Account behavior prediction model training method and account behavior prediction method

Abstract

The invention discloses an account behavior prediction model training method and an account behavior prediction method, and belongs to the technical field of artificial intelligence. According to the account behavior prediction model training method, the account behavior prediction model can be trained according to the consistency difference between the correlation between the behavior prediction results corresponding to a plurality of samples and the correlation between the behavior level labels of the end behaviors corresponding to the plurality of samples in the account behavior link. Different from the related technology of predicting the behavior in the account behavior link only based on the own tag of a certain behavior, the embodiment of the invention not only considers the own modeling result of a certain behavior to be predicted in the account behavior link, but also needs to consider the consistency of the behavior to be predicted and the behavior level tag of the end behavior; the comprehensive utilization rate of the account behavior link topology structure and the information at different nodes is fully improved, and the accuracy of account behavior prediction is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly relates to a method for training an account behavior prediction model and a method for predicting account behavior. Background Art

[0002] In related technologies, an account behavior can be modeled based on artificial intelligence, and a trained account behavior prediction model is used to estimate the probability of a certain behavior occurring in an account behavior link, and based on this probability, the recommendation or placement of relevant media content is guided. However, this conventional account behavior prediction method focuses on modeling information related to a single behavior in the account behavior link, and the ability to construct losses during the training process is not strong, resulting in low accuracy and completeness of account behavior prediction. Summary of the Invention

[0003] Embodiments of this application provide a method for training an account behavior prediction model and a method for predicting account behavior to solve the technical problem of low accuracy and completeness of account behavior prediction.

[0004] According to one aspect of the embodiments of this application, a method for training an account behavior prediction model is provided, and the method includes:

[0005] Obtain a first sample and a second sample, where both the first sample and the second sample include account basic data and a behavior level label corresponding to the end behavior of the account behavior link;

[0006] Determine a relative constraint label, where the relative constraint label is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample;

[0007] Input the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes predicted values corresponding to multiple behaviors in the account behavior link;

[0008] Input the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set;

[0009] Determine at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the predicted values with corresponding relationships in the first prediction result set and the second prediction result set;

[0010] Adjust the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

[0011] According to one aspect of the embodiments of the present application, there is provided an account behavior prediction method, the method comprising:

[0012] Obtaining the account basic data of the target account;

[0013] Inputting the account basic data into an account behavior prediction model to obtain the probabilities that the target account executes at least two behaviors in a preset account behavior link respectively, and the account behavior prediction model is trained based on the aforementioned account behavior prediction model training method.

[0014] According to one aspect of the embodiments of the present application, there is provided an account behavior prediction model training device, the device comprising:

[0015] A sample acquisition module, configured to acquire a first sample and a second sample, where both the first sample and the second sample include account basic data and a behavior level label corresponding to the end behavior of the account behavior link;

[0016] A training module, configured to perform the following operations:

[0017] Determining a relative constraint label, where the relative constraint label is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample;

[0018] Inputting the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes prediction values corresponding to multiple behaviors in the account behavior link respectively;

[0019] Inputting the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set;

[0020] Determining at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the prediction values with corresponding relationships in the first prediction result set and the second prediction result set;

[0021] Adjusting the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

[0022] According to one aspect of the embodiments of the present application, there is provided an account behavior prediction device, the device comprising:

[0023] A data acquisition module, configured to acquire the account basic data of the target account;

[0024] A behavior prediction module, configