Game loss user recall method, system, equipment and medium

Abstract

The invention is suitable for the technical field of computers, and provides a game churn user recall method, which comprises the following steps: analyzing a behavior sequence mode before user churn, and determining a churn type of a churn user in combination with game event data before user churn; based on the churn type, making a personalized touch strategy for the churn user for sending periodic touch content to the churn user; whether the lost user triggers the in-game backflow activity or not is judged, a personalized backflow gift bag is configured for the backflow user, the recall efficiency and effect of the lost user can be improved, and powerful technical support is provided for long-line operation of the game.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, system, device, and medium for recalling lost game users. Background Art

[0002] Throughout the long-term operation of a game, users, as the core asset of the gaming ecosystem, are of undeniable importance. From pre-launch user testing and pre-registration stages to post-launch user acquisition and promotional activities, game developers are constantly iterating and optimizing their operational strategies to acquire more and higher-quality users. However, with increasingly fierce competition in the game acquisition market, the difficulty and cost of user acquisition have increased significantly, making user maintenance, especially in the long-term operation of the game in the later stages, particularly important.

[0003] Existing user recall methods and strategies often lack specificity and personalization, making it difficult to accurately reach lapsed users and stimulate their return. For example, traditional recall methods might simply attract users through unified push notifications or promotional offers, but these methods often overlook the specific reasons for user churn and their individual needs, resulting in poor recall effectiveness. Furthermore, some recall solutions lack automation and regularity, making them difficult to sustain over the long term of a game's operations.

[0004] In order to solve the above problems, this application proposes a method for recalling lost game users, aiming to improve the recall efficiency and effect, and provide strong technical support for the long-term operation of the game. Summary of the Invention

[0005] The embodiments of the present application provide a method, system, device, and medium for recalling lost game users, which can solve one of the above-mentioned problems in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for recalling lost game users, comprising: Analyze the behavioral sequence patterns before user churn and combine them with the game event data before user churn to determine the churn type of churned users; Based on the churn type, a personalized reach strategy is developed for the churned users, which is used to send periodic reach content to the churned users; Determine whether churned users trigger in-game return activities and configure personalized return gift packages for returning users.

[0007] Furthermore, analyzing the behavioral sequence patterns before user churn and combining them with the game event data before user churn to determine the churn type of the churned user includes: Extract the behavior sequences within a preset time period before user churn from the game logs, arrange each behavior sequence in timestamp order, and generate a behavior sequence table; The behavior sequence table is input into a pre-built deep learning model to output the churn contribution and churn type of each game behavior event.

[0008] Furthermore, after determining the churn type of churned users, the following steps are also included: We use each gaming behavior event as a node and the co-occurrence relationship between them as an edge to construct an event influence graph. We also calculate the edge weights based on the churn contribution output by the deep learning model. The edge weights are associated with corresponding game behavior events to determine game behavior events that are strongly correlated with churn.

[0009] Furthermore, based on the churn type, a personalized reach strategy is formulated for the churned users, for sending periodic reach content to the churned users, including: Constructing a profile of churned users based on their user characteristics, including gender, age, game level, preference pattern, churn days, churn type, and historical achievements; Based on the churned user profile, select a content strategy from the game content library, wherein the content strategy includes recall copywriting and material types; Select key data from the user characteristics and insert it into the recall copy, and generate contact content by combining the material type; Collect user behavior data, extract channel feature data, and combine it with the pre-trained channel preference model to generate channel prediction results; Based on the channel prediction results, the reach content is periodically sent to the lost users.

[0010] Furthermore, the periodic sending of the contact content to the churned users includes: Analyze the behavioral sequence patterns before user churn to obtain the activity decay degree, and combine operational events and user characteristics to generate a context vector; By combining the context vector with a multi-task learning model, the model prediction results are output, including the predicted user response probability and the optimal recall period; Determine the final reach cycle based on model prediction results and business rules.

[0011] Furthermore, configuring a personalized return gift package for returning users includes: Calculate the user value score based on the game level, game activity and social influence of the lost users; Based on the user value points, the user's value level is determined, and the user's return gift package is determined according to the value level.

[0012] Furthermore, the method further comprises: When a churned user triggers a return activity in the game, the user ID is marked for the churned user, and the triggering timestamp, triggering channel, and return gift package configuration are recorded to generate a return user record table; Calculate the retention and payment status of each returning user in the returning user record table within the preset time period; Based on the reach cost and gift package cost, calculate the recall cost of returning users, and combine the retention and payment status to calculate the revenue of returning users.

