User loss prediction method and device, and storage medium
By obtaining user target behavior information and using the churn prediction model to generate user portraits for prediction, the problem of unexplainable user churn prediction in the prior art is solved, and a more accurate and clear user churn prediction and retention strategy is achieved.
Patent Information
- Application Number
- CN202411675195.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-04-25
AI Technical Summary
The prior art is difficult to provide explainable results in user churn prediction, and it is impossible to accurately know the cause of user churn.
By obtaining user target behavior information, the image generation network and prediction network are used to generate user portraits and make churn predictions. When the prediction result meets the preset conditions, the corresponding user retention strategy is output.
It improves the interpretability of user churn prediction, makes user portraits and prediction results have clear behavior description information, and improves the accuracy of determining user retention strategies.
Smart Images
Figure CN119887271A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of churn prediction, and in particular to a user churn prediction method, device and storage medium. Background Art
[0002] At present, with the rapid development of information technology, enterprises are facing increasingly fierce market competition and changing user needs. In order to maintain competitiveness, enterprises must deeply understand and predict user behavior, especially predict user churn, such as using user churn prediction technology to predict user churn information.
[0003] However, most of the related technologies are based on machine learning algorithms or deep learning models to predict user churn information, which can easily lead to a situation where they can only determine whether a user has churned, but cannot know the reason for the user churn, resulting in poor interpretability of user churn prediction. Summary of the invention
[0004] The main purpose of this application is to provide a user churn prediction method, device and storage medium, aiming to improve the interpretability of user churn prediction.
[0005] In a first aspect, the present application provides a user churn prediction method, comprising:
[0006] Obtain the user's target behavior information;
[0007] Based on a preset portrait generation network of a churn prediction model and a preset first prompt text, determining a user portrait of the user according to the target behavior information;
[0008] Based on the prediction network of the churn prediction model and the preset second prompt text, churn prediction is performed on the user according to the user portrait to obtain a predicted churn result of the user;
[0009] When it is determined that the predicted churn result meets the preset user retention condition, a user retention strategy corresponding to the user portrait is output according to the user portrait.
[0010] In a second aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;
[0011] The memory is used to store computer programs;
[0012] The processor is used to execute the computer program and implement the steps of the user churn prediction method as described above when executing the computer program.
[0013] In a third aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the user churn prediction method as described above are implemented.
[0014] The present application provides a user churn prediction method, device and storage medium. The user churn prediction method includes: obtaining target behavior information of the user; determining the user portrait of the user according to the target behavior information based on a portrait generation network of a preset churn prediction model and a preset first prompt text; predicting the churn of the user according to the user portrait based on a prediction network of the churn prediction model and a preset second prompt text, and obtaining a predicted churn result of the user; when the predicted churn result meets the preset user retention condition, outputting a user retention strategy corresponding to the user portrait according to the user portrait.
[0015] For example, the portrait generation network of the churn prediction model can analyze the target behavior information according to the instructions of the first prompt text, such as summarizing, concluding, evaluating and describing the target behavior information, etc., to determine the user portrait of the user, so that the user portrait is interpretable. Correspondingly, the prediction network of the churn prediction model can analyze the factors that are likely to cause user churn, the trend of user churn, etc. according to the instructions of the second prompt text, so as to predict the churn of the user and obtain the predicted churn result of the user, so that the predicted churn result of the user is interpretable. When both the user portrait of the user and the predicted churn result of the user are interpretable, it is conducive to improving the interpretability of user churn prediction. Accordingly, based on the interpretability of the user portrait, when the corresponding user retention strategy is output according to the user portrait, a user retention strategy that is more suitable for the user behavior needs can be output, which is conducive to improving the accuracy of determining the user retention strategy. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0017] Figure 1 It is a flowchart of a user churn prediction method provided in an embodiment of the present application;
[0018] Figure 2 is a flowchart of a user churn prediction method according to an embodiment of the present application;
[0019] Figure 3is a schematic block diagram of a user churn prediction device provided in an embodiment of the present application;
[0020] Figure 4 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0022] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.
[0023] The embodiments of the present application provide a user churn prediction method, device and storage medium. The user churn prediction method can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and the like. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0024] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.
[0025] See also Figure 1 , Figure 1 It is a flow chart of a user churn prediction method provided by an embodiment of the present application. It should be noted that the control method of the interactive task provided by the embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain the user's target behavior information from the computer device, and process the target behavior information accordingly according to the user churn prediction method to obtain a user retention strategy. Exemplarily, the server can, for example, send the user retention strategy to the computer device so that the computer device executes the user retention strategy to retain the user; of course, it is not limited to this and is not limited here.
[0026] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.
[0027] like Figure 1 As shown, the user churn prediction method includes steps S101 to S104.
[0028] Step S101: Obtain user's target behavior information.
