Game user loss prediction method and device
By constructing a behavior vector pre-training model, the loss of game users is predicted based on the timing of behavior information in the game, the problems of large data demand and subjectivity in the existing technology are solved, and the accurate prediction of the loss trend of game users is achieved.
Patent Information
- Application Number
- CN202411135599.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-12-13
AI Technical Summary
The prior art has problems with large data demand, subjectivity and one-sidedness when predicting game user churn, making it difficult to accurately predict user churn trends in small games or new games.
By constructing a behavior vector pre-training model, the behavior representation vector is determined based on the timing of the behavior information in the game, combined it into a behavior vector, and input a preset churn prediction model for prediction, dynamically capture the timing of user behavior information, and reduce the amount of sample data rely.
It improves the accuracy and generalization ability of game user churn prediction, can better understand user life cycle and churn trends, and does not require a large amount of sample data to participate in training, and is suitable for small games and new games.
Smart Images
Figure CN119113527A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of Internet technology, and in particular to a method and device for predicting game user churn. Background Art
[0002] With the rapid development of the gaming industry, the types and number of games have continued to increase, attracting more and more game users. The loss of game users has become a common problem for all games. Predicting game user loss can help game operators better understand game user needs, optimize game design and operation strategies, and improve game user retention and activity, thereby achieving more refined and personalized operation management. Summary of the invention
[0003] In view of the above problems, embodiments of the present application are proposed to provide a method and device for predicting game user churn that overcomes the above problems or at least partially solves the above problems.
[0004] According to a first aspect of an embodiment of the present application, a method for predicting game user churn is provided, which includes:
[0005] Construct a behavior vector pre-training model to determine the behavior representation vectors corresponding to different behavior information based on the time sequence between behavior information in the game;
[0006] Obtaining information about each behavior of the game user to be predicted in the game, and obtaining corresponding representation vectors of each behavior;
[0007] Combine the behavior representation vectors to obtain the behavior vector of the game user to be predicted;
[0008] The behavior vector of the game user to be predicted is input into a preset churn prediction model to predict whether the game user to be predicted will churn.
[0009] Optionally, determining the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game further includes:
[0010] Mapping each behavior information of historical game users in the game into behavior text, and constructing the behavior sequence text of historical game users based on the time sequence;
[0011] Input the behavior sequence text into the behavior vector pre-training model to obtain the behavior sequence vector of the historical game user;
[0012] Slice the behavior sequence vector and perform dimensionality reduction clustering to obtain the representation vector of each behavior.
[0013] Optionally, the behavior information includes the game user's behavior operation information and corresponding behavior result information; the behavior information includes the game user's active behavior information and the game user's passive behavior information.
[0014] Optionally, mapping each behavior information of the historical game user in the game into a behavior text, and constructing the behavior sequence text of the historical game user based on the time sequence further includes:
[0015] Mapping the historical game user's behavior operation information in the game and the corresponding behavior result information into behavior texts one by one, and combining the behavior texts of the behavior operation information with the behavior texts of the behavior result information to obtain a set of behavior texts;
[0016] Based on the operation time sequence of the behavior operation information, each group of behavior texts is combined to obtain the behavior sequence text of the historical game user.
[0017] Optionally, the behavior text includes Chinese text;
[0018] Inputting the behavior sequence text into the behavior vector pre-training model to obtain the behavior sequence vector of the historical game user further includes:
[0019] The behavior sequence text is input into a behavior vector pre-training model so as to convert the behavior sequence text into a corresponding behavior sequence vector; the behavior vector pre-training model includes a natural language processing model.
[0020] Optionally, slicing is performed according to the behavior sequence vector, and dimension reduction clustering is performed to obtain each behavior representation vector, which further includes:
[0021] Slice the behavior sequence vector according to the behavior order to obtain multiple vectors containing each behavior;
[0022] Perform dimensionality reduction clustering on multiple vectors to obtain the behavior representation vector corresponding to the behavior.
[0023] Optionally, the method further comprises:
[0024] The preset churn prediction model is trained based on the sample behavior vectors and sample churn labels of sample game users.
[0025] According to a second aspect of an embodiment of the present application, a game user churn prediction device is provided, comprising:
[0026] A pre-training module is used to construct a behavior vector pre-training model, and determine the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game;
[0027] A representation vector module, adapted to obtain information on various behaviors of the game user to be predicted in the game, and obtain corresponding representation vectors of each behavior;
[0028] A behavior vector module, adapted to combine various behavior representation vectors to obtain a behavior vector of the game user to be predicted;
[0029] The prediction module is adapted to input the behavior vector of the game user to be predicted into a preset churn prediction model to predict whether the game user to be predicted will churn.
[0030] According to a third aspect of an embodiment of the present application, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;
[0031] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned game user churn prediction method.
[0032] According to a fourth aspect of an embodiment of the present application, a computer storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned game user churn prediction method.
[0033] According to a fifth aspect of an embodiment of the present application, a computer program product is provided, comprising at least one executable instruction, wherein the executable instruction enables a processor to perform operations corresponding to the above-mentioned game user churn prediction method.
[0034] According to the game user churn prediction method and device provided by the present application, according to the time sequence between the behavior information occurring in the game and the behavior vector pre-training model, through context learning, the behavior representation vectors corresponding to the different behavior information can be obtained, the behavior information of the game user to be predicted in the game is converted into the corresponding behavior representation vector, and the behavior representation vectors are combined to obtain the behavior vector of the game user to be predicted, so as to better understand the life cycle and churn trend of the game user, and to obtain the churn result based on the behavior vector prediction.
[0035] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0037] Figure 1 A flowchart of a method for predicting game user churn according to an embodiment of the present application is shown;
[0038] Figure 2 A flowchart of a method for predicting game user churn according to another embodiment of the present application is shown;
[0039] Figure 3 A schematic diagram showing a behavior text sequence corresponding to behavior information is shown;
[0040] Figure 4 A schematic diagram of position coding is shown;
[0041] Figure 5 A schematic diagram showing a behavior sequence vector obtained by a behavior vector pre-training model;
[0042] Figure 6 A schematic diagram showing the structure of a device for predicting game user churn according to an embodiment of the present application is shown;
[0043] Figure 7 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. DETAILED DESCRIPTION
[0044] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.
[0045] First, the terms involved in one or more embodiments of the present application are explained.
[0046] Game user retention: Game users continue to stay in the game and participate in the game after completing the game. In the game industry, retention rate is one of the important indicators to measure the success of a game.
[0047] Figure 1 A flowchart of a method for predicting game user churn according to an embodiment of the present application is shown. Figure 1 As shown, the method comprises the following steps:
[0048] Step S101, constructing a behavior vector pre-training model, and determining the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game.
[0049] Predicting the churn of game users can help game operators better understand the retention of game users. Furthermore, they can respond to game users with high churn risks in a timely manner, such as taking corresponding customized measures and promotional activities to retain game users and improve the retention rate and activity of the game.
[0050] When predicting or preventing the loss of game users, the following methods are mostly used:
[0051] For example, by analyzing the gaming behavior data of game users, such as login frequency, game duration, task completion, consumption, etc., we can identify the game users' loyalty and activity to the game. Game users with high login frequency, long game duration, active task completion and high completion, high consumption, etc., have high loyalty to the game and are active game users. By analyzing the above data, we can predict the loss of game users. However, this method requires a lot of data support. For small games or new games, there are problems such as insufficient game user data, which may lead to inaccurate analysis and prediction results. Here, the behavior of game users will change with the game time, and the behavior data will be constantly updated. Correspondingly, the prediction model also needs to be constantly updated, which requires a lot of workload and cost.
[0052] Alternatively, based on the social relationships of game users in the game, the number of friends, family participation, and speech of game users can be analyzed to understand the social status of game users. When game users make friends, participate in families, and actively speak in the game, the game users have a high stickiness to the game. Game friends and families will also attract game users to continue to participate in the game, thereby judging whether game users are likely to churn based on social relationships. However, this method relies on social relationship data. If the user does not actively participate in social activities but actively plays games, the predicted results are limited and cannot be accurately predicted. When the game can be participated normally without social interaction, the correlation between social relationships and user churn may not be high, and other factors must be combined for accurate analysis. Obviously, this method cannot obtain accurate results.
