenuation factor of the previous two days = 0.85², and so on.

[0184] Step S322: Input the updated login and play data, consumption data, and the game feature vector of the corresponding game into the Transformer model to obtain the behavior preference score of each feature data in the game feature vector.

[0185] Step S323 sums the behavioral preference scores for the same feature data in each game feature vector to obtain a feature preference score. Specifically, if the game feature vectors for each game played by the user share the same feature, for example, if both Game A and Game B have the feature data "open world," the behavioral preference scores for the "open world" feature data in these two game feature vectors are summed, thus merging the same feature data and corresponding behavioral preference scores.

[0186] Step S324: sort each feature data according to the feature preference score.

[0187] Step S325 : taking a preset number of feature data in the first column of the sort as a behavior preference vector.

[0188] The preset number may be 5-10, that is, the top 5 or top 10 feature data are used as the behavior preference vector.

[0189] Step S33: Combine the subjective preference vector and the behavioral preference vector to obtain a user preference vector.

[0190] The method of extracting the user preference vector in the above embodiment can accurately locate the characteristic content that the user prefers or focuses on when selecting games to download and play, thereby improving the accuracy of locating the recommended games to be selected.

[0191] In some embodiments, the method further comprises:

[0192] Step S71: Obtain the latest update log of the game downloaded by the user.

[0193] Step S72: Determine whether there is a time-limited activity within a preset time period based on the latest update log.

[0194] The preset time period may be a period between the current moment (ie, the moment when the game recommendation interface is to be generated) and the start time of the limited-time activity, or a week or a month before the current moment.

[0195] Step S73: If yes, obtain the average game flow data and the second flow data within a preset time period.

[0196] Furthermore, the limited-time activities here can be further divided into resource acquisition limited-time activities and resource consumption limited-time activities, that is, the activity content of game updates can often be divided into two categories. One is that players spend time to complete tasks to obtain virtual resources within the game (such as game virtual currency, game props, etc.), and the other is that players spend game virtual resources to obtain another type of game virtual resources (such as spending card drawing props in a card drawing game to draw rewards from the card pool); this application focuses on determining whether there is a resource consumption limited-time activity, because resource acquisition limited-time activities often require players to spend time and energy, and when players participate in resource consumption limited-time activities, there is a high probability that they will recharge in order to obtain virtual props.

[0197] To this end, when this application detects that there is a resource consumption limited-time activity corresponding to the user's recharge behavior, it is necessary to consider the relationship between the user's recharge behavior and the resource consumption limited-time activity, that is, whether there is an impact of the resource consumption limited-time activity.

[0198] Step S74: Calculate the activity attenuation factor based on the average flow data and the second flow data.

[0199] Specifically, the average flow data is divided by the second flow data to obtain the activity attenuation factor.

[0200] Step S75: Update consumption data according to the activity attenuation factor.

[0201] In order to eliminate the impact of the game's limited-time resource consumption activities on the user's recharge behavior, so that the user's consumption data is more referenceable and reliable, the present application further obtains the average flow data of the game and the second flow data within a preset time period, divides the average flow data by the second flow data to obtain the activity attenuation factor, and then multiplies the activity attenuation factor by the consumption data to update the consumption data and eliminate the impact of the game's limited-time resource consumption activities in the consumption data.

[0202] It is understandable that the greater the difference between the second transaction data and the average transaction data, the more obvious the effect of this limited-time resource consumption activity in encouraging players to recharge, the smaller the corresponding activity attenuation factor, and the less reliable the consumption data.

[0203] The above embodiment updates the user's consumption data based on the limited-time activities and game turnover in the game update log, eliminates the influence of limited-time activities on the player's recharge behavior as much as possible, and improves the accuracy of the behavior preference score.

[0204] In some embodiments, the method further comprises:

[0205] Step S81: Collect the downloaded games of the user on the game recommendation interface according to a preset period.