[0013] In a second aspect, an embodiment of the present application provides a system for recalling lost game users, including: The first processing module is used to analyze the behavioral sequence patterns before user churn and determine the churn type of churned users by combining the game event data before user churn; The second processing module is used to formulate a personalized contact strategy for the churned users based on the churn type, and to send periodic contact content to the churned users; The third processing module is used to determine whether the lost users trigger the return activity in the game, and configure personalized return gift packages for the returning users.

[0014] 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, wherein the processor implements the above-mentioned method for recalling lost game users when executing the computer program.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, comprising a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned method for recalling lost game users is implemented.

[0016] Compared with the prior art, the embodiments of the present application have the following beneficial effects: This application discloses a method for recalling lost game users. By analyzing the behavioral sequence patterns and game event data before user loss, it can accurately identify the specific reasons for user loss and then determine the corresponding loss type, thereby providing a solid foundation for subsequent personalized reach strategies. Furthermore, based on the user's loss type, formulating a personalized reach strategy can effectively stimulate user interest and increase the return rate. At the same time, according to the user's loss type and behavioral characteristics, the content and value of the return gift package can be dynamically adjusted, which can maximize the attractiveness of the return gift package and improve user retention rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 This is a flow chart of a method for recalling lost game users provided by one embodiment of the present invention; Figure 2 This is a structural diagram of a system for recalling lost game users provided by one embodiment of the present invention; Figure 3 It is a structural diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0020] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0021] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0022] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0023] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0024] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0025] See also Figure 1 As shown, the present invention is a method for recalling lost game users, comprising the following steps: S100, analyzing the behavioral sequence pattern of users before churn, and combining the game event data before churn to determine the churn type of churned users; By analyzing the behavioral sequence patterns and game event data before user churn, this application can accurately identify the specific reasons for user churn and then determine the corresponding churn type, thereby providing a solid foundation for subsequent personalized reach strategies.

[0026] In some embodiments, step S100 includes: Extract the behavior sequences within a preset time period before user churn from the game logs, arrange each behavior sequence in timestamp order, and generate a behavior sequence table; The behavior sequence table is input into a pre-built deep learning model to output the churn contribution and churn type of each game behavior event.

[0027] In one embodiment, the behavior sequences of the user in the 30 days before churn are extracted from the game logs and sorted in timestamp order to generate a behavior sequence table, such as login behavior, game behavior, and purchase behavior. At the same time, the behavior sequences in the behavior sequence table are statistically analyzed for time series features, such as using a sliding window to calculate time series features within a preset time period, such as the copy failure rate in the past 7 days, the standard deviation of daily login duration, etc.

[0028] In this embodiment, each behavior in the behavior sequence table is labeled as an event to facilitate the subsequent processing of game behavior events that are strongly related to churn. Specifically, each type of behavior is labeled as general events, frustration events, social events, payment events, game events, etc., among which general events are basic game behaviors that are not directly related to core gameplay or emotional fluctuations, specifically daily game tasks, login or exit events, etc. Frustration events are specifically behaviors that will cause negative emotions to players, usually failure to pass levels, resource loss, etc. Social events are behaviors involving multi-person interaction, which can reflect the social stickiness of players. Game events are behaviors related to core gameplay, which are used to reflect the depth of player investment, generally including copy clearance, skill upgrades, equipment enhancement, achievement achievement, etc.

[0029] In this embodiment, the behavior sequence list is input into the deep learning model, and the deep learning model is used to obtain the churn contribution of each game behavior and the player's churn type. Specifically, the deep learning model includes a shared feature extraction layer. When each game behavior event is input, it can process the behavior sequence in both directions and capture the dependency relationship between the previous and subsequent behaviors. Furthermore, the deep learning model can calculate the contribution of each game behavior event to churn and the user's churn type. Specific churn types include frustration type, social disconnection type, payment burnout type, gameplay fatigue type, etc.

[0030] In one embodiment, during the construction of the deep learning model, the original behavior sequences of different game players are obtained and converted into structured sequences, and each behavior in the structured sequence is subjected to event labeling, and the time difference between adjacent behaviors is calculated as an event feature. At the same time, the game players are subjected to sample labeling. In one possible embodiment, players who have not logged in for 7 consecutive days are used as positive samples of the deep learning model and are marked as churned users, while players who have logged in for 3 consecutive days are used as negative samples of the deep learning model and are marked as active users. This is used to ensure the comprehensiveness of the deep learning model training data, and the churn types are labeled at the same time. In this embodiment, they are specifically frustration type, social disconnection type, payment fatigue type, and gameplay fatigue type. Specifically, a behavior ID is set for each game behavior event in the structured sequence, and the discrete behavior ID is converted into a continuous vector to capture the semantic relationship between behaviors. For example, the words "failure" and "victory" are mapped to vectors, so that the model understands that "failure" is closer to "frustration" rather than "login", and then labels each game behavior event. Furthermore, a bidirectional LSTM network is used to analyze the dependencies between the behavior sequences and capture long-term patterns. Specifically, the sequence is analyzed from left to right and from right to left at the same time. From left to right, it means that historical behavior affects current behavior, and from right to left, it means that current behavior affects future behavior. Specifically, If "social help-seeking" behavior is not triggered after "level failure", the churn risk increases to a certain extent. This is used to measure the contribution of each game behavior event to churn, and then different weights are assigned to each game behavior event through the attention mechanism to quantify the churn contribution. Furthermore, the churn contribution of each game behavior event is independently calculated through the Sigmoid activation function, and the distribution of churn types is dynamically adjusted through the Softmax activation function. Therefore, the multi-task learning requirements of the deep learning model are realized through the two activation functions, specifically the churn contribution of each game behavior event and the determination of the churn type.

[0031] Specifically, the churn contribution of each game behavior event is calculated independently. In this process, the output of each game behavior event is independent and does not affect each other. Specifically, after each game behavior data passes through the hypothesis model, a bidirectional LSTM network and attention mechanism, a shared feature vector z∈R^d is obtained, which represents the abstract features extracted by the deep learning model from the input game behavior event. Among them, d is the feature dimension, which is measured by the amount of input data of the deep learning model, usually 64 dimensions, 128 dimensions, or 256 dimensions. For n behavior labels, the output layer performs linear calculations on each behavior label i to obtain a raw score: s i =W i ×z+b i , where W i Represents the weight vector of behavior label i, bi Represents the bias term. Further, the original scores of all behavior labels are converted into probability distributions through the Sigmoid activation function, and the output is p i ∈(0,1), represents the probability of behavior label i contributing to churn, i.e., churn contribution. For m churn types, the output layer calculates the original score of each churn type j through linear transformation: s j =W j ×z+b j , where W j represents the weight vector of churn type j, b j Represents the bias term, and then through Softmax activation, the scores of all churn types are converted into probability distributions, and the output is p j ∈(0,1), and , and finally the churned users are clearly classified into the most likely churn type through this probability distribution.

[0032] In some embodiments, after determining the churn type of the churned user, the method further includes: We use each gaming behavior event as a node and the co-occurrence relationship between them as an edge to construct an event influence graph. We also calculate the edge weights based on the churn contribution output by the deep learning model. The edge weights are associated with corresponding game behavior events to determine game behavior events that are strongly correlated with churn.

[0033] In this embodiment, after the behavior sequence list is input into the deep learning model, the deep learning model outputs each game behavior event after event labeling and the contribution degree of the corresponding game behavior event to churn. Furthermore, each game behavior event is used as a node, and the co-occurrence relationship between game behavior events is used as an edge. At the same time, the edge weight between two game behavior events is calculated based on the corresponding churn contribution. Specifically, the edge weight = P1 + P2 − P1 × P2 × correlation coefficient, where P1 and P2 respectively represent the churn contribution degrees of the two game behavior events, and the correlation coefficient is specifically the correlation between the two game behavior events obtained by calculating historical data.

[0034] In this embodiment, the edge weight is associated with the corresponding game behavior event. Specifically, if the edge weight between two game behavior events is greater than a preset threshold, it means that the correlation between the two game behavior events is strong. Then, it is associated with the user's churn type and the two game behavior events are marked as game behavior events with strongly correlated events of the churn type.

[0035] S200: Developing a personalized reach strategy for the churned user based on the churn type, for sending periodic reach content to the churned user; In this embodiment, formulating a personalized reach strategy based on the user's churn type can effectively stimulate user interest and increase the return rate.

[0036] In some embodiments, step S200 includes: Constructing a profile of churned users based on their user characteristics, including gender, age, game level, preference pattern, churn days, churn type, and historical achievements; Based on the churned user profile, select a content strategy from the game content library, wherein the content strategy includes recall copywriting and material types; Select key data from the user characteristics and insert it into the recall copy, and generate contact content by combining the material type; Collect user behavior data, extract channel feature data, and combine it with the pre-trained channel preference model to generate channel prediction results; Based on the channel prediction results, the reach content is periodically sent to the lost users.

[0037] In this embodiment, user characteristics of churned users are obtained, and based on this, a churned user profile is constructed to facilitate the subsequent generation of personalized reach strategies. Specifically, user characteristics include static characteristics and dynamic characteristics, wherein static characteristics include basic user data such as name, gender, age, region, login device type, and social relationship data, and dynamic characteristics include game level, preference mode, churn days, churn type, historical achievements, payment data, and other data related to game behavior. Furthermore, based on the above user characteristics, a portrait label system is constructed for the churned users, which specifically includes basic attributes, game ability, behavioral preferences, churn type, and value level. The basic attributes are specifically user age, gender, and device type. Game ability is measured by historical achievements, game level, etc., and the user's game ability is determined to be novice, beginner, intermediate, or advanced, etc. The behavioral preference is specifically a game preference mode, such as PVP enthusiast, social type, or single-line action type, etc. The churn type is specifically the churn type obtained by the deep learning model in step S100, and the value level is quantified by game level, game activity, and social influence.

[0038] Furthermore, based on the churned user portrait obtained above, a suitable content strategy is selected from the game content library, such as recall copy and material type. Specifically, multimodal materials such as pictures, texts, videos, etc. are stored in the game content library, from which the material type that matches the churned user portrait is selected. Specifically, the CLIP model is used to calculate the matching degree between each portrait label and the material in the churned user portrait, and the material with the highest similarity is selected, such as matching the most relevant game image or video according to the user behavior preference label. Furthermore, according to the user's churn type, a recall copy with a higher matching degree is selected, and key data such as user name, historical achievements, etc. are selected from the user characteristics, inserted into the corresponding recall copy, and combined with the material to generate the reach content.

[0039] Specifically, the churned user portrait is converted into a structured label list, and then the structured label list is further converted into a label vector, for example, churned user A = ["advanced player", "PVP enthusiast", "frustrated"]. In addition, each material type in the game content library is annotated with material attributes, specifically associated user features, effect data and multimodal embedding, where the associated user features are specifically features associated with the user's portrait label, such as a possible material type is associated with "PVP" and "novice", and the effect data are specifically the historical click-through rate, conversion rate and applicable user churn type of the material type, such as "suitable for frustrated users". For multimodal embedding, it is specifically the text and visual joint feature vector pre-extracted by the CLIP model for subsequent similarity calculation. Further specifically, the label vector of the churned user is converted into natural language, and the above churned user A is converted into self- Then the language is followed by "an advanced player who prefers PVP mode and is a frustrated user", and the similarity is calculated between it and the multimodal feature vector of each material type to determine the material type. Furthermore, based on the user's churn type, the recall copy is determined. It can be understood that each recall copy in the game content library is associated with the churn type, such as "

King Exclusive

[0040] In this embodiment, by predicting the user's preference for different channels, the content delivery channels are optimized and the content is sent to the relevant devices of the lost users. Specifically, the user's behavior data on the relevant devices is collected in real time, and features such as device, time period, and geographic location are extracted. The relevant features are input into the pre-trained channel preference model to generate channel prediction results, and the channel prediction results are cached for channel decision service calls.

[0041] Specifically, the user's login device type is obtained from the user's game log, such as Android mobile terminal, Apple mobile terminal, PC client, web terminal, etc., and user behavior events are reported in real time through the game client SDK or Web embedding point. Flume / Kafka is used to collect server logs and filter channel-related behaviors, such as Push notification clicks, SMS link access, etc. In some embodiments, the user's network type and device performance during the game are also obtained, specifically through the CPU / GPU occupancy conversion collected by the user client SDK. In addition, the user's game time period characteristics are obtained, that is, the user's game active time period. The peak period of behavior statistics before user loss can be used as the game active time period. For geographical location characteristics The feature is mapped through the geographic location IP to determine the city where the user is located, such as Beijing, Shanghai, etc., and different cities are encoded. Then, the above features are input into the channel preference model. It can be understood that the channel preference model specifically uses the device information, login time, geographic location and channel information of different game users when they first log in to the game in history as training data to generate a model that can predict the user's channel preference. In a possible embodiment, the training data is selected from new users who logged into the game for the first time in the past 6 months, and the data is stratified according to channel distribution, device type, login time and geographic location. The channel distribution specifically includes all contact channels such as Push, SMS, Email, and pop-up windows in the game.