[0053] Alternatively, based on the analysis of in-game events, such as game users failing to participate in a level, insufficient game currency, etc., game users may lose interest in the game and then lose users. By analyzing the frequency and distribution of the above events, the possibility of losing game users can be predicted. However, when analyzing events for specific events, it is easy to ignore other behaviors of game users and the influence of game factors, etc., which affects the accuracy of the prediction results. Furthermore, sometimes there are accidental factors in the occurrence of events themselves, which are not directly related to the loss of game users, making the prediction results less accurate.
[0054] Alternatively, by using machine learning, data mining and other technologies, such as decision trees, random forests, logistic regression and other algorithms, a large amount of game user data can be modeled and analyzed to predict the possibility of game user churn. However, machine learning and data mining require a large amount of labeled sample data, and the selection of suitable algorithms and algorithm parameters based on experience, which requires a large amount of data and is not suitable for small games and new games. In addition, the generalization ability of machine learning models, overfitting or underfitting, also affects the accuracy of the prediction results.
[0055] Alternatively, by regularly conducting user surveys and collecting user feedback information, we can understand the satisfaction and needs of game users, promptly discover problems in products or services, and take corresponding measures to prevent user loss. However, this method relies on the active participation and real feedback of users. If some users are unwilling to participate in surveys or feedback, the data volume will be insufficient or biased. User feedback is also affected by subjective factors and may not objectively reflect the reasons for the loss of game users, resulting in inaccurate data.
[0056] The above prediction methods all have various problems, such as large data demand, subjectivity, and one-sidedness, which affect the accuracy of the prediction results. This embodiment takes the above problems into consideration, and pre-trains the behavior vector pre-training model according to the time sequence of the game user's behavior information in the game to obtain the behavior representation vector corresponding to the behavior information, which can be trained without a large amount of sample data. Dynamically capture behavior information, taking into account the time sequence of game user behavior information, and more convenient to understand the interests and needs of game users.
[0057] Specifically, the behavior vector pre-training model can adopt a natural language processing model. Each behavior information in the game can be mapped into different behavior texts, and then the behavior sequence text corresponding to the behavior information sequence can be obtained after arranging them in chronological order. The behavior sequence text is input into the behavior vector pre-training model, and the behavior sequence text is converted into a behavior sequence vector through language processing, analysis, etc. Then the behavior sequence vector is sliced, and the behavior sequence vector is sliced into multiple vectors, each vector corresponding to different information, and the behavior representation vector corresponding to each behavior information is obtained. The above processing can use the behavior information of historical game users to train the behavior vector pre-training model, and obtain the behavior sequence vector corresponding to the behavior sequence text through context learning by using the context association relationship of the behavior sequence text. Through slicing processing, the behavior sequence vector is divided to obtain the behavior representation vector corresponding to different behavior information in the game. There is no need to collect a large amount of sample data in the whole process, and the behavior sequence method is used to serialize and vectorize the behavior information of game users. The obtained behavior sequence vector is more convenient for mining the behavior details of game users, and improves the accuracy and generalization ability of the subsequent preset churn prediction model.
[0058] The behavior representation vector corresponds to the behavior information, such as a multi-dimensional vector representation, and different behavior representation vectors correspond to vectors of different dimensions, such as the behavior representation vector of login behavior is [100000], the behavior representation vector of emailing behavior is [010000], the behavior representation vector of recruitment behavior is [001000], the behavior representation vector of shopping behavior is [000100], etc., and the behavior representation vector of click behavior is [000010]. The above is an example, and other vector forms can also be used to represent the behavior representation vector, which is not limited here.
[0059] Furthermore, before obtaining the behavior information for mapping, the obtained behavior information can also be cleaned and preprocessed, including removing invalid data, processing missing values, and normalizing data, and then the cleaned and preprocessed behavior information is mapped into behavior text to obtain a standardized and valid behavior sequence text.
[0060] Step S102, obtaining information on various behaviors of the game user to be predicted in the game, and obtaining corresponding representation vectors of each behavior.
[0061] For the game users to be predicted, various behavior information of the game users to be predicted in the game can be obtained by, for example, embedding points in the game code, etc. The behavior information includes the behavior information actively initiated by the game users in the game, as well as the game behavior information passively generated in the game. When obtaining, it can be obtained according to a preset time period, such as obtaining various behavior information of the game users to be predicted within a day, to avoid too much behavior information increasing too much data calculation, affecting the prediction accuracy, etc.
[0062] According to the obtained behavior information of the game user to be predicted in the game, and the behavior representation vector corresponding to the behavior information obtained by the behavior vector pre-training model, the behavior representation vector of the game user to be predicted can be obtained. For example, if the behavior information of the game user to be predicted in the game includes login behavior, recruitment behavior, shopping behavior, etc., the obtained behavior representation vectors include [100000], [001000], [000100]. The above is an example, which is set according to the implementation situation and is not limited here.
[0063] Furthermore, after obtaining the behavior information of the game user to be predicted, the obtained behavior information can also be cleaned and preprocessed, including removing invalid data, processing missing values, and normalizing data, and then the cleaned and preprocessed behavior information is mapped into a behavior representation vector.
[0064] It should be noted that the user information (including but not limited to user terminal information, user account information, etc.) and user data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.
[0065] Step S103: combine the behavior representation vectors to obtain the behavior vector of the game user to be predicted.
[0066] The behavior representation vectors of the game user to be predicted are combined. The behavior representation vector is a vector corresponding to different behaviors. When combining, the behavior representation vectors are added. For example, if the obtained behavior representation vectors are [100000], [001000], and [000100], the behavior vector [101100] of the game user to be predicted is added.
[0067] Step S104, inputting the behavior vector of the to-be-predicted game user into a preset churn prediction model to predict whether the to-be-predicted game user will churn.
[0068] The preset churn prediction model can be trained with sample data of the behavior vector and user churn labels to obtain a trained preset churn prediction model. The preset churn prediction model can also be evaluated and tuned through evaluation indicators such as cross-validation, ROC curve, AUC value, etc., to timely discover the deficiencies of the preset churn prediction model and make improvements.
[0069] According to the preset churn prediction model obtained through training, the behavior vector of the game user to be predicted is input into the preset churn prediction model, and the game user to be predicted is predicted based on the preset churn prediction model to obtain a prediction result. According to the prediction result, it can be determined whether the game user to be predicted will churn.
[0070] The game users to be predicted can be new game users. In the case of less behavioral information about them, churn prediction can be made for them to timely find out whether new game users will churn, so that game operations can better understand the retention of new game users, take timely measures, deeply understand the needs and preferences of new game users, and provide personalized recommendations and suggestions, such as preferential activities, customized tasks, etc., to improve the participation and retention rate of new game users. Furthermore, for game users with high churn risk, their churn reasons and behavioral data can be analyzed, and problems in game design, such as excessive difficulty and lack of interaction, can be found in time, and adjustments and improvements can be made in time to improve the playability and attractiveness of the game. Game operations can also achieve more refined operations based on churn predictions, such as providing different rewards and activities for game users at different levels to improve the activity and shopping willingness of game users at all levels.
[0071] According to the game user churn prediction method provided by the present application, based on the time sequence of the behavior information occurring in the game and based on the behavior vector pre-training model, through context learning, the behavior representation vectors corresponding to the different behavior information can be obtained, the behavior information of the game user to be predicted in the game is converted into the corresponding behavior representation vector, and the behavior representation vectors are combined to obtain the behavior vector of the game user to be predicted. This can better understand the life cycle and churn trend of game users, so as to obtain the churn results based on the behavior vector prediction.
[0072] Figure 2 A flowchart of a method for predicting game user churn according to an embodiment of the present application is shown. Figure 2 As shown, the method comprises the following steps:
[0073] Step S201, mapping each behavior information of the historical game user in the game into a behavior text, and constructing a behavior sequence text of the historical game user based on the time sequence.
[0074] In order to obtain the behavior representation vectors corresponding to different behavior information in the game, training can be performed through different behavior information of historical game users in the game. Specifically, the behavior information of historical game users in the game is obtained, and each behavior information can be mapped into a corresponding behavior text. The behavior text includes, for example, Chinese text, that is, different behavior information is mapped into different Chinese texts, and then the mapped Chinese texts are combined in sequence according to the occurrence time of the behavior information to construct the behavior sequence text of the historical game users.
[0075] Furthermore, the behavior information includes the behavior operation information event_id of the game user and the corresponding behavior result information target_id. For example, the behavior operation information includes the game user's login, and the behavior result information includes the login time, login duration, etc.; the behavior operation information includes the game user sending messages to friends, and the behavior result information includes the game user's social result, etc. When mapping the behavior information to behavior texts, the behavior operation information and behavior result information of historical game users can be mapped separately. Different behavior operation information and behavior result information are mapped to different behavior texts. For example, the behavior operation information is mapped to the behavior text "ge", and the behavior result information is mapped to the behavior text "ding", etc. The behavior text can be a single text, or a phrase, a word group, etc. The above is for illustrative purposes, and the specific mapping of behavior texts can be set according to the implementation situation and is not limited here.
[0076] Furthermore, the behavior information includes the active behavior information of the game user and the passive behavior information of the game user. The active behavior information of the game user is the behavior initiated by the game user actively, such as the game user logging in, the game user viewing emails, etc. The passive behavior information is the behavior generated passively by the game user, such as the game user receiving an email (an email sent by the game system), etc. In the active behavior information, the active behavior operation information of the game user will generate behavior result information, which is collected according to the active behavior operation; in the passive behavior information, the passive behavior operation information of the game user will not generate a result, and the behavior result information can be set to be empty, etc.
[0077] For historical game users, after mapping the behavior operation information and the corresponding behavior result information in the game to behavior texts respectively, the behavior text of the behavior operation information and the behavior text of the behavior result information are combined correspondingly to obtain a group of behavior texts. For example, the behavior operation information is mapped to the behavior text "ge", and the behavior result information is mapped to the behavior text "ding". Combining the behavior text of the behavior operation information and the behavior text of the behavior result information, we get "geding". Then, based on the chronological order of the operation time of the behavior operation information, the groups of behavior texts are combined to obtain the behavior sequence text of the historical game user. As Figure 3 shown, the behavior operation information includes login, sending emails, receiving system emails, shopping, etc., and the corresponding behavior result information are login time duration, social relationship, empty (passive behavior), increased funds, etc. Map the corresponding behavior texts to each behavior operation information and behavior result information, and then combine the behavior text of the behavior operation information and the behavior text of the behavior result information correspondingly. The behavior sequence text obtained by combining them in order is gedingyichouluanqixiba. Among them, the orange box represents the active behavior of the game user, and the blue box represents the passive behavior of the game user. The above is for illustrative purposes and is specifically set according to the implementation situation and is not limited here.
[0078] Step S202: input the behavior sequence text into the behavior vector pre-training model to obtain the behavior sequence vector of the historical game user.
[0079] The behavior text includes, for example, Chinese text, and the behavior sequence text is a sequence of Chinese text. In this embodiment, the behavior vector pre-training model uses, for example, a natural language processing model, such as the MLM task (Masked Language Model) of the Bert model, and inputs the behavior sequence text into the behavior vector pre-training model, such as the MLM model, and converts the behavior sequence text into a corresponding behavior sequence vector by analyzing and processing the Chinese text. The behavior sequence text can be encoded in a positional manner, such as Figure 4 As shown in the figure, the input behavior sequence text includes a start tag CLS, an end tag SEP or a segmentation tag SEP, and the input behavior sequence text is between the two. Through token embedding TokenEmbeddings+segment embedding Segment Embedding+position embedding Position Embedding, it is input into the behavior vector pre-training model to obtain the corresponding behavior sequence vector. Figure 4 My dog is cute, he likes playing are used as examples. The corresponding behavior sequence texts are set according to the implementation situation and will not be explained in detail here.
[0080] The behavior sequence vector of game users converts the behavior information of game users into high-dimensional feature representation, digging out more details and patterns about the behavior of game users, making it easier to use the feature representation ability to improve the accuracy and generalization ability of the preset churn prediction model. The behavior sequence vector also takes time factors into consideration and analyzes the changes in the behavior of game users over time, which can help better understand the life cycle and churn trend of game users. The behavior sequence vector combines contextual information, such as the user profile, historical behavior, social relationships, etc. of game users, to provide more comprehensive and accurate data support for the preset churn prediction model.
[0081] This embodiment adopts the BERT model, such as Figure 5 As shown in the figure, each word contained in the behavior text sequence (i.e., the single text sequence in the figure) is input into the BERT model, and the corresponding representations are obtained based on the MLM task. After passing through the dense layer, the final label, i.e., the behavior sequence vector, is obtained.
[0082] Step S203: Slice the behavior sequence vectors and perform dimensionality reduction clustering to obtain each behavior representation vector.
[0083] The behavior representation vector represents the vector corresponding to different behavior information. The behavior sequence vector contains multiple vectors of different behaviors. After obtaining the behavior sequence vector, it can be sliced according to the behavior sequence to obtain multiple vectors containing each behavior. The vector is high-dimensional. Multiple vectors are clustered by dimensionality reduction to obtain the behavior representation vector corresponding to each behavior. When reducing the dimension, t-SNE (t-Distributed Stochastic Neighbor Embedding) can be used to reduce the vector dimension to a low-dimensional visualization to obtain the behavior representation vector. The behavior representation vector represents different behavior information.
[0084] Step S204, obtaining information on various behaviors of the game user to be predicted in the game, and obtaining corresponding representation vectors of each behavior.
[0085] For the game users to be predicted, various behavior information of the game users to be predicted in the game is obtained based on game embedding points, game platforms, third-party platforms, etc., and various behavior representation vectors of the game users to be predicted are obtained according to the behavior representation vectors corresponding to the obtained various behavior information.
[0086] Step S205: combine the behavior representation vectors to obtain the behavior vector of the game user to be predicted.
[0087] The behavior representation vectors of the game user to be predicted are combined according to the corresponding bit number, such as adding the behavior representation vectors bit by bit, to obtain the behavior vector of the game user to be predicted. The behavior vector of the game user to be predicted indicates various behavior information, behavior timing, social information generated by the behavior, event information generated by the behavior, etc. of the game user to be predicted, which is determined based on the user's objective behavior.
[0088] Step S206, training a preset churn prediction model based on the sample behavior vectors and sample churn labels of the sample game users.
[0089] The preset churn prediction model can predict whether a game user has a churn risk. The preset churn prediction model can be pre-trained, and the training process includes: collecting sample behavior vectors of sample game users, and marking sample churn labels for the sample game users according to their corresponding labels, and the sample churn labels include churn and non-churn. The sample behavior vectors and sample churn labels of the sample game users are input into the preset churn prediction model, and the preset churn prediction model is trained to obtain the trained preset churn prediction model.
[0090] Furthermore, the preset churn prediction model can be fine-tuned in real time using data from new game users, and can also be optimized and improved through evaluation indicators such as cross-validation, ROC curve, and AUC value.
[0091] Step S207: inputting the behavior vector of the to-be-predicted game user into a preset churn prediction model to predict whether the to-be-predicted game user will churn.
[0092] The obtained behavior vector of the game user to be predicted is input into a preset churn prediction model, and a prediction result is obtained according to the preset churn prediction model. The prediction result can predict whether the game user to be predicted will churn.
[0093] Furthermore, for game users with a high risk of churn, corresponding measures can be taken to retain game users, such as sending customized push notifications, providing preferential activities, etc. At the same time, feedback and behavior information of game users can be continuously collected to further update and optimize the models used in this embodiment for more accurate prediction.
[0094] According to the game user churn prediction method provided by the present application, the behavior information occurring in the game is mapped into different behavior texts, and the behavior texts are combined to obtain a behavior text sequence according to the time sequence of the behavior information. The behavior text vector is obtained by using the behavior vector pre-training model through context learning. The behavior text vector is sliced and processed, and after dimensionality reduction and clustering, the behavior representation vector corresponding to the different behavior information is obtained. The behavior information of the game user to be predicted in the game can be converted into their respective behavior representation vectors, and the behavior representation vectors are combined and added to obtain the behavior vector of the game user to be predicted. The behavior vector is input into the preset churn prediction model so that the preset churn prediction model can better predict based on the behavior vector according to the game user behavior, behavior sequence, behavior-related social relationships, event occurrence and other information contained in the behavior vector to obtain the churn result.
[0095] Figure 6 FIG. 1 is a schematic diagram showing the structure of a device for predicting game user churn provided by an embodiment of the present application. Figure 6 As shown, the device comprises:
[0096] A pre-training module 610 is adapted to construct a behavior vector pre-training model, and determine the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game;
[0097] The characterization vector module 620 is adapted to obtain information on various behaviors of the game user to be predicted in the game and obtain corresponding characterization vectors of each behavior;
[0098] A behavior vector module 630, adapted to combine various behavior representation vectors to obtain a behavior vector of the game user to be predicted;
[0099] The prediction module 640 is adapted to input the behavior vector of the game user to be predicted into a preset churn prediction model to predict whether the game user to be predicted will churn.
[0100] Optionally, the pre-training module 610 is further adapted to:
[0101] Mapping each behavior information of historical game users in the game into behavior text, and constructing the behavior sequence text of historical game users based on the time sequence;
[0102] Input the behavior sequence text into the behavior vector pre-training model to obtain the behavior sequence vector of the historical game user;
[0103] Slice the behavior sequence vector and perform dimensionality reduction clustering to obtain the representation vector of each behavior.
[0104] Optionally, the behavior information includes the game user's behavior operation information and corresponding behavior result information; the behavior information includes the game user's active behavior information and the game user's passive behavior information.
[0105] Optionally, the pre-training module 610 is further adapted to:
[0106] Mapping the historical game user's behavior operation information in the game and the corresponding behavior result information into behavior texts one by one, and combining the behavior texts of the behavior operation information with the behavior texts of the behavior result information to obtain a set of behavior texts;
[0107] Based on the operation time sequence of the behavior operation information, each group of behavior texts is combined to obtain the behavior sequence text of the historical game user.
[0108] Optionally, the behavior text includes Chinese text;
[0109] The pre-training module 610 is further adapted to:
[0110] The behavior sequence text is input into a behavior vector pre-training model so as to convert the behavior sequence text into a corresponding behavior sequence vector; the behavior vector pre-training model includes a natural language processing model.
[0111] Optionally, the pre-training module 610 is further adapted to:
[0112] Slice the behavior sequence vector according to the behavior order to obtain multiple vectors containing each behavior;
[0113] Perform dimensionality reduction clustering on multiple vectors to obtain the behavior representation vector corresponding to the behavior.
[0114] Optionally, the device further includes: a training module 650, adapted to obtain a preset churn prediction model through training based on sample behavior vectors and sample churn labels of sample game users.
[0115] The description of each module above refers to the corresponding description in the method embodiment and will not be repeated here.
[0116] According to the game user churn prediction device provided by the present application, based on the time sequence of the behavior information occurring in the game and based on the behavior vector pre-training model, through context learning, the behavior representation vectors corresponding to different behavior information can be obtained, the behavior information of the game user to be predicted in the game can be converted into the corresponding behavior representation vector, and the behavior representation vectors can be combined to obtain the behavior vector of the game user to be predicted. This can better understand the life cycle and churn trend of game users, so as to predict the churn results based on the behavior vectors.
[0117] The present application also provides a non-volatile computer storage medium, which stores at least one executable instruction, and the executable instruction can execute operations corresponding to the game user churn prediction method in any of the above method embodiments.
[0118] The present application also provides a computer program product, which includes at least one executable instruction or computer program, and the executable instruction or computer program can enable a processor to perform operations corresponding to the game user churn prediction method in any of the above method embodiments.
[0119] Figure 7 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. The specific embodiment of the present application does not limit the specific implementation of the computing device.
[0120] like Figure 7 As shown, the computing device may include: a processor (processor) 702 , a communications interface (Communications Interface) 704 , a memory (memory) 706 , and a communication bus 708 .
[0121] in:
[0122] The processor 702 , the communication interface 704 , and the memory 706 communicate with each other via a communication bus 708 .
[0123] The communication interface 704 is used to communicate with other devices such as clients or other servers.
[0124] The processor 702 is used to execute the program 710, and specifically can execute the relevant steps in the above-mentioned game user churn prediction method embodiment.
[0125] Specifically, the program 710 may include program codes, which include computer operation instructions.
[0126] The processor 702 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the present application. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.
[0127] The memory 706 is used to store the program 710. The memory 706 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0128] The program 710 can be specifically used to enable the processor 702 to execute the game user churn prediction method in any of the above method embodiments. The specific implementation of each step in the program 710 can refer to the corresponding descriptions in the corresponding steps and units in the above game user churn prediction embodiments, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above-mentioned method embodiments, which will not be repeated here.
[0129] The algorithm or display provided here is not inherently related to any specific computer, virtual system or other device. Various general systems can also be used together with the teaching based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the application is not directed to any specific programming language either. It should be understood that various programming languages can be utilized to realize the content of the application described here, and the above description of specific languages is to disclose the preferred embodiment of the application.
[0130] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.
[0131] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.
[0132] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.
[0133] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.
[0134] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some or all components of the present application. The present application can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing a part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.
[0135] It should be noted that the above embodiments illustrate the present application rather than limit the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be constructed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The present application may be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.
Claims
1. A method for predicting game user churn, comprising: Construct a behavior vector pre-training model to determine the behavior representation vectors corresponding to different behavior information based on the time sequence between behavior information in the game; Obtaining information about each behavior of the game user to be predicted in the game, and obtaining corresponding representation vectors of each behavior; Combining the behavior representation vectors to obtain the behavior vector of the game user to be predicted; The behavior vector of the game user to be predicted is input into a preset churn prediction model to predict whether the game user to be predicted will churn.
2. The method according to claim 1, wherein: The step of determining the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game further includes: Mapping each behavior information of the historical game user in the game into a behavior text, and constructing a behavior sequence text of the historical game user based on the time sequence; Inputting the behavior sequence text into a behavior vector pre-training model to obtain a behavior sequence vector of the historical game user; Slicing processing is performed according to the behavior sequence vector, and dimension reduction clustering is performed to obtain each behavior representation vector.
3. The method according to claim 2, wherein: The behavior information includes the game user's behavior operation information and the corresponding behavior result information; the behavior information includes the game user's active behavior information and the game user's passive behavior information.
4. The method according to claim 3, wherein: The mapping of each behavior information of the historical game user in the game into a behavior text, and constructing the behavior sequence text of the historical game user based on the time sequence further includes: Mapping the behavior operation information of the historical game user in the game and the corresponding behavior result information into behavior texts one by one, and combining the behavior texts of the behavior operation information with the behavior texts of the behavior result information to obtain a set of behavior texts; Based on the operation time sequence of the behavior operation information, each group of behavior texts is combined to obtain the behavior sequence text of the historical game user.
5. The method according to claim 2, wherein: The said text of the act includes Chinese text; Inputting the behavior sequence text into the behavior vector pre-training model to obtain the behavior sequence vector of the historical game user further includes: The behavior sequence text is input into a behavior vector pre-training model so as to convert the behavior sequence text into a corresponding behavior sequence vector; the behavior vector pre-training model includes a natural language processing model.
6. The method according to any one of claims 2 to 5, wherein: The slicing process and dimensionality reduction clustering according to the behavior sequence vector to obtain each behavior representation vector further includes: Slicing the behavior sequence vector according to the behavior sequence to obtain multiple vectors containing each behavior; Perform dimensionality reduction clustering on the multiple vectors to obtain a behavior representation vector corresponding to the behavior.
7. The method according to any one of claims 1 to 6, wherein: The method further comprises: The preset churn prediction model is trained based on the sample behavior vectors and sample churn labels of sample game users.
8. A device for predicting game user churn, comprising: A pre-training module is used to construct a behavior vector pre-training model, and determine the behavior representation vectors corresponding to different behavior information according to the time sequence between the behavior information in the game; A representation vector module, adapted to obtain information on various behaviors of the game user to be predicted in the game, and obtain corresponding representation vectors of each behavior; A behavior vector module, adapted to combine the behavior representation vectors to obtain the behavior vector of the game user to be predicted; The prediction module is adapted to input the behavior vector of the to-be-predicted game user into a preset churn prediction model to predict whether the to-be-predicted game user will churn.
9. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the game user churn prediction method according to any one of claims 1-7.
10. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the game user churn prediction method according to any one of claims 1 to 7.
11. A computer program product, comprising at least one executable instruction, wherein the executable instruction enables a processor to execute operations corresponding to the game user churn prediction method according to any one of claims 1 to 7.