[0206] Step S82: Obtain the cumulative playing time of the user for the downloaded game within a preset period.

[0207] Step S83: Compare the download time and cumulative playing time of the game to obtain the conversion rate.

[0208] Step S84: Determine the downloaded games with a conversion rate higher than a preset conversion threshold and the corresponding target game feature vectors.

[0209] Step S85: Update the user preference vector according to the target game feature vector.

[0210] Specifically, the target game feature vector may be directly placed in the user preference vector, or the weights corresponding to features in the user preference vector that are repeated with the target game feature vector may be increased, for example, by 0.1.

[0211] Furthermore, a first weight can be set for a game feature vector whose conversion rate is higher than a preset conversion threshold, a second weight can be set for a game feature vector that is clicked by the user but not downloaded, and a third weight can be set for a game feature vector that is clicked by the user but downloaded but has a conversion rate lower than the preset conversion threshold, where the first weight is greater than the second weight and the third weight is greater than the third weight. Then, based on the set weights, the feature weights of the features in the user preference vector that are repeated in the game feature vector are updated. When subsequently calculating the similarity between the user preference vector and each game feature vector, a weighted cosine similarity calculation can be used based on the weights.

[0212] The above embodiment adds a feedback mechanism to the generated game recommendation interface, and reversely updates the user preference vector based on the user's clicks and downloads on the game recommendation interface, further improving the accuracy of the user preference vector.

[0213] In some embodiments, the method further comprises:

[0214] Step S91: monitor the log update date of each game in real time.

[0215] Step S92: Update the game reference data of the game within a preset time period after the game log is updated.

[0216] The preset duration can be three days or a week. Understandably, not all players will log in at the exact time of a game update; a significant number of players will wait a while before updating and entering the game. Therefore, the impact of game updates on game features will take some time to be reflected in the game's reference data.

[0217] Step S93: Update the game feature vector of the game according to the updated game reference data.

[0218] Specifically, the game feature vector needs to be determined through big data analysis of a large amount of text data such as application evaluation data and social comment data, so the update process requires a certain amount of time and computing resources; and the characteristics of the game will not change significantly when it is not updated. For this reason, the present application updates the game feature vector within a preset time after the game is updated; while saving computing resources, it can also ensure the real-time and reliability of the game feature vector.

[0219] In some embodiments, the login and play data further includes a historical IP address for each login; the method further includes:

[0220] Get the current IP address and filter out abnormal login records whose historical IP addresses are not equal to the current IP address; filter and eliminate the number of logins and the game duration of each login based on the abnormal login records.

[0221] Specifically, there is no shortage of game proxy services in the gaming industry, such as game proxy services. That is, a long-term login to a game may not be performed by the user himself. If the login and play data is added to the process of determining the user preference vector at this time, it will be inaccurate. Therefore, this application deletes login records suspected of not being operated by the user himself based on the IP address before analyzing the login and play data.

[0222] See Figure 4 Another embodiment of the present application provides a game information recommendation device, comprising:

[0223] Acquisition module 101 is used to obtain user behavior data of users and game reference data of various games; wherein user behavior data includes user subjective evaluation data, login and play data, consumption data and game social data; game reference data includes game application evaluation data, game tags and social comment data.

[0224] The game analysis module 102 is used to analyze the game reference data of each game to obtain the game feature vector of each game.

[0225] The user analysis module 103 is used to analyze user behavior data to obtain user preference vectors.

[0226] The matching module 104 is used to match the user preference vector with each game feature vector to obtain a plurality of recommended games to be selected.

[0227] The reordering module 105 is configured to reorder the recommended games to be selected according to the game reference data to obtain a recommended game list.

[0228] The recommendation module 106 is configured to generate a game recommendation interface for the user based on the recommended game list and interface recommendation areas of different sizes.

[0229] In some embodiments, the game analysis module 102 includes:

[0230] The text analysis unit is used to perform text analysis on application evaluation data and social comment data to obtain multiple keyword vectors.

