Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, the "plurality" generally includes at least two.
The words "if", as used herein, may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to a detection", depending on the context. Similarly, the phrase "if determined" or "if detected (stated condition or event)" may be interpreted as "when determined" or "in response to determination" or "when detected (stated condition or event)" or "in response to detection (stated condition or event), depending on the context.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a product or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such product or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a commodity or system comprising such elements.
In addition, the sequence of steps in the method embodiments described below is only an example and is not strictly limited.
The user recall method provided by the embodiment of the invention can be executed by an electronic device, and the electronic device can be a terminal device such as a PC, a notebook computer, a smart phone and the like, and can also be a server. The server may be a physical server comprising an independent host, or may be a virtual server carried by a host cluster, or may be a cloud server.
In practical application, the user recall method provided by the embodiment of the invention can be applied to any game, such as open world games, shooting games, player recall scenes in card games, and the like.
In fact, in different games, the setting modes of recall targets are different in order to adapt to the characteristics of different games. For example, recall targets are set in open world games based on data such as game copy checkpoints, recall targets are set in shooting games based on weapon use, and recall targets are set in card games based on card selection preferences. The recall target setting bases are examples, and the recall target setting bases of the game in actual use can also be different.
The following describes the execution of the user recall method in conjunction with the following embodiments.
At present, a data analysis method is generally adopted to obtain all the losing users, and then recall information is sent to all the losing users according to the contact modes such as mobile phone numbers, mailboxes and the like, so that the losing users log in the game again through rewards or new function introduction in the recall information. However, this method has high recall cost, such as huge short message fee, rewarding fee, etc., and lacks pertinence, and has poor recall effect.
For this reason, in the related art, it is also possible to select paid users, high-level users, from among all the attrition users based on recall policies set by the technician, and send recall information to these users. Although the method improves the pertinence of the recall of the user, the requirements on the capability and experience of technicians are higher, some recall-capable users are easy to miss, and the recall effect is difficult to guarantee.
In view of at least one of the above-mentioned technical problems, the following describes the execution of the user recall method provided herein in connection with the following embodiments.
Fig. 1 is a flowchart of a user recall method according to an embodiment of the present invention, as shown in fig. 1, where the method includes the following steps:
101. and obtaining game data of a plurality of users, wherein the users meeting the recall target in the plurality of users are taken as reference users, and the users meeting the reference target and not meeting the recall target in the plurality of users are taken as users to be analyzed.
102. And inputting the game data into a recall analysis model, so that the recall analysis model calculates the user to be analyzed based on the reference user, and the similarity of the user to be analyzed and the reference user in game performance is obtained.
103. And determining the user to be analyzed, the similarity of which meets the preset condition, as a target recall user, and pushing game resources corresponding to the recall target to the target recall user.
In the method provided by the figure 1, the similarity of the user to be analyzed and the reference user in game performance is analyzed through the recall analysis model, and the recall potential of the user to be analyzed is judged based on the similarity, so that the automatic screening of the target recall user is realized, the user with larger recall potential is avoided from missing without relying on artificial experience analysis, the recall efficiency is greatly improved, the scale of the target recall user can be controlled, and the recall cost is reduced. Meanwhile, the method also enables the target recall user under different recall targets to acquire more interesting game resources by pushing game resources corresponding to the recall targets to the target recall user, and improves recall effects.
Specific implementations of the individual steps are described below in connection with specific examples.
First, game data of a plurality of users is acquired in 101. Since the plurality of users includes users who remain in the game (i.e., reference users described below) and users who do not remain in the game (i.e., churner users described below), preliminary division of the user group is achieved by recall targets and reference targets among the plurality of users. Optionally, the recall target and the reference target are determined according to a judgment condition of the churn user. For example, if a user who has not logged in to the game for 7 days, or has not completed a specified game performance for 7 days, is determined to be a churn user, recall targets and reference targets may be formulated based on the time of the game logging or execution of the specified game performance (e.g., completion of a task). For example, the recall target may be a user who is logged in for 15 days and completes a specified game performance, and the reference target may be a user who is logged in for 3 days or completes a specified game performance.