[0029] For example, the user churn prediction method can be applied to different application scenarios. For example, the application scenarios may include e-commerce platforms, social media platforms, financial service platforms, etc., which are not limited here. User behaviors in different application scenarios may be different. For example, user behaviors on e-commerce platforms may include commodity purchasing behaviors, commodity return behaviors, commodity browsing behaviors, etc., which are not limited here. For example, user behaviors on social media platforms may include content publishing behaviors, content interaction behaviors, etc., which are not limited here. For another example, user behaviors on financial service platforms may include investment behaviors, risk tolerance assessment behaviors, etc., which are not limited here.
[0030] The computer device may acquire the target behavior information of the user in response to the user behavior performed by the user.
[0031] In some implementations, the preset behavior information of the user is obtained; the non-text information in the preset behavior information is converted into text information to obtain the target behavior information of the user.
[0032] For example, when a computer device detects a user behavior performed by a user, it can record the user behavior to determine the user's preset behavior information. For example, the computer device can record the user behavior in the form of tables, texts, logs, pictures, buried data, etc. to obtain the user's preset behavior information. When the computer device records the user behavior in the form of text, corresponding text information can be formed in the preset behavior information. Accordingly, when the computer device records the user behavior in a form other than text, corresponding non-text information can be formed in the preset behavior information.
[0033] Compared with non-text information, when viewing text information, relevant personnel can better obtain and understand the meaning expressed by text information, and it can be determined that the interpretability of text information is better than the interpretability of non-text information. Based on this, when the computer device obtains the user's preset behavior information, it can convert the non-text information in the preset behavior information to text information, thereby determining the user's target behavior information. Accordingly, the computer device can determine that the interpretability of the target behavior information is better than the interpretability of the preset behavior information. The target behavior information can be used for subsequent user churn prediction to improve the interpretability of user churn prediction.
[0034] The computer device can obtain the user's target behavior information through the churn prediction model. The churn prediction model can include a Large Language Model (LLM) model. For example, the churn prediction model can be provided with an information processing network. The computer device can input the user's preset behavior information into the information processing network, and the information processing network can convert the non-text information in the preset behavior information into text information to obtain the user's target behavior information. For example, the information processing network can convert the non-text information in the preset behavior information into text information through a preset code script for determining the user's target behavior information, which is not limited here.
[0035] In this way, when the user's target behavior information is obtained, the target behavior information can be used for subsequent prediction of user churn. Accordingly, in the process of determining the user's target behavior information, the computer device can convert the non-text information included in the obtained preset behavior information into text information, which is conducive to improving the diversity of information sources corresponding to the target behavior information and the comprehensiveness of information sources. Accordingly, the user churn prediction can be made based on the target behavior information with better diversity and comprehensiveness of information sources, which is conducive to improving the accuracy of user churn prediction in the future. Moreover, since the target behavior information is all text information, the text information is easier for relevant personnel to view and understand, which is conducive to improving the interpretability of user churn prediction in the future.
[0036] Step S102: Based on a preset portrait generation network of a churn prediction model and a preset first prompt text, a user portrait of the user is determined according to target behavior information.
[0037] For example, when obtaining the user's target behavior information, the computer device can input the target behavior information into the portrait generation network of the churn prediction model, so that the portrait generation network combines the instructions of the preset first prompt text to perform portrait generation processing on the target behavior information to obtain the user portrait of the user.
[0038] The first prompt text can be used to instruct the portrait generation network how to summarize, conclude, evaluate, and describe the target behavior information, etc., so as to determine the user portrait of the user.
[0039] In some implementations, the target behavior information is processed by text description based on the first prompt text to obtain the user's behavior description information; and a user portrait of the user is generated based on the behavior description information.
[0040] For example, the portrait generation network may perform text description processing on the target behavior information in response to the instruction of the first prompt text to determine the user's behavior description information.
[0041] Taking the e-commerce platform as an example, the first prompt text can be used to indicate the evaluation and description of the user's commodity purchasing habits, commodity purchasing frequency, commodity repurchase rate, etc. Based on this, the portrait generation network can respond to the instruction of the first prompt text, determine the user's commodity purchasing habits, commodity purchasing frequency, commodity repurchase rate, etc. according to the commodity purchasing behavior information, commodity return behavior information, commodity browsing behavior information, etc. included in the target behavior information, and describe them in the form of text to obtain the user's behavior description information.
[0042] Taking the social media platform as an example, the first prompt text can be used to indicate the evaluation and description of the user's content publishing frequency, content interaction volume, etc. Based on this, the portrait generation network can respond to the prompt of the first prompt text, determine the user's content publishing frequency, content interaction volume, etc. according to the content publishing behavior information, content interaction behavior information, etc. included in the target behavior information, and describe them in the form of text to obtain the user's behavior description information.
[0043] Taking the financial service platform as an example, the first prompt text can be used to indicate the evaluation and description of the user's investment amount, the user's investment risk, the user's investment frequency, etc. Based on this, the portrait generation network can respond to the prompt of the first prompt text, determine the user's investment amount, investment risk, investment frequency, etc. according to the investment behavior information, risk tolerance assessment behavior information, etc. included in the target behavior information, and describe them in the form of text to obtain the user's behavior description information.