[0231] The clustering unit is used to cluster each keyword vector to obtain multiple feature clusters and the core vector of each feature cluster.

[0232] The sentiment analysis unit is used to perform sentiment analysis on each keyword vector and obtain the corresponding sentiment score.

[0233] The score unit is used to calculate the sentiment score of each feature cluster based on the sentiment score of each keyword vector.

[0234] The target unit is used to take the core vector of the feature cluster whose sentiment score is greater than a preset positive threshold as the target vector.

[0235] The combining unit is used to combine the target vector and the game label to obtain the game feature vector.

[0236] In some embodiments, the above-mentioned sentiment analysis unit is specifically used to obtain user evaluation data and user rating data in the application download platform; train the TextCNN model based on the user evaluation data and user rating data to obtain a sentiment labeling model; obtain a general sentiment dictionary and a game-specific dictionary; score the keyword vector based on the general sentiment dictionary and the game-specific dictionary to obtain a dictionary score; input the keyword vector into the sentiment labeling model to obtain a model score; weight the dictionary score and the model score to obtain a sentiment score of the keyword vector.

[0237] In some embodiments, the user analysis module 103 includes:

[0238] The subjective preference unit is used to extract a subjective preference vector based on the subjective evaluation data and the social behavior data.

[0239] The behavior preference unit is used to determine a behavior preference vector based on login and play data and consumption data.

[0240] The user combination unit is used to combine the subjective preference vector and the behavioral preference vector to obtain the user preference vector.

[0241] In some embodiments, the user analysis module 103 also includes an activity decay determination unit for obtaining the latest update log of the game downloaded by the user; determining whether there is a limited-time activity within a preset time period based on the latest update log; if so, obtaining the average transaction data of the game and the second transaction data within the preset time period; calculating the activity decay factor based on the average transaction data and the second transaction data; and updating the consumption data based on the activity decay factor.

[0242] In some embodiments, the device also includes a user preference update module, which is used to collect the downloaded games of the user on the game recommendation interface according to a preset period; obtain the cumulative playing time of the user on the downloaded games within the preset period; compare the download time and cumulative playing time of the user on the downloaded games to obtain the conversion rate; determine the downloaded games with a conversion rate higher than the preset conversion threshold and the corresponding target game feature vector; and update the user preference vector according to the target game feature vector.

[0243] In some embodiments, the device also includes a game feature update module for monitoring the log update date of each game in real time; updating the game reference data of the game within a preset time after the game log is updated; and updating the game feature vector of the game according to the updated game reference data.

[0244] The specific limitations of the game information recommendation device provided in this embodiment can be found in the above embodiment of the game information recommendation method, and will not be repeated here. Each module in the above-mentioned game information recommendation device can be implemented in whole or in part through software, hardware, or a combination thereof.

[0245] The above modules may be embedded in or independent of the processor in the computer device in the form of hardware, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0246] An embodiment of the present application provides a computer device, which may include a processor, memory, network interface, and database connected via a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the processor executes the steps of a game information recommendation method as described in any of the above embodiments.

[0247] The working process, working details and technical effects of the computer device provided in this embodiment can be found in the above embodiment of a game information recommendation method, which will not be described in detail here.

[0248] The embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of a method for recommending game information as described in any of the above embodiments are implemented. The computer-readable storage medium refers to a carrier for storing data, which may include, but is not limited to, a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick. The computer may be a general-purpose computer, a dedicated computer, a computer network, or other programmable device. The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiment of a method for recommending game information above, and will not be repeated here.