Specifically, a reference user is screened out through a recall target, and game data of the reference user is the reference standard in the recall process. In practice, recall targets may be set according to one or a combination of the following criteria, including but not limited to: user level, login days, number of friends, friend activity, recharge status, game system data, prop use status, weapon use status, battle change status, duplicate checkpoints. The game system is, for example, a pet system, a face pinching system, a combat system, a copy task system, and a collection system. For example, the game system data is, for example, the use cases of various game systems. Specifically, the gaming system data includes, but is not limited to, the following examples: the number of pets in the pet system, the number of pet culture values, the pet attributes, the task access condition, the completion progress, the task success rate and the task failure rate in the task system, and the user arrived area, the user arrival frequency, the user preference route and the user preference place type in the map system.
For example, the recall target is set to log in for 7 days and reach level 40, whereby the attrition users with the potential to reach a higher level are selected by the recall target. Or the recall target is set to log in for 7 days, the number of friends exceeds 10 and the charge value exceeds 100 yuan, so that the lost user with the potential of achieving higher activity is selected through the recall target. Optionally, to screen out the lost users with more recall potential, the lost users are classified by setting recall targets. In practice, recall targets can be adjusted iteratively to optimize selected target recall users, so that recall effects are further improved.
By way of example, assuming that the recall target is a player rating of 40, this means that during this recall, the user whose player rating is 40 belongs to the user who meets the recall target.
To facilitate recall analysis, optionally, a minimum login interval is also set in the recall target that should be reached by the surviving user. For example, the minimum login interval is set to 3 days, and then a user who has logged in the game within 3 days is considered to be a retention user, and a user who has not logged in the game for 3 or more days is considered to be a loss user.
It should be noted that the churn user refers to a user who does not remain in the game. In short, the lost user is a user whose time difference between the last login time and the current time exceeds a set threshold. For example, a user that has not logged into the game for n days may be determined to be a churn user. Wherein n is a set time threshold, which can be set according to different application scenarios.
Taking the hand-tour field as an example, users who have no active behavior for 7 days are generally referred to as churn users. Where active behavior refers to logging in, or completing a specified game performance (e.g., 30 minutes online, daily punch-card tasks, specified tasks, etc.). Among different games, the designated games behave differently. Alternatively, the relevance of various game performances and the retention user is acquired, and the game performance with the highest relevance is used as the designated game performance for judging the loss user.
In fact, there are also users with low recall potential among the lost users, such as users who have not logged in again after registration, or users with an online time of less than 30 minutes, or users with a ranking below 10, it is evident that these users have not only low recall potential, but also lack game data for analyzing recall potential. In the embodiment of the invention, in order to improve recall efficiency, a reference target for screening the users to be analyzed from all the lost users is further provided, and in addition, 101, the users which meet the reference target and do not meet the recall target are the users to be analyzed. Therefore, the user with lower recall potential (i.e. the user which does not reach the standard condition) in the plurality of users can be eliminated, and the data volume of the game data to be analyzed is reduced, so that the subsequent recall analysis efficiency is improved.
In general, the difficulty of achieving the benchmark is lower than the recall goal, such as the recall goal is set to log in for 7 days and up to level 40, and the benchmark can be set to log in for 3 days and up to level 15. Of course, in practical application, the setting of the reference target is not only referred to the recall target, but also can be set according to specific application occasions, for example, specific game types, play characteristics and the like.
Further, after the reference user and the user to be analyzed are divided, in 102, game data is input into a recall analysis model, so that the recall analysis model calculates the user to be analyzed based on the reference user, and the similarity between the user to be analyzed and the reference user in game performance is obtained.
In short, from the viewpoint of game performance, the higher the similarity degree with the reference user, namely the more similar the behavior of the user to be analyzed and the reference user in the game, the more likely the user to be analyzed reaches the recall target, the greater the recall potential of the user to be analyzed, so that the recall potential of the user to be analyzed can be quantitatively represented through the similarity between the user to be analyzed and the reference user in the game performance, thereby assisting the subsequent screening of target recall users.
The game performance in the present invention includes the user's interaction with the game system. In particular, game performances include, but are not limited to, login behavior, interaction with NPC, interaction with other players, and behavior in various game systems (e.g., combat system, collection system, weapon system, skill system, face pinching system, fostering system, etc.).
The recall analysis model is selected according to the recall requirement of the user. For example, the recall analysis model may be a machine learning model for similarity calculation, such as various similarity models.
Specifically, assuming that the recall analysis model is a machine learning model for similarity calculation, inputting game data into the recall analysis model based on the machine learning model, so that the recall analysis model calculates a user to be analyzed based on a reference user to obtain the similarity between the user to be analyzed and the reference user, and the method can be specifically implemented as follows:
Selecting first game data generated before the reference user meets a reference target from game data of the reference user; selecting third game data generated before the user to be analyzed meets a reference target from game data of the user to be analyzed; and calculating the similarity of the third game data and the first game data through a machine learning model for calculating the similarity, and taking the calculated similarity of the third game data and the first game data as the similarity of the user to be analyzed and the reference user.
In the above step, assume that the recall target is set to be registered for 7 days and reaches the level 40, and that the reference target is set to be registered for 3 days and reaches the level 15.
Based on the above assumption, a user who logs in for 7 days and reaches 40 levels is taken as a reference user, and a user who logs in for 3 days to seven days and reaches 15 levels but does not reach 40 levels is taken as a user to be analyzed. And selecting game data generated by the reference user 3 days before logging or 15 stages before the reference user is met from the game data of the reference user as first game data. Similarly, from among the above-described game data of the user to be analyzed, game data generated before the user to be analyzed satisfies login for 3 days or before reaching level 15 is selected as third game data.
And inputting the first game data and the third game data into a similarity model, calculating the similarity between the third game data and the first game data through the similarity model, and taking the calculated similarity between the third game data and the first game data as the similarity between the user to be analyzed and the reference user.
Optionally, to simplify the similarity calculation process and improve the calculation efficiency, the similarity may be calculated using one or more of the following game data types: the method comprises the steps of friend number, friend activity, recharging condition, game system data, prop use condition, weapon use condition, war change condition and duplicate gate.
Of course, to ensure accuracy of the similarity analysis, the similarity may also be calculated using the full-scale game data. For example, the similarity is calculated using all game data generated before the user satisfies the reference target.
Finally, after obtaining the similarity between the user to be analyzed and the reference user, in 103, the user to be analyzed, the similarity of which meets the preset condition, is determined to be the target recall user, and game resources corresponding to the recall target are pushed to the target recall user.
The preset conditions to be met by the similarity include, but are not limited to, any one of the following: the similarity is not smaller than the similarity threshold, and the order obtained based on the similarity ordering belongs to a preset target order.
For example, assume that the preset condition is that the similarity is not less than the similarity threshold. Let the similarity threshold be 60%. Assume that the similarity between each of the users a, b, c to be analyzed and the reference user is 83%, 40%, 66%. Based on the above assumption, in 103, users a, c to be analyzed whose similarity is not less than 60% are determined as target recall users, and game resources corresponding to recall targets are pushed to the target recall users a, c.
In another example, assume that the preset condition is that the order obtained based on the similarity sorting belongs to a preset target order. Assume that the target cis position is 100. Assuming that the number of users to be analyzed is m (where m is greater than 100), based on the above assumption, the users to be analyzed 1,2, 3, … …, m are ordered from large to small according to the similarity.
Further, assume that the sorting result is users 1, 2,3, … …, m to be analyzed, based on which the first 100 bits are taken as target recall users, i.e., users 1 to 100 to be analyzed are selected as target recall users.
It should be noted that if there are multiple target recall users, the same type of game resource may be pushed to the multiple target recall users. To increase game interest, optionally, game resources are randomly selected from the same type of game resources and pushed to different target recall users.
Or the target recall users can be divided according to other dimension data, so that different types of game resources are pushed for different target recall user groups. For example, game resources defined by the region are pushed to the target recall user in the same region according to the game region in which the target recall user is located. Or pushing game resources such as skin, activities, pets and the like related to respective roles for target recall users of different role types according to the role types of the target recall users.
Here, the manner of determining the game resource corresponding to the recall target may be referred to below, which is not developed here.
Optionally, after 103, the number of successful recalled users in the target recall users and corresponding game data are obtained, and the obtained game data and the historical game data are subjected to iterative analysis to optimize a recall analysis model. The number of successfully recalled users in the target recall users can be used as one of reference data of iterative analysis.
The iterative analysis result can be optionally used for adjusting the display mode or the display content of the game resource. For example, assuming that the game resource is a plurality of groups of new function introduction words, based on the successful recall rate of the target recall user corresponding to each group of new function introduction words, the new function introduction words with higher success rate are selected as the game resource for subsequent use.
In practical application, optionally, the iterative analysis result can be used for dynamically sequencing the game resources, so that the pushing proportion of the game resources with higher recall success rate is increased, the game resources are enriched, and the recall efficiency is further improved.
Through 101 to 103, through the automatic screening of the target recall user, not only is the user with larger recall potential avoided without relying on human experience analysis, but also the recall efficiency is greatly improved, the scale of the target recall user can be controlled, and the recall cost is reduced. Meanwhile, by pushing game resources corresponding to recall targets for the target recall users, the target recall users under different recall targets can acquire more interesting game resources, game playability is improved, and recall effects are improved.
In the above or the following embodiments, to improve the pertinence of the game resources, the method for determining the game resources shown in fig. 2 may also be executed, and specific steps are as follows:
201. Selecting historical game data generated before meeting a reference target from game data of a plurality of users, wherein the historical game data comprises: referencing first game data generated before the user satisfies the reference target, and second game data generated before the user who does not satisfy the recall target satisfies the reference target;
202. Respectively marking the first game data and the second game data by adopting different values to obtain marked historical game data;
203. inputting the marked historical game data into a correlation analysis model so that the correlation analysis model calculates the correlation between the historical game data and the recall target;
204. game resources are determined based on a correlation between historical game data and recall targets.
Wherein the game data of the historical game data includes, but is not limited to, any one or combination of the following: user level, login days, number of friends, friend activity, recharge status, game system data, prop use status, weapon use status, battle change status, duplicate checkpoints. The specific parameters are similar to those above and will not be expanded here.
Continuing with the example above in the above step, it is still assumed that the recall target is set to log in for 7 days and reaches level 40, and the reference target is set to log in for 3 days and reaches level 15.
Based on the above assumption, the user who logs in for 7 days and reaches 40 levels is taken as the reference user. In 201, historical game data generated before a reference target is satisfied is selected from game data of a plurality of users. Specifically, from among the above-described game data of the reference user, game data generated before the reference user satisfies login for 3 days or before reaching level 15 is selected as the first game data. From the game data of the users who do not satisfy the recall target (including the users who reach the reference target and the users who do not reach the reference target), the game data generated 3 days before the user to be analyzed satisfies the login or 15 stages before is selected as the second game data. 202, the first game data is marked with 1, the second game data is marked with 0, and the marked historical game data is obtained.
Further, assuming that the correlation analysis model is a regression analysis model, in 203, the marked historical game data is input into the regression analysis model so that the regression analysis model calculates a correlation matrix between the historical game data and the recall target, and the correlation matrix can reflect the correlation between each historical game data and the recall target.
Finally, assuming that there are multiple historical game data, based on this, in 204, determining game resources based on the correlation of the historical game data with recall targets may be implemented as:
Arranging the historical game data according to the correlation between the historical game data and the recall target from large to small, and acquiring the data type of the historical game data with the sequence at the front target position; and taking the game resource corresponding to the acquired data type as the game resource.
For example, the plurality of history game data are arranged in such a manner that the correlation with the recall target is large to small, and the data types to which the history game data in the first 2-digit order belong are acquired. The data types of the historical game data of the first 2 bits are pet culture data and pet capturing data, so that the correlation between the use condition of the pet system and a recall target is strong, and game resources corresponding to the pet system are taken as game resources.
In practical application, the correlation analysis model can be implemented by other models or algorithms besides the regression analysis model, and the method is not limited in the invention.
In this embodiment, the correlation analysis model is used to analyze the correlation between the game data and the recall target from each dimension, so that the game data with stronger correlation is used as the basis for selecting the game resources, thereby improving the pertinence of the game resources and further improving the recall effect.
In the above or below embodiments, optionally, a plurality of recall targets are set to recall user groups of different types and different levels, so as to further improve recall efficiency and enhance recall effect.
Specifically, if the recall analysis model is provided with a plurality of recall targets, the plurality of users are further divided into a plurality of reference user groups and user groups to be analyzed corresponding to the plurality of reference user groups based on the plurality of recall targets.
Further, after the multiple reference user groups and the user groups to be analyzed corresponding to the multiple reference user groups are divided, in 102, the recall analysis model calculates the user to be analyzed based on the reference user, so as to obtain the similarity between the user to be analyzed and the reference user, which can be implemented as follows:
and calculating the corresponding user groups to be analyzed based on the plurality of reference user groups through the recall analysis model to obtain the similarity between each user group to be analyzed and the corresponding reference user group. The similarity between each user group to be analyzed and the corresponding reference user group comprises the following steps: similarity between each user to be analyzed in each user group to be analyzed and the corresponding reference user group.
Furthermore, in the similarity between each user group to be analyzed and the corresponding reference user group output based on the recall analysis model, 103, determining the user to be analyzed whose similarity meets the preset condition as the target recall user may be implemented as:
and selecting users to be analyzed, the similarity of which meets preset conditions, from the user groups to be analyzed, and forming target recall user groups corresponding to the user groups to be analyzed.
Optionally, assuming that each user group to be analyzed corresponds to each recall target one by one, based on this, in 103, before game resources corresponding to the recall targets are pushed to target recall users, it is further determined whether the same target recall users exist in different target recall user groups.
If the same target recall user exists in different target recall user groups, the fact that repeated data exists in different target recall user groups is indicated, and in order to avoid harassment caused by repeated game resource pushing, duplicate removal processing can be performed on the target recall user groups corresponding to all user groups to be analyzed based on the difficulty level of recall targets, so that the same target recall user does not exist in different target recall user groups, and each target recall user is guaranteed to push game resources only once.
Of course, if multiple pushing needs to be implemented for some target recall users to increase the recall probability, other duplication eliminating processing modes can be set. If a plurality of recall targets with stronger relevance to the target recall user are reserved, game resources corresponding to the recall targets are pushed for the recall targets.
In the embodiment, the types and the levels of the target recall users are enriched by setting a plurality of recall targets, so that the user recall effect is further improved.
In the above or below embodiments, it is assumed that a plurality of reference users exist among a plurality of users, optionally game data of the plurality of reference users is acquired, and a reference performance is selected from game performances of the plurality of reference users based on the game data of the plurality of reference users. Wherein the reference performance includes, but is not limited to, one or a combination of the following actions: the behavior executed by the reference users exceeding the preset times, the behavior of the reference users reaching the recall target in the set time, the behavior of the reference users with higher grades, the behavior of the reference users with higher liveness and the behavior of the reference users with more recharging.
For example, assuming that 1000 reference users exist, based on this, game data of 1000 reference users are acquired, and game expressions associated with the respective reference users are marked based on the game data of 1000 reference users, respectively, and from among the game expressions of 1000 reference users, a game expression of a target position at the top of the number of marking ranks is selected as the reference expression. Or selecting the game performance associated with the reference user with the target position before the activity ranking and the game performance of the reference user reaching the recall target within 2 days from the game performances of 1000 reference users to form the reference performance.
Thus, further, the recall analysis model in 102 calculates the game performance of the user to be analyzed based on the reference performance, and obtains the similarity between the user to be analyzed and the reference user in the game performance. Or screening game resources to be pushed based on the correlation of the game data of the user and the reference expression.
In practical applications, the game performance in this embodiment may refer to the game performance described above, that is, the interaction behavior between the user and the game system. Wherein the reference user associated game performance is, for example, one or a combination of the following: logging actions, interactions with NPC, interactions with other players, and actions in various gaming systems (e.g., combat systems, collection systems, weapons systems, skills systems, face pinching systems, fostering systems, etc.).
In the above or the following embodiments, in addition to the recall target described in the above embodiments, in practical application, the recall target may alternatively be set by using a recall target guide map. Specifically, the recall target guide graph comprises a tree hierarchy structure based on recall target partitioning and a plurality of nodes, wherein the plurality of nodes respectively represent recall conditions in different dimensions.
Based on the recall target guide graphs introduced above, optionally, behavior features in game data of each user to be analyzed are extracted, and recall target guide graphs matched with each user to be analyzed are determined according to the behavior features of each user to be analyzed. Therefore, the recall target guide graphs corresponding to the users to be analyzed are matched, so that recall analysis is more targeted, the judgment accuracy of the users to be analyzed is further improved, and the users to be analyzed which are possibly recalled are avoided.
In the above steps, firstly, the behavior characteristics in the game data of each user to be analyzed are extracted through a machine learning model or a preset algorithm. For example, game data of each user to be analyzed is input into a classification prediction model, and behavior feature labels of each user to be analyzed are output through the classification prediction model.
Furthermore, in the above steps, determining, according to the behavior characteristics of each user to be analyzed, a recall target guide graph matched with each user to be analyzed may be implemented as:
Inquiring graph guide branches or nodes corresponding to the behavior characteristics of each user to be analyzed in a preset standard recall target graph according to the behavior characteristics of each user to be analyzed; and establishing recall target guide graphs matched with all users to be analyzed based on the queried guide graph branches or nodes.
In the preset standard recall target guide graph, the nodes comprise root nodes, intermediate nodes and leaf nodes. The next layer node (i.e., intermediate node) corresponding to the root node includes recall policy nodes, each of which is used to indicate a user churn reason or a user reflow reason, respectively. In fact, the intermediate nodes can also be refined into multiple layers, thereby providing a more targeted recall analysis strategy. Optionally, the recall policy node includes a drain-back policy node for indicating a user drain-back reason and/or a drain-back maintenance policy node for indicating a user drain-back reason. Each recall policy node corresponds to at least one leaf node for indicating a game resource corresponding to the user churn reason or the user reflow reason.
Based on the above example, in the above steps, through the behavior feature labels of the users to be analyzed, from the preset standard recall target guide graph, the guide graph branches or nodes matched with the behavior feature labels are queried and used as a plurality of nodes corresponding to the behavior features of the users to be analyzed. And further, based on the queried multiple nodes, respectively constructing recall target guide graphs matched with the users to be analyzed.
In practical application, the recall target guide graphs matched with the users to be analyzed and determined through the steps can be part or all branches of the standard recall target guide graphs.
For example, in a standard recall target guide graph, the reasons for user churn in the churn recall policy node include, but are not limited to, one or a combination of the following: insufficient version content, serious loss, poor interaction experience, poor game experience, hardware problems, unexpected game and conventional loss. Poor gaming experience is, for example, poor payment experience, poor numerical experience, severe combat faults (ecological imbalance in games). The hardware problems are, for example, unstable server environment, unsmooth operation, stuck, flash back, excessive memory occupancy rate, and severe equipment heating. The game is not expected to be, for example, advertisement is greatly distinguished from game, and the game is greatly separated from the designed prototype experience. Conventional churn is, for example, a sales prop problem in a game, a game account problem, a friend churn, and no time to play the game. User reflow reasons in the reflow maintenance policy node include, but are not limited to, one or a combination of the following: version content updates (e.g., new profession, new play, new game system), reflow benefits (e.g., reflow package, top-up benefits, reflow game privileges, etc.).
Based on the above example, in the above steps, it is assumed that the behavior feature tag of the user a to be analyzed is that the number of active friends is small and the game is flashing. Based on the behavior feature labels of the user a to be analyzed, searching for graph guide branches corresponding to the behavior feature labels of the user a to be analyzed in a preset standard recall target graph, such as hardware problems including flashing back and conventional loss and friend loss. And further, establishing a recall target guide graph matched with the user a to be analyzed based on the guide graph branches, namely a recall target guide graph a' formed by the two branches.
Further, after determining the recall target guide graphs matched with the users to be analyzed, inputting the game data into a recall analysis model in 102, so that the recall analysis model calculates the users to be analyzed based on the reference user to obtain the similarity of the users to be analyzed and the reference user in game performance, and the method can be further realized as follows:
Determining reference users which are matched with recall target guide graphs of all the users to be analyzed from a plurality of users as the reference users matched with all the users to be analyzed; and inputting the behavior characteristics of each user to be analyzed, the recall target guide graphs matched with each user to be analyzed and the game data of the reference users into a recall analysis model, so that the recall analysis model calculates the similarity of each user to be analyzed and each matched reference user in game performance.
Continuing with the above example, a reference user of the plurality of users that meets recall target guide image a' is determined to be a reference user that matches user a to be analyzed. And further, inputting the behavior characteristics of the user a to be analyzed, the recall target guide graph a' matched with the user to be analyzed and the game data of the reference user into a recall analysis model, so that the recall analysis model calculates the similarity of the user a to be analyzed and the matched reference user in game performance.
Therefore, by introducing the recall target guide graph, more targeted reference users can be screened in advance, the analysis flow of the recall analysis model is assisted, and more accurate recall analysis results are provided for the users to be analyzed.
Optionally, the recall target guide map matched by each user to be analyzed further comprises game resources corresponding to the recall targets. If the user to be analyzed is determined to be the target recall user, determining recall target guide graphs matched with the users to be analyzed according to the behavior characteristics of the users to be analyzed, and determining recall strategy nodes with highest matching degree from the matched recall target guide graphs according to the behavior characteristics of the target recall users; and taking the game resource indicated in the recall strategy node as the game resource corresponding to the recall target.
Continuing with the example above, the recall target guide graph a' also includes game resources (i.e., leaf nodes) corresponding to the two branches described above, hardware issues-flashing-version update package, regular churn-new churn match.
In practical applications, the game resources corresponding to the loss recall policy node include, but are not limited to, one or a combination of directional recall information, call back, designated tasks, recharge benefits, designated game areas. For example, the game resources corresponding to the back flow maintenance policy node include, but are not limited to, one or a combination of welfare activities, rewards packages, designated tasks, recharge welfare, designated game areas. It should be noted that the reflux maintenance policy node may also be used to make a decision for issuing game resources after the target recall user logs in again to the game.
The directional recall information is, for example, advertisement, short message and push information which are targeted to the target recall user. Optionally, the directional recall information carries a reflux gift bag activation mode for improving recall possibility. In practice, the content of the document that directs recall information may be set according to the correlation between historical game data and recall targets (see the above embodiments). Further, the document format and the word number of the directional recall information are preset, so that automatic generation of auxiliary directional recall information is realized, the document is ensured to be concise, parade is avoided, and the user's sense of well being is further improved.
Optionally, the recall telephone content is determined based on a correlation between historical game data and recall targets. For example, based on the correlation between historical game data and recall targets, game props or tasks preferred by the target recall user are determined and relevant content is prompted to the user in the recall telephone. The relevant content is, for example, game props giving away user preferences, or tasks subsequent to updating user preferences, or target recall users in profession where the user likes to select.
Further, voice information in the call-back telephone can be generated by adopting character sounds preferred by the user, so that the game experience recall of the user is checked, and the interest of the user is improved. Or voice materials with higher historical recall rate (such as voice packets of the persona) can also be used for generating voice information in the call-back telephone.
User recall devices in accordance with one or more embodiments of the present invention are described in detail below. Those skilled in the art will appreciate that these user recall devices can each be configured by the steps taught by the present solution using commercially available hardware components.
Fig. 3 is a schematic structural diagram of a user recall device according to an embodiment of the present invention, where, as shown in fig. 3, the user recall device includes: the device comprises an acquisition module 11, an analysis module 12 and a pushing module 13.