[0044] Of course, the first prompt text and behavior description information are not limited to this and are not restricted here.
[0045] When determining the user's behavior description information, the portrait generation network can integrate all the behavior description information to generate a user portrait of the user. For example, the portrait generation network can set the user's behavior description information around the preset user image or inside the user image based on the preset user image. Of course, it is not limited to this. The portrait generation network can also determine the information category corresponding to the behavior description information based on the first prompt text, such as product purchasing habits, product purchase frequency, product repurchase rate, etc., and set the corresponding behavior description information under the corresponding information category, which is not limited here.
[0046] In this way, when the churn prediction model determines the user portrait of the user based on the target behavior information based on the portrait generation network and the preset first prompt text, the churn prediction model can intuitively display the behavior description information included in the user portrait to relevant personnel through the user portrait, which is conducive to improving the interpretability of the user portrait.
[0047] Step S103: Based on the prediction network of the churn prediction model and the preset second prompt text, churn prediction is performed on the user according to the user portrait to obtain a predicted churn result of the user.
[0048] For example, when the churn prediction model determines the user portrait of the user through the portrait generation network, the churn prediction model can input the user portrait into the prediction network so that the prediction network combines the instructions of the second prompt text to perform churn prediction processing on the user portrait to obtain the user's predicted churn result.
[0049] The second prompt text can be used to instruct the prediction network on how to analyze the possibility of user churn, such as analyzing factors that are likely to cause user churn, whether there is behavioral description information in the user portrait that can characterize the trend of user churn, etc., so as to analyze the causal relationship related to user churn prediction and determine the user's predicted churn result.
[0050] In some implementations, based on the second prompt text, the causal relationship between the user portrait and the preset churn type is analyzed to obtain user churn analysis information corresponding to the user portrait; based on the user churn analysis information, the predicted churn result of the user is determined.
[0051] For example, the preset churn type may include at least one of user churn and user non-churn. The prediction network may analyze the possibility of user churn in response to the prompt of the second prompt text.
[0052] In an exemplary embodiment, the prediction network can analyze the causal relationship between the user portrait and user churn. For example, the second prompt text can be used to indicate whether the behavior description information in the user portrait analyzed by the prediction network will lead to user churn, and the second prompt text can also be used to indicate whether the behavior description information in the user portrait analyzed by the prediction network shows a trend of user churn, etc., which is not limited here.
[0053] In an exemplary embodiment, the prediction network can analyze the causal relationship between the user portrait and the user's non-churn. For example, the second prompt text can be used to indicate whether the prediction network analyzes the behavior description information in the user portrait to cause the user to not churn, and the second prompt text can also be used to indicate whether the prediction network analyzes the behavior description information in the user portrait to show a trend of user non-churn, etc., which is not limited here.
[0054] Accordingly, the prediction network can comprehensively analyze the causal relationship between the user profile and different preset churn types to obtain the user churn analysis information, and determine the user's predicted churn result. For example, the user's predicted churn result may include the user churn analysis information, and the user churn type determined according to the user churn analysis information. The user churn type may include any one of the preset churn types.
[0055] Taking the e-commerce platform as an example, the second prompt text can be used to indicate the behavioral description information included in the user portrait, such as the user's commodity purchasing habits, commodity purchasing frequency, commodity repurchase rate, etc., to analyze the causal relationship between the user portrait and the preset churn type to obtain user churn analysis information. For example, when the commodity purchasing habits indicate that the user often purchases commodity A, and the user's commodity purchase frequency for commodity A is once a month, if the user has not purchased commodity A for several consecutive months, the prediction network can determine that the user portrait shows a trend of user churn. The prediction network can further analyze the reasons for user churn in combination with the user portrait. For example, the prediction network can determine the user's historical evaluation information of commodity A based on the user portrait. If the most recent historical evaluation information indicates that the user has doubts about the authenticity of the source of commodity A, the prediction network can determine that one of the reasons for user churn is that the authenticity of the source of commodity A is unclear.
[0056] Taking the social media platform as an example, the second prompt text can be used to indicate the behavioral description information included in the user portrait, such as the user's content publishing frequency, content interaction volume, etc., to analyze the causal relationship between the user portrait and the preset churn type to obtain user churn analysis information. For example, when the user's content publishing frequency is greater than or equal to the preset publishing frequency threshold, if the time interval between the time when the user last published content and the current time is greater than or equal to the preset time interval threshold, the prediction network can determine that the user portrait shows a trend of user churn. The prediction network can further analyze the reasons for user churn in combination with the user portrait. For example, the prediction network can determine that one of the reasons for user churn is the social media platform's flow restriction when the content interaction volume in the user portrait shows a downward trend and the user portrait indicates that the user's most recent complaint about the social media platform's restriction on others viewing published content has been fed back through the complaint channel of the app store or social media platform.
[0057] By analogy, the prediction network can analyze the causal relationship between the user portrait and the preset churn type according to the instruction of the second prompt text, such as determining the trend of user churn and the reasons for user churn. The prediction network can determine the user churn analysis information by combining the trend of user churn and the reasons for user churn. Of course, the second prompt text and user churn analysis information are not limited to this and are not limited here.