[0249] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0250] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0251] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for recommending game information, characterized in that: include: Obtain user behavior data and game reference data of each game; The user behavior data includes the user's subjective evaluation data, login and play data, consumption data, and game social data; the game reference data includes the game's application evaluation data, game tags, and social comment data; Analyze the game reference data of each game to obtain a game feature vector of each game; specifically, perform text analysis on the application evaluation data and the social comment data to obtain multiple keyword vectors; cluster each keyword vector to obtain multiple feature clusters and a core vector of each feature cluster; perform sentiment analysis on each keyword vector to obtain a corresponding sentiment score; calculate the sentiment score of each feature cluster based on the sentiment score of each keyword vector; use the core vector of the feature cluster whose sentiment score is greater than a preset positive threshold as a target vector; combine the target vector and the game tag to obtain the game feature vector; Analyzing the user behavior data to obtain a user preference vector; Matching the user preference vector with each of the game feature vectors to obtain a plurality of recommended games to be selected; Reorder the recommended games to be selected according to the game reference data to obtain a recommended game list; A game recommendation interface for the user is generated according to the recommended game list and interface recommendation areas of different sizes.

2. The game information recommendation method according to claim 1, characterized in that: The subjective evaluation data includes game evaluation data published by users on the application download platform; the game social data includes media data with game-related tags published by users on the social platform and interaction data published in the game chat channel.

3. The game information recommendation method according to claim 1, characterized in that: The application evaluation data includes the evaluation text, number of raters and comprehensive score of the game on various application download platforms; the social comment data includes media data released by the game-related tags of the game on at least one social platform within a preset time period.

4. The game information recommendation method according to claim 1, characterized in that: The sentiment analysis of each keyword vector is performed to obtain a corresponding sentiment direction, including: Obtain user evaluation data and user rating data from the application download platform; Training a TextCNN model based on the user evaluation data and the user rating data to obtain a sentiment labeling model; Get a general sentiment dictionary and a game-specific dictionary; Scoring the keyword vectors according to the general sentiment dictionary and the game-specific dictionary to obtain a dictionary score; Inputting the keyword vector into the sentiment tagging model to obtain a model score; The dictionary score and the model score are weighted to obtain a sentiment score of the keyword vector.

5. The game information recommendation method according to claim 1, characterized in that: The analyzing the user behavior data to obtain a user preference vector includes: Extracting a subjective preference vector based on the subjective evaluation data and the social behavior data; Determining a behavior preference vector based on the login and play data and the consumption data; The subjective preference vector and the behavioral preference vector are combined to obtain the user preference vector.

6. The game information recommendation method according to claim 5, characterized in that: The extracting of the subjective preference vector based on the subjective evaluation data and the social behavior data includes: Extracting at least one keyword vector from the subjective evaluation data using a BERT model; Determining whether the keyword vector includes a preset visual vocabulary; If yes, ResNet50 is used to extract the style features of the game corresponding to the subjective evaluation data; The painting style feature and the keyword vector are used as the subjective preference vector.

7. The game information recommendation method according to claim 5, characterized in that: The login and play data includes the number of times the user logs into each downloaded game within a preset time period and the game duration of each login; The consumption data includes the cumulative recharge amount of the user for each downloaded game within a preset time period.

8. The game information recommendation method according to claim 7, characterized in that: The determining of the behavior preference vector based on the login and play data and the consumption data includes: Updating the login play data according to a preset time decay factor; Input the updated login and play data, consumption data, and the game feature vector of the corresponding game into the Transformer model to obtain the behavior preference score for each feature data in the game feature vector; Summing up the behavioral preference scores of the same feature data in each of the game feature vectors to obtain a feature preference score; sorting each of the feature data according to the feature preference score; A preset number of feature data in the first column of the sort is used as the behavior preference vector.

9. The game information recommendation method according to claim 8, characterized in that: Also includes: Get the latest update log of the game downloaded by the user; Determine whether there is a time-limited activity within a preset time period based on the latest update log; If yes, then obtain the average turnover data of the game and the second turnover data within a preset time period; Calculate an activity attenuation factor based on the average flow data and the second flow data; The consumption data is updated according to the activity decay factor.