An acquisition module 11, configured to acquire game data of a target object, where the game data includes game system data of the target object in a first period of time;
the analysis module 12 is used for inputting game data into the recall analysis model so that the recall analysis model calculates the user to be analyzed based on the reference user to obtain the similarity of the user to be analyzed and the reference user in game performance;
And the pushing module 13 is used for determining the user to be analyzed, the similarity of which meets the preset condition, as the target recall user and pushing the game resource corresponding to the recall target to the target recall user.
Optionally, the apparatus further comprises a determining module for:
Selecting historical game data generated before the reference target is met from game data of the plurality of users, wherein the historical game data comprises: the reference user satisfies first game data generated before the reference target and second game data generated before the recall target is satisfied by the user who does not satisfy the recall target; respectively marking the first game data and the second game data by adopting different values to obtain marked historical game data; inputting the marked historical game data into a correlation analysis model so that the correlation analysis model calculates the correlation between the historical game data and the recall target; the game resource is determined based on a correlation between the historical game data and the recall target.
Wherein, optionally, if there are a plurality of historical game data, the determining module is specifically configured to, when determining the game resource based on the correlation between the historical game data and the recall target:
Arranging the historical game data according to the correlation between the historical game data and the recall target from large to small, and acquiring the data type of the historical game data with the sequence at the previous target position; and taking the game resource corresponding to the acquired data type as the game resource.
Wherein the game data of the historical game data comprises any one or combination of the following: user level, login days, number of friends, friend activity, recharge status, game system data, prop use status, weapon use status, battle change status, duplicate checkpoints.
Optionally, the recall analysis model is a machine learning model for similarity calculation.
The analysis module 12 inputs the game data into a recall analysis model, so that the recall analysis model is specifically configured to, when calculating the user to be analyzed based on the reference user to obtain a similarity between the user to be analyzed and the reference user in game performance:
Selecting first game data generated before the reference user meets the reference target from the game data of the reference user;
Selecting third game data generated before the user to be analyzed meets the reference target from the game data of the user to be analyzed;
And calculating the similarity of the third game data and the first game data through the machine learning model for calculating the similarity, and taking the calculated similarity of the third game data and the first game data as the similarity of the user to be analyzed and the reference user in game performance.
Optionally, if the recall analysis model is provided with a plurality of recall targets, the apparatus further comprises a preprocessing module for:
dividing the users into a plurality of reference user groups and user groups to be analyzed corresponding to the reference user groups based on the recall targets.
Wherein, optionally, each user group to be analyzed corresponds to each recall target one by one.
The preprocessing module is further configured to, before the pushing module 13 pushes, to the target recall user, a game resource corresponding to the recall target:
judging whether the same target recall user exists in different target recall user groups; if yes, performing repeated elimination processing on the target recall user groups corresponding to the user groups to be analyzed based on the difficulty level of the recall target, so that the same target recall user does not exist in different target recall user groups.
Optionally, the preset condition includes any one or a combination of the following: the similarity is not less than a similarity threshold; the order obtained based on the similarity ordering belongs to a preset target order.
The user recall device shown in fig. 3 may perform the method provided in the foregoing embodiments, and for the portions of this embodiment not described in detail, reference may be made to the relevant descriptions of the foregoing embodiments, which are not repeated here.
In one possible design, the configuration of the user recall device shown in FIG. 3 described above may be implemented as an electronic device. As shown in fig. 4, the electronic device may include: a processor 21, and a memory 22. Wherein said memory 22 has stored thereon executable code which, when executed by said processor 21, at least enables said processor 21 to implement a user recall method as provided in the previous embodiments.
The electronic device may further include a communication interface 23 for communicating with other devices or a communication network.
Additionally, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code that, when executed by a processor of an electronic device, causes the processor to perform the user recall method provided in the previous embodiments.
The apparatus embodiments described above are merely illustrative, wherein the various modules illustrated as separate components may or may not be physically separate. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by adding necessary general purpose hardware platforms, or may be implemented by a combination of hardware and software. Based on such understanding, the foregoing aspects, in essence and portions contributing to the art, may be embodied in the form of a computer program product, which may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.