[0058] Accordingly, the prediction network can determine the user churn type according to the user churn analysis information, thereby determining the predicted churn result of the user. For example, the prediction network can combine the user churn analysis information and the user churn type to determine the predicted churn result, so that the predicted churn result indicates the process of obtaining the conclusion of the user churn type from the analysis process of the user churn analysis information.
[0059] In this way, the churn prediction model predicts churn for the user based on the prediction network and the preset second prompt text according to the user portrait. When the predicted churn result of the user is obtained, the churn prediction model can display the user churn trend, the reasons for user churn and other user churn analysis information through the predicted churn result. The user churn analysis information can be viewed and understood by relevant personnel, which is conducive to improving the interpretability of user churn prediction.
[0060] Step S104: when it is determined that the predicted churn result meets the preset user retention condition, output a user retention strategy corresponding to the user portrait according to the user portrait.
[0061] For example, when the churn prediction model determines the predicted churn result, it can determine whether the predicted churn result meets the preset user retention condition. The preset user retention condition can be pre-set or set by relevant personnel.
[0062] When the predicted churn results meet the preset user retention conditions, the churn prediction model can determine that the user needs to be retained. The churn prediction model can then output a user retention strategy corresponding to the user portrait based on the user portrait.
[0063] In some implementations, when the predicted churn result indicates user churn, and the predicted churn result is the same as the actual churn result of the user, it is determined that the predicted churn result meets a preset user retention condition.
[0064] For example, the churn prediction model can obtain the actual churn result of the user in the process of judging whether the predicted churn result meets the preset user retention conditions. For example, when the time interval between the occurrence time corresponding to the user's most recent target behavior information and the current time is greater than or equal to the first time interval threshold, the churn prediction model can determine that the actual churn result of the user is user churn. When the time interval between the occurrence time corresponding to the user's most recent target behavior information and the current time is less than the first time interval threshold, the churn prediction model can determine that the actual churn result of the user is that the user has not churned. Of course, it is not limited to this. For example, the churn prediction model can select key behavior information from the user's target behavior information. Key behavior information includes, for example, the user's platform login behavior information, platform access behavior information, platform transaction behavior information, platform core function usage behavior information, etc., which are not limited here. When the time interval between the occurrence time corresponding to the user's most recent key behavior information and the current time is greater than or equal to the second time interval threshold, the churn prediction model can determine that the actual churn result of the user is user churn. When the time interval between the occurrence time corresponding to the user's most recent key behavior information and the current time is less than the second time interval threshold, the churn prediction model can determine that the actual churn result of the user is that the user has not churned. This is not limited here. The second time interval threshold may be the same as the first time interval threshold, or may be different from the first time interval threshold, which is not limited here. The first time interval threshold and the second time interval threshold may be preset, or may be set by relevant personnel themselves, which is not limited here.
[0065] In the case of determining the predicted churn result of the user and obtaining the actual churn result of the user, the churn prediction model can comprehensively predict the churn result and the actual churn result to determine whether the predicted churn result meets the preset user retention condition. For example, based on the consideration of retaining lost users and temporarily not retaining non-lost users, the churn prediction model can set a preset user retention condition including the predicted churn result indicating user churn. Based on the consideration of improving the efficiency of the strategy output of the churn prediction model, the churn prediction model can also determine whether the predicted churn result is the same as the actual churn result of the user. If they are the same, it can be determined that the user is a lost user and needs to be retained. If they are not the same, it can be determined that the user is a non-lost user and does not need to be retained temporarily, or it can be determined that the accuracy of the user portrait of the user is not good, and the user portrait needs to be regenerated to re-determine the predicted churn result, so that the user does not need to be retained temporarily. Accordingly, the churn prediction model can set a preset user retention condition including the predicted churn result being the same as the actual churn result of the user, so as to eliminate the predicted churn results that are different from the actual churn results of the user from the predicted churn results indicating user churn, and obtain the predicted churn results that meet the preset user retention conditions. The churn prediction model can then output a user retention strategy corresponding to the user portrait based on the user portrait when the predicted churn results meet the preset user retention conditions.
[0066] In this way, based on the setting of the preset user retention conditions, the churn prediction model can screen out the predicted churn results that meet the preset user retention conditions. When the predicted churn results of the user meet the preset user retention conditions, the churn prediction model can determine that the relevant personnel have a retention demand for the user who meets the preset user retention conditions, and the churn prediction model can then determine the user retention strategy corresponding to the user profile of the user whose predicted churn results meet the preset user retention conditions, thereby facilitating the retention convenience and accuracy of the churn users.
[0067] In some implementations, based on the behavior description information included in the user portrait, a user retention strategy corresponding to the user portrait is determined from preset retention strategies.
[0068] The preset retention strategy can be set for the historical user portrait or the historical user behavior description information included in the historical user portrait. The preset retention strategy can be based on the targeted operational suggestions and strategies provided by experts or relevant personnel for historical churn users. For example, the preset retention strategy can be determined based on the historical user portrait corresponding to the historical churn user or the historical user behavior description information included in the historical user portrait corresponding to the historical churn user. Then the churn prediction model can establish an association between the historical user portrait and the preset retention strategy, and can also establish an association between the historical user behavior description information and the preset retention strategy. Historical churn users can be used to indicate users who have been retained from a churn state to a non-churn state. The preset retention strategy may include marketing strategies, advertising promotion strategies, notification promotion strategies, etc., which are not limited here.
[0069] In some embodiments, the churn prediction model may be provided with a strategy output network. When the predicted churn result meets the preset user retention condition, the predicted churn result may input the user profile into the strategy output network, so that the strategy output network outputs the user retention strategy corresponding to the user profile according to the user profile.
[0070] For example, the churn prediction model can determine the user retention strategy corresponding to the user portrait from the preset retention strategy based on at least one of the association relationship between the historical user portrait and the preset retention strategy and the association relationship between the historical user behavior description information and the preset retention strategy. The process of determining the user retention strategy can be explainable. For example, the churn prediction model can analyze the derivation process and derivation conclusion of obtaining the user retention strategy corresponding to the user portrait from the user portrait in the form of text information, so that relevant personnel can view and understand the user retention strategy determined by the churn prediction model.
[0071] In an exemplary embodiment, the churn prediction model may determine the first degree of match between the user portrait and the historical user portrait in the process of determining the user retention strategy corresponding to the user portrait. In the case where the first degree of match between the historical user portrait and the user portrait is greater than or equal to the first degree of match threshold, the strategy output model may determine the preset retention strategy associated with the historical user portrait whose first degree of match is greater than or equal to the first degree of match threshold as the user retention strategy corresponding to the user portrait.
[0072] In another exemplary implementation, the churn prediction model may determine the second degree of match between the behavior description information in the user portrait and the historical behavior description information in the process of determining the user retention strategy corresponding to the user portrait. In the event that the second degree of match between the historical behavior description information and the behavior description information in the user portrait is greater than or equal to the second matching threshold, the churn prediction model may determine the preset retention strategy associated with the historical behavior description information whose second matching degree is greater than or equal to the second matching threshold as the user retention strategy corresponding to the user portrait. The second matching threshold may be the same as the first matching threshold, or may be different from the first matching threshold, and the first matching threshold and the second matching threshold may be pre-set, or may be set by relevant personnel themselves, and are not limited here.
[0073] Of course, it is not limited to this. The churn prediction model can also integrate all preset retention strategies associated with historical user portraits with a first matching degree greater than or equal to a first matching degree threshold, and the preset retention strategies associated with historical behavior description information with a second matching degree greater than or equal to a second matching degree threshold, to determine the user retention strategy corresponding to the user portrait, which is not limited here.
[0074] In this way, the churn prediction model determines the user retention strategy corresponding to the user portrait from the preset retention strategy based on the behavior description information included in the user portrait, and the user retention strategy corresponding to the user portrait can be implemented by the computer device to retain lost users, which is conducive to improving the convenience of retaining lost users. Accordingly, when the user portrait and the behavior description information included in the user portrait are both interpretable, the churn prediction model can output the user retention strategy corresponding to the user portrait in an interpretable form, so that relevant personnel can view and understand the user retention strategy, which is conducive to improving the convenience of implementing the user retention strategy and improving the convenience of retaining lost users.
[0075] For example, when the predicted churn result is different from the actual churn result of the user, the churn prediction model needs to regenerate the user's user portrait to re-determine the predicted churn result corresponding to the user portrait. Based on this, the churn prediction model needs to improve the user portrait generation process accordingly.
[0076] In some embodiments, when the predicted churn result is different from the actual churn result of the user, key behavior information is selected from the target behavior information according to the behavior type corresponding to the target behavior information; based on the portrait generation network and the first prompt text, the user portrait is regenerated according to the key behavior information.
[0077] For example, when the predicted churn result is different from the actual churn result of the user, the churn prediction model can determine that there is interference information in the user portrait that is less relevant or irrelevant to the prediction process of the predicted churn result, thereby adversely affecting the prediction accuracy of the predicted churn result. Accordingly, since the user portrait is determined by the churn prediction model based on the portrait generation network and the first prompt text according to the target behavior information, the churn prediction model can determine that there is interference information in the target behavior information that is less relevant or irrelevant to the prediction process of the predicted churn result, resulting in poor prediction accuracy of the predicted churn result. In order to improve the accuracy of the churn prediction model in determining the predicted churn result, the churn prediction model can select key behavior information from the target behavior information based on the behavior type corresponding to the target behavior information, so as to regenerate the user portrait of the user based on the key behavior information.
[0078] In the process of selecting key behavior information from target behavior information, the selection strategies for key behavior information corresponding to different behavior types are different.
[0079] Taking the e-commerce platform as an example, the target behavior information may include product purchase behavior information and product browsing behavior information, and the behavior type corresponding to the target behavior information related to the e-commerce platform may be a product transaction type. If a user only browses the products provided by the e-commerce platform but does not purchase the products, then for the e-commerce platform, the user has not conducted a product transaction. Accordingly, if the user has not conducted a product transaction for a long time, the possibility of user churn is relatively high. Based on this, the churn prediction model can select the product purchase behavior information as the key behavior information from the product purchase behavior information and product browsing behavior information included in the target behavior information according to the behavior type corresponding to the target behavior information related to the e-commerce platform.
[0080] Taking the social media platform as an example, the target behavior information may include content publishing behavior information, content interaction behavior information, and behavior information about following other users. The behavior type corresponding to the target behavior information related to the social media platform may be content communication. If a user only follows other users through the social media platform, but does not publish content or interact with other users, then for the social media platform, the user has not conducted content communication. Accordingly, if a user has not conducted content communication for a long time, the possibility of user churn is relatively high. Based on this, the churn prediction model can select content publishing behavior information and content interaction behavior information as key behavior information from the content publishing behavior information, content interaction behavior information, and behavior information about following other users included in the target behavior information according to the behavior type corresponding to the target behavior information related to the social media platform.
[0081] By analogy, when the predicted churn result is different from the actual churn result of the user, the churn prediction model can select key behavior information from the target behavior information according to the behavior type corresponding to the target behavior information.
[0082] In an exemplary embodiment, the churn prediction model may be provided with a portrait improvement network. In the case where the predicted churn result is different from the actual churn result of the user, the churn prediction model may input the target behavior information into the portrait improvement network, so that the portrait improvement network outputs the corresponding key behavior information according to the behavior type corresponding to the target behavior information, so that the subsequent portrait generation network can regenerate the user portrait of the user. Of course, this is not limited to this and is not limited here.
[0083] Accordingly, when the key behavior information is determined, the churn prediction model can input the key behavior information into the portrait generation network, so that the portrait generation network performs portrait generation processing on the key behavior information in response to the instruction of the first prompt text, and then regenerates the user portrait of the user.
[0084] In this way, in the process of regenerating the user's user portrait, the churn prediction model can eliminate the information in the target behavior information that is less relevant or irrelevant to the prediction process of the user's predicted churn result, so as to select the key behavior information, and the churn prediction model can regenerate the user portrait based on the key behavior information. The regenerated user portrait can be used by the churn prediction model to re-determine the user's predicted churn result, and when the information in the target behavior information that is less relevant or irrelevant to the prediction process of the user's predicted churn result is eliminated, it is beneficial to improve the prediction accuracy of the predicted churn result.
[0085] In some implementations, key behavior information is selected from the target behavior information according to the behavior type corresponding to the target behavior information and the historical key behavior corresponding to the behavior type.
[0086] For example, each time the churn prediction model determines that the predicted churn result is different from the actual churn result of the user, the churn prediction model can select key behavior information from the target behavior information based on the behavior type corresponding to the target behavior information. Accordingly, the churn prediction model can record the key behavior information selected each time, such as recording it as a historical key behavior. Based on this, the churn prediction model can record the historical key behaviors corresponding to different behavior types. Accordingly, the churn prediction model can establish a correspondence between different behavior types and historical key behaviors. The correspondence between different behavior types and historical key behaviors can be used to guide the churn prediction model to determine the key points that need to be considered for regenerating the user portrait, so as to select key behavior information from the target behavior information, and then regenerate the user portrait of the user, thereby improving the prediction accuracy of the predicted churn result determined based on the user portrait.
[0087] When the churn prediction model once again determines that the predicted churn result is different from the actual churn result of the user, the churn prediction model can determine the historical key behavior corresponding to the behavior type from the corresponding relationship between different behavior types and historical key behaviors according to the behavior type corresponding to the target behavior information. If the historical key behavior exists in the target behavior information, the churn prediction model can determine the historical key behavior as the key behavior information.
[0088] For example, when the churn prediction model selects key behavior information 1 from target behavior information 1 based on the behavior type corresponding to target behavior information 1 being behavior type 1, and selects key behavior information 2 from target behavior information 2 based on the behavior type corresponding to target behavior information 2 being behavior type 2, the churn prediction model can determine key behavior information 1 as the historical key behavior corresponding to behavior type 1, and determine key behavior information 2 as the historical key behavior corresponding to behavior type 2, so as to establish a corresponding relationship between different behavior types and historical key behaviors. Accordingly, when the predicted churn result of the user corresponding to target behavior information n by the churn prediction model is different from the actual churn result of the user, the churn prediction model can determine the behavior type corresponding to target behavior information n. For example, when the behavior type corresponding to target behavior information n is behavior type 1, and target behavior information n includes key behavior information 1, the churn prediction model can select key behavior information 1 from target behavior information n as key behavior information. Accordingly, the churn prediction model can also select behavior information other than key behavior information 1 from target behavior information n as key behavior information based on the behavior type corresponding to target behavior information n, which is not limited here. For another example, when the behavior type corresponding to the target behavior information n is behavior type 2, and the target behavior information n includes key behavior information 2, the churn prediction model can select key behavior information 2 from the target behavior information n as the key behavior information. Accordingly, the churn prediction model can also select behavior information other than key behavior information 2 from the target behavior information n as the key behavior information based on the behavior type corresponding to the target behavior information n, which is not limited here. When the key behavior information corresponding to the target behavior information n is determined, the churn prediction model can input the key behavior information into the portrait generation network, so that the portrait generation network performs portrait generation processing on the key behavior information in response to the instruction of the first prompt text, and then regenerates the user portrait of the user corresponding to the target behavior information n, so as to subsequently re-determine the predicted churn result of the user corresponding to the target behavior information n.
[0089] In an exemplary embodiment, the churn prediction model may be provided with an improved memory network. The churn prediction model may record, through the improved memory network, the key behavior information selected from the target behavior information by the churn prediction model based on the behavior type corresponding to the target behavior information each time the churn prediction model determines that the predicted churn result is different from the actual churn result of the user, that is, the corresponding relationship between different behavior types and historical key behaviors is recorded. Accordingly, when the churn prediction model once again determines that the predicted churn results of other users are different from the actual churn results of the user, the churn prediction model may determine the key points to be considered for regenerating the user portraits of other users through the improved memory network, such as outputting the corresponding historical key behaviors according to the behavior types corresponding to the target behavior information of other users, so that the churn prediction model can select key behavior information from the target behavior information of other users, and then regenerate the user portraits of other users.
[0090] In this way, when the churn prediction model selects key behavior information from the target behavior information according to the behavior type corresponding to the target behavior information and the historical key behaviors corresponding to the behavior type, the churn prediction model can comprehensively consider the key points required to regenerate the user's user portrait to determine the key behavior information, which is conducive to improving the convenience and accuracy of selecting key behavior information for subsequent regeneration of the user's user portrait.
[0091] In an exemplary embodiment, the churn prediction model may be provided with an information processing network, a portrait generation network, a prediction network, a strategy output network, a portrait improvement network, and an improved memory network. Among them, the strategy output network, the portrait improvement network, and the improved memory network may be an integrated review network or multiple separately provided review networks, and the network architectures of review networks with different functions may be the same or different, which is not limited here.
[0092] like Figure 2As shown, the churn prediction model can obtain the preset behavior information of the user. The churn prediction model can input the preset behavior information into the information processing network to convert the non-text information in the preset behavior information into text information through the information processing network, thereby obtaining the target behavior information of the user. The churn prediction model can input the target behavior information into the portrait generation network to determine the user portrait of the user according to the target behavior information based on the portrait generation network and the preset first prompt text. The churn prediction model can input the user portrait into the prediction network to predict the churn of the user according to the user portrait based on the prediction network and the preset second prompt text, and obtain the predicted churn result of the user. The churn prediction model can determine whether the predicted churn result meets the preset user retention condition. When it is determined that the predicted churn result meets the preset user retention condition, the churn prediction model can input the user portrait into the strategy output network so that the strategy output network outputs the user retention strategy corresponding to the user portrait according to the user portrait. When it is determined that the predicted churn result does not meet the preset user retention condition, the churn prediction model can determine whether the predicted churn result is the same as the actual churn result. In the case where it is determined that the predicted churn result is different from the actual churn result, the churn prediction model can input the target behavior information into the portrait improvement network, so that the portrait improvement network selects key behavior information from the target behavior information according to the behavior type corresponding to the target behavior information, so that the subsequent portrait generation network can regenerate the user portrait of the user. Of course, it is not limited to this. The churn prediction model can record the key behavior information selected by the portrait improvement network according to the behavior type corresponding to the target behavior information each time through the improved memory network, so as to establish the association relationship between different behavior types and key behavior information, so as to select key behavior information from the target behavior information in the future, and then the subsequent portrait generation network can regenerate the user portrait of the user.
[0093] The user churn prediction method provided in the above embodiment includes obtaining the user's target behavior information; determining the user's user portrait according to the target behavior information based on a portrait generation network of a preset churn prediction model and a preset first prompt text; predicting the user's churn according to the user portrait based on a prediction network of the churn prediction model and a preset second prompt text, and obtaining a predicted churn result for the user; when the predicted churn result meets the preset user retention condition, outputting a user retention strategy corresponding to the user portrait according to the user portrait.
[0094] For example, the portrait generation network of the churn prediction model can analyze the target behavior information according to the instructions of the first prompt text, such as summarizing, concluding, evaluating and describing the target behavior information, etc., to determine the user portrait of the user, so that the user portrait is interpretable. Correspondingly, the prediction network of the churn prediction model can analyze the factors that are likely to cause user churn, the trend of user churn, etc. according to the instructions of the second prompt text, so as to predict the churn of the user and obtain the predicted churn result of the user, so that the predicted churn result of the user is interpretable. When both the user portrait of the user and the predicted churn result of the user are interpretable, it is conducive to improving the interpretability of user churn prediction. Accordingly, based on the interpretability of the user portrait, when the corresponding user retention strategy is output according to the user portrait, a user retention strategy that is more suitable for the user behavior needs can be output, which is conducive to improving the accuracy of determining the user retention strategy.