[0042] In some embodiments, periodically sending the reaching content to churned users includes: Analyze the behavioral sequence patterns before user churn to obtain the activity decay degree, and combine operational events and user characteristics to generate a context vector; By combining the context vector with a multi-task learning model, the model prediction results are output, including the predicted user response probability and the optimal recall period; Determine the final reach cycle based on model prediction results and business rules.

[0043] In this embodiment, by real-time analysis of user behavior before churn, combined with in-game operational events and user characteristics, the optimal recall cycle is dynamically predicted, thereby optimizing user reach strategies, improving recall effects, and reducing negative feedback.

[0044] In one embodiment, login events before user churn are selected, and the user's activity decay is measured by the login events. Specifically, all login events within one month before user churn are selected and marked with corresponding login timestamps to generate a login behavior sequence. The login behavior sequence is divided into multiple login sequences, and the login frequency of the user in each login sequence is calculated. For example, in one embodiment, the login behavior sequence within one month before user churn is divided into natural weeks, specifically the login sequence of the first week, the login sequence of the second week, the login sequence of the third week, and the login sequence of the fourth week. Furthermore, the corresponding activity decay is obtained by the activity decay calculation formula. .

[0045] In this embodiment, the activity decay degree is combined with the operational events before user churn to reflect the external triggers of churn, such as insufficient activity rewards leading to user disappointment, where the operational events are specifically version updates or activity launches, etc., and then combined with user characteristics to generate multi-dimensional contextual information. By fusing the above-mentioned features and using a multi-task learning model to learn the associations between the features, the accuracy, explainability and operational decision-making efficiency of churn prediction are improved.

[0046] In this embodiment, the multi-task learning model is used to predict the user response probability and the optimal recall period. Specifically, the activity decay degree and the historical response rate to the corresponding channel in the user characteristics are combined to predict the user's acceptance probability of the recall strategy. When recalling users, priority is given to recalling users with a high response probability to improve the recall ROI and avoid sending push notifications to users who have completely lost their users. In addition, the recall period is determined by combining the game active period and activity decay degree in the user characteristics.

[0047] Specifically, the above-mentioned active decay, historical response rate and game active period are feature processed. Specifically, the active decay range [0,1] is mapped to [0,10] points. The faster the decay, the higher the score. For the historical response rate, it is divided into three levels according to the quantile: low (0-0.3), medium (0.3-0.7), and high (0.7-1). The 24 hours are divided into multiple periods, such as 0-6 o'clock, 6-12 o'clock, etc., and the main active periods of users are represented by One-Hot encoding. Then, the active decay degree × historical response rate are feature combined, and the active decay degree is feature combined with the active period. Through the multi-task learning model, the user response probability and the optimal recall period are predicted. In the multi-task learning model, the model is constructed with the task goal of predicting the user's acceptance probability of the recall strategy and the optimal recall period.

[0048] Furthermore, the model prediction results are converted into executable reach strategies, and the cycle is optimized in combination with business rules. Specifically, if the user response probability is greater than the preset threshold, the generation of the recall strategy is triggered, and then the optimal recall cycle is dynamically adjusted in combination with the business rules to generate the final reach cycle, and the corresponding reach content is sent to the lost users. For example, in one possible embodiment, the business rule is that when the lost user is a high-value user and the response probability is greater than 0.7, the optimal recall cycle is shortened by 2 days. If a new version is released, the global optimal recall cycle is shortened by 30%. Thus, the corresponding final reach cycle is determined.

[0049] S300: Determine whether the lost user triggers a return activity in the game, and configure a personalized return gift package for the returning user.

[0050] In this embodiment, after receiving the reflow contact content, the user logs in to the game again. Based on whether there is a reflow contact time and status record, it is determined whether the in-game reflow activity is triggered. If so, the gift package is issued. Specifically, the user login data is obtained through the game log. Combined with the reflow contact related records, the user's contact time and status record information is extracted to determine whether the user is a first-time contact. If it is a first-time contact, the corresponding reflow gift package is configured for the reflow user based on information such as the user's value level. If the user is not a first-time contact, the previous contact time and status record are obtained to determine whether the contact time meets the preset reflow gift package triggering conditions. If the contact time exists and the status record meets the preset conditions, it is determined that the user meets the conditions for triggering the reflow activity. Then, the corresponding gift package distribution configuration information is obtained for the user, the gift package content and distribution method are determined, and the gift package distribution configuration information is transmitted to the game client. The display status and distribution status feedback from the client are obtained to determine whether the execution is successful. If an exception occurs during the execution process, the log is updated based on the feedback status record, and the gift package distribution information is re-pushed using the backup mechanism to obtain the final execution result. By analyzing the final execution results, user behavior data and status records are updated to form a closed-loop data flow for subsequent return flow and dynamic adjustment of activity conditions.