10. The game information recommendation method according to claim 1, characterized in that: Also includes: collecting downloaded games from the user on the game recommendation interface according to a preset period; Obtaining the cumulative playing time of the user for the downloaded game within the preset period; Comparing the download time and cumulative playing time of the user for the downloaded game to obtain a conversion rate; Determining the downloaded games with a conversion rate higher than a preset conversion threshold and the corresponding target game feature vectors; The user preference vector is updated according to the target game feature vector.

11. The game information recommendation method according to claim 1, characterized in that: Also includes: Real-time monitoring of the log update date of each game; Updating game reference data of the game within a preset time period after the game log is updated; The game feature vector of the game is updated according to the updated game reference data.

12. The game information recommendation method according to claim 7, characterized in that: The login and play data also includes the historical IP address of each login; the method further includes: Obtain the current IP address and filter out abnormal login records whose historical IP addresses are not equal to the current IP address; The number of logins and the game duration of each login are screened and eliminated based on the abnormal login records.

13. The game information recommendation method according to claim 1, characterized in that: The step of reordering the recommended games to be selected according to the game reference data to obtain a recommended game list includes: Obtain the number of ratings and comprehensive ratings of the recommended game on any application download platform; Calculating the credibility score of the recommended game to be selected based on the number of scorers and the comprehensive score; Determine the corresponding trust weight based on the average daily active users of each app download platform; Weighting the credibility scores of the recommended games on various application download platforms according to the credibility weights; The recommended games are reordered according to the weighted total of the credibility scores to obtain a recommended game list.

14. A game information recommendation device, characterized in that: include: The acquisition module is used to obtain the user behavior data of users and the game reference data of each game; The user behavior data includes the user's subjective evaluation data, login and play data, consumption data, and game social data; the game reference data includes the game's application evaluation data, game tags, and social comment data; A game analysis module is configured to analyze game reference data of each game to obtain a game feature vector of each game; the game analysis module specifically includes: a text analysis unit, configured to perform text analysis on application evaluation data and social comment data to obtain multiple keyword vectors; a clustering unit, configured to cluster each keyword vector to obtain multiple feature clusters and a core vector of each feature cluster; a sentiment analysis unit, configured to perform sentiment analysis on each keyword vector to obtain a corresponding sentiment score; a score unit, configured to calculate a sentiment score for each feature cluster based on the sentiment score of each keyword vector; a target unit, configured to use the core vector of the feature cluster whose sentiment score is greater than a preset positive threshold as a target vector; and a combination unit, configured to combine the target vector and the game tag to obtain the game feature vector. A user analysis module, configured to analyze the user behavior data to obtain a user preference vector; A matching module is used to match the user preference vector with each game feature vector to obtain multiple recommended games; A reordering module is used to reorder each recommended game to be selected according to the game reference data to obtain a recommended game list; The recommendation module is used to generate a game recommendation interface for users based on the recommended game list and interface recommendation areas of different sizes.

15. The game information recommendation device according to claim 14, characterized in that: The sentiment analysis unit is used to: Obtain user evaluation data and user rating data from the application download platform; Training a TextCNN model based on the user evaluation data and the user rating data to obtain a sentiment labeling model; Get a general sentiment dictionary and a game-specific dictionary; Scoring the keyword vectors according to the general sentiment dictionary and the game-specific dictionary to obtain a dictionary score; Inputting the keyword vector into the sentiment tagging model to obtain a model score; The dictionary score and the model score are weighted to obtain a sentiment score of the keyword vector.

16. The game information recommendation device according to claim 14, characterized in that: The user analysis module includes: A subjective preference unit, configured to extract a subjective preference vector based on the subjective evaluation data and the social behavior data; a behavior preference unit, configured to determine a behavior preference vector based on the login and play data and the consumption data; The user combination unit is configured to combine the subjective preference vector and the behavioral preference vector to obtain a user preference vector.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the game information recommendation method according to any one of claims 1 to 13 are implemented.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the game information recommendation method according to any one of claims 1 to 13 are implemented.

Images