[0095] See also Figure 3 , Figure 3 1 is a schematic block diagram of a user churn prediction device provided in an embodiment of the present application. The user churn prediction device can be configured in a server or a computer device to execute the aforementioned user churn prediction method.
[0096] like Figure 3 As shown, the user churn prediction device includes: an information acquisition module 110, a portrait generation module 120, a prediction module 130 and a strategy output module 140.
[0097] The information acquisition module 110 is used to acquire the user's target behavior information.
[0098] The portrait generation module 120 is used to determine the user portrait of the user according to the target behavior information based on the portrait generation network of the preset churn prediction model and the preset first prompt text.
[0099] The prediction module 130 is used to perform churn prediction on the user according to the user portrait based on the prediction network of the churn prediction model and the preset second prompt text, so as to obtain the predicted churn result of the user.
[0100] The strategy output module 140 is used to output the user retention strategy corresponding to the user portrait according to the user portrait when it is determined that the predicted churn result meets the preset user retention condition.
[0101] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0102] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0103] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.
[0104] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.
[0105] like Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.
[0106] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any user churn prediction method.
[0107] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
[0108] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any user churn prediction method.
[0109] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0110] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be 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, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0111] In one embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:
[0112] Obtain the user's target behavior information;
[0113] Based on a preset portrait generation network of a churn prediction model and a preset first prompt text, determining a user portrait of the user according to the target behavior information;
[0114] Based on the prediction network of the churn prediction model and the preset second prompt text, churn prediction is performed on the user according to the user portrait to obtain a predicted churn result of the user;
[0115] When it is determined that the predicted churn result meets the preset user retention condition, a user retention strategy corresponding to the user portrait is output according to the user portrait.
[0116] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the control of the interactive task described above can refer to the corresponding process in the aforementioned user churn prediction method embodiment, and will not be repeated here.
[0117] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The method implemented when the computer program is executed by a processor can refer to the various embodiments of the user churn prediction method of the present application.
[0118] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.
[0119] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0120] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.
[0121] It should also be understood that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.
[0122] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A method for predicting user churn, characterized in that: include: Obtain the user's target behavior information; Based on a preset portrait generation network of a churn prediction model and a preset first prompt text, determining a user portrait of the user according to the target behavior information; Based on the prediction network of the churn prediction model and the preset second prompt text, churn prediction is performed on the user according to the user portrait to obtain a predicted churn result of the user; When it is determined that the predicted churn result meets the preset user retention condition, a user retention strategy corresponding to the user portrait is output according to the user portrait.
2. The user churn prediction method according to claim 1, characterized in that: The profile generation network based on the preset churn prediction model and the preset first prompt text determines the user profile of the user according to the target behavior information, including: According to the first prompt text, performing text description processing on the target behavior information to obtain the behavior description information of the user; A user profile of the user is generated based on the behavior description information.
3. The user churn prediction method according to claim 1, characterized in that: The prediction network based on the churn prediction model and the preset second prompt text predicts churn of the user according to the user portrait to obtain the predicted churn result of the user, including: Analyze the causal relationship between the user portrait and the preset churn type according to the second prompt text to obtain user churn analysis information corresponding to the user portrait; Determine the predicted churn result of the user according to the user churn analysis information.
4. The user churn prediction method according to claim 1, characterized in that: Determining that the predicted churn result meets a preset user retention condition includes: When the predicted churn result indicates the user churn, and the predicted churn result is the same as the actual churn result of the user, it is determined that the predicted churn result meets a preset user retention condition.
5. The user churn prediction method according to claim 4, characterized in that: Outputting a user retention strategy corresponding to the user portrait according to the user portrait includes: According to the behavior description information included in the user portrait, a user retention strategy corresponding to the user portrait is determined from preset retention strategies.
6. The user churn prediction method according to claim 1, characterized in that: The obtaining of the user's target behavior information includes: Obtaining preset behavior information of the user; The non-text information in the preset behavior information is converted into text information to obtain the target behavior information of the user.
7. The user churn prediction method according to any one of claims 1 to 6, characterized in that: The user churn prediction method further includes: When the predicted churn result is different from the actual churn result of the user, selecting key behavior information from the target behavior information according to the behavior type corresponding to the target behavior information; Based on the portrait generation network and the first prompt text, and according to the key behavior information, the user portrait of the user is regenerated.
8. The user churn prediction method according to claim 7, characterized in that: The selecting key behavior information from the target behavior information according to the behavior type corresponding to the target behavior information includes: According to the behavior type corresponding to the target behavior information and the historical key behaviors corresponding to the behavior type, key behavior information is selected from the target behavior information.
9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the user churn prediction method according to any one of claims 1 to 8 when executing the computer program.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the user churn prediction method according to any one of claims 1 to 8 are implemented.