[0051] Furthermore, dynamically adjusting the content and value of the return gift package based on the user's churn type and behavioral characteristics can maximize the attractiveness of the return gift package and improve user retention rate.

[0052] In some embodiments, configuring a personalized return gift package for a returning user includes: Calculate the user value score based on the game level, game activity and social influence of the lost users; Determine the user's value level based on the user's value score, and determine the user's return gift package based on the value level; Analyze the behavioral sequence pattern before user churn, combine it with user characteristics, determine the severity of user churn, and dynamically adjust the distribution strategy of the return gift package.

[0053] In this embodiment, according to the value level in the lost user portrait generated in step S200, a personalized return gift package is configured for the returning user, thereby improving the recall effect.

[0054] In this embodiment, the user value level is quantified by the game level, game activity and social influence of the lost user. Specifically, the user value score = (LTV×activity weight)×(1+social influence coefficient)×churn attenuation factor, where LTV represents the game level value. It is measured by predicting the payment situation within a preset time by the historical payment generated when the user reaches the game level. Specifically, the historical payment records of users when they reach a specific game level are sorted to obtain a preliminary user payment data set. Based on the preliminary user payment data set, data analysis technology is used to mine the correlation between historical payment and game level to determine the payment behavior characteristics of users at different levels. If the historical payment record of the user at a certain game level exceeds the preset threshold, then Mark them as users with high payment tendency, obtain their payment trend information, process the payment trend information of users with high payment tendency, combine the behavior patterns in the preset time period, use the random forest model to predict the payment behavior in the time period, and obtain the prediction results. For game activity, measure it by calculating the active weight value, where active weight = 1 + (average daily online time / standard deviation) × 0.2. Social influence is measured by the user's position in the game, chat frequency, etc. Specifically, the total social influence score = α × position influence + β × chat frequency influence. Assign a value to the user's position in the game to obtain the position influence As for the influence of chat frequency, we obtain the speech records and interaction records of lost users in the period before they lost, and perform deduplication and filtering on the corresponding speech records and interaction records. For example, if the interval between consecutive speeches of the same user in the same channel is less than 5 seconds, they are merged into 1 time to avoid screen interference, and delete pure expressions, single words, such as "oh", "good" and other meaningless speeches. Furthermore, we count the chat frequency in the past period of time, sort the total number of speeches of all users, divide the intervals by quantiles, and determine the chat frequency score of lost users. Specifically, different quantiles correspond to different chat frequency scores, such as a possible In the embodiment, when the quantile is ≥ the 80th percentile, the chat frequency score is 80; when the chat frequency score is between the 50th and 79th percentiles, the chat frequency score is 50; and when the quantile is < the 50th percentile, the chat frequency score is 10. Furthermore, the chat frequency score is modified based on the interaction quality of churned users, such as the proportion of replies to speeches by other users, which reflects the attractiveness of the user's chat content. Specifically, for each speech, the number of IDs to which it is replied is counted, and the reply rate = the number of replied speeches / the total number of speeches. The chat frequency influence = the chat frequency score × the reply rate, and the churn attenuation factor = e^(-the number of churn days × 0.1) In this embodiment, different user value scores are divided into multiple value levels based on a preset stratification strategy. At the same time, different return gift packages are pre-allocated for different value levels. In a possible embodiment, when the user value score is greater than 90 points, it is divided into S level, and the corresponding return gift package is "rare props + exclusive title + double currency". When the user value score is in the range of 70-89 points, the corresponding value level is A level, and the corresponding return gift package is "premium gift package + limited-time discount coupon". When the user value score is in the range of 50-69 points, the corresponding value level is B level, and the corresponding return gift package is "basic currency + ordinary props". When the user value score is less than 50 points, the corresponding value level is C level, and the return gift package is only basic currency. Based on this, according to the user value score of the lost user, the return gift package of the user's desired method can be determined.

[0055] In some embodiments, the method further comprises: When a churned user triggers a return activity in the game, the user ID is marked for the churned user, and the triggering timestamp, triggering channel, and return gift package configuration are recorded to generate a return user record table; Calculate the retention and payment status of each returning user in the returning user record table within the preset time period; Based on the reach cost and gift package cost, calculate the recall cost of returning users, and combine the retention and payment status to calculate the revenue of returning users.

[0056] In this embodiment, when a user clicks on a return activity link via SMS, email, or Push notification, the user ID of the churned user is marked, and the corresponding trigger timestamp, trigger channel, and return gift package configuration are recorded to generate a return user record table. During this process, Kafka is used to receive the trigger event, which is then cleaned by Flink and stored in the return user record table partitioned by user ID user_id for subsequent analysis.

[0057] In this embodiment, a retention analysis window and a payment analysis window are set. Specifically, the retention analysis window is used to analyze the retention situation, and the payment analysis window is used to analyze the payment situation. In a possible embodiment, the retention analysis windows are D1, D3, D7, D15, and D30, respectively, and specifically analyze the user's login status in the game on the second day, third day, seventh day, fifteenth day, and thirtieth day after returning, while the payment analysis window is specifically for the first week of user return and the second month after return.

[0058] Furthermore, based on the aforementioned retention analysis window, the number of logins by returning users within each time window is counted to determine the retention level of returning users within a preset time period. Specifically, the login data of returning users is obtained, and the user's in-game behavior records are compiled for each time point within the analysis window. By counting the login frequencies one by one, the initial retention performance of returning users within different time periods is determined. Based on the initial retention performance, the login data within the time period is stratified and classified using a preset threshold. If the login frequency at a certain time point is below the preset threshold, the user is marked as a low-activity user group; if it is above or equal to the preset threshold, the user is marked as a high-activity user group, thereby obtaining the user's activity level classification result. Based on the activity level classification result and the specific operational data in the behavior records, the user's behavior patterns at each time point are analyzed to determine the distribution characteristics of the user's activity level within different time periods. Based on the activity level distribution characteristics, the user's behavior change data within consecutive time periods is extracted, and the changing trends of the user at different time points are analyzed to obtain a quantitative indicator of the change trend. For the quantitative indicators of changing trends, the time series analysis method is used to compare the changes in user retention performance on a period-by-period basis. If the trend indicator of changing trends shows a decline for multiple consecutive time periods, and the decline exceeds the preset range, it is marked as a continuous decline signal, and the risk level of user retention performance is determined, which are high risk level, medium risk level, and low risk level.

[0059] Furthermore, based on the above payment analysis window, the payment situation of returning users in each time window is counted. Specifically, for returning users with medium risk level and low risk level, it means that the probability of user retention in a short period of time is high, and for high risk level, it means that the probability of user retention is low. When calculating the revenue, the calculation weight is dynamically adjusted based on the payment situation in the corresponding time window to obtain the recall cost. Specifically, different calculation weights are configured for risk levels. In one possible embodiment, the high risk level is 0.5. Due to its retention probability and cost sensitivity, the weight for the medium risk level is 0.7, and for the low risk level Its weight is 1.0, and the overall payment situation is calculated based on the payment situation of retained users in the first week and the next month. In one possible embodiment, there is a retained user A with a high risk level. He pays 200 yuan in the first week and 10 yuan in the next month. The overall payment situation = (200+10)×0.5=105 yuan. In some other embodiments, the time decay coefficient is introduced for calculation of the payment situation in different time windows, indicating that the user's payment behavior decreases linearly over time. In one possible embodiment, the time decay coefficient for the first week is 1, and the time decay coefficient for the next month is 0.7. The overall payment situation = (200×1.0×0.5 + 10×0.7×0.5)=103.5 yuan.

[0060] Furthermore, the recall cost is calculated based on the gift package cost in the return gift package configuration and the reach cost during the reach process, that is, recall cost = gift package cost + reach cost. Combined with the overall payment situation, the revenue of the returning users is calculated. Specifically, revenue = overall payment situation - recall cost. In this way, the corresponding return effect can be quantitatively presented, which is convenient for game developers to conduct subsequent user return analysis and strategy adjustments.

[0061] See also Figure 2 As shown, the present invention also provides a game lost user recall system, the system comprising: The first processing module 201 is used to analyze the behavioral sequence pattern of users before churn, and determine the churn type of churned users by combining the game event data before churn; The second processing module 202 is configured to formulate a personalized reach strategy for the churned users based on the churn type, and to send periodic reach content to the churned users; The third processing module 203 is used to determine whether the lost user triggers the return activity in the game, and configure a personalized return gift package for the returning user.

[0062] It is understandable that if Figure 1 The contents of the game lost user recall method embodiment shown in the figure are applicable to the game lost user recall system embodiment. The functions specifically implemented by the game lost user recall system embodiment are similar to those in the example above. Figure 1 The embodiment of the method for recalling lost game users is the same as that shown in FIG. Figure 1 The beneficial effects achieved by the embodiment of the method for recalling lost game users are also the same.

[0063] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0064] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0065] See also Figure 3 As shown, an embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, a method for recalling lost game users as described in any one of the above methods is implemented.

[0066] The computer device 3 may be a desktop computer, a notebook computer, a PDA, a cloud server or other computing devices. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that Figure 3 This is merely an example of the computer device 3 and does not constitute a limitation on the computer device 3 . The computer device 3 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 3 may also include input and output devices, network access devices, etc.

[0067] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0068] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard drive or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the computer device 3. Furthermore, the memory 302 may include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is about to be output.

[0069] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for recalling lost game users as described in any one of the above methods is implemented.

[0070] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a camera / computer device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0071] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A method for recalling lost game users, characterized in that: include: Analyze the behavioral sequence patterns before user churn and combine them with the game event data before user churn to determine the churn type of churned users; Based on the churn type, a personalized reach strategy is developed for the churned users, which is used to send periodic reach content to the churned users; Determine whether churned users trigger in-game return activities and configure personalized return gift packages for returning users.

2. The method according to claim 1, wherein The analysis of the behavioral sequence pattern before user churn, combined with the game event data before user churn, to determine the churn type of the churned user includes: Extract the behavior sequences within a preset time period before user churn from the game logs, arrange each behavior sequence in timestamp order, and generate a behavior sequence table; The behavior sequence table is input into a pre-built deep learning model to output the churn contribution and churn type of each game behavior event.

3. The method according to claim 2, wherein After determining the churn type of churned users, also include: We use each gaming behavior event as a node and the co-occurrence relationship between them as an edge to construct an event influence graph. We also calculate the edge weights based on the churn contribution output by the deep learning model. The edge weights are associated with corresponding game behavior events to determine game behavior events that are strongly correlated with churn.

4. The method according to claim 1, wherein The method of formulating a personalized reach strategy for churned users based on the churn type, and sending periodic reach content to the churned users, includes: Constructing a profile of churned users based on their user characteristics, including gender, age, game level, preference pattern, churn days, churn type, and historical achievements; Based on the churned user profile, select a content strategy from the game content library, wherein the content strategy includes recall copywriting and material types; Select key data from the user characteristics and insert it into the recall copy, and generate contact content by combining the material type; Collect user behavior data, extract channel feature data, and combine it with the pre-trained channel preference model to generate channel prediction results; Based on the channel prediction results, the reach content is periodically sent to the lost users.

5. The method according to claim 4, wherein The periodically sending the reaching content to the lost users includes: Analyze the behavioral sequence patterns before user churn to obtain the activity decay degree, and combine operational events and user characteristics to generate a context vector; By combining the context vector with a multi-task learning model, the model prediction results are output, including the predicted user response probability and the optimal recall period; Determine the final reach cycle based on model prediction results and business rules.

6. The method according to claim 1, wherein The configuration of a personalized return gift package for returning users includes: Calculate the user value score based on the game level, game activity and social influence of the lost users; Based on the user value points, the user's value level is determined, and the user's return gift package is determined according to the value level.

7. The method according to claim 1, wherein The method further comprises: When a churned user triggers a return activity in the game, the user ID is marked for the churned user, and the triggering timestamp, triggering channel, and return gift package configuration are recorded to generate a return user record table; Calculate the retention and payment status of each returning user in the returning user record table within the preset time period; Based on the reach cost and gift package cost, calculate the recall cost of returning users, and combine the retention and payment status to calculate the revenue of returning users.

8. A game lost user recall system, characterized by: include: The first processing module is used to analyze the behavioral sequence patterns before user churn and determine the churn type of churned users by combining the game event data before user churn; The second processing module is used to formulate a personalized contact strategy for the churned users based on the churn type, and to send periodic contact content to the churned users; The third processing module is used to determine whether the lost users trigger the return activity in the game, and configure personalized return gift packages for the returning users.

9. 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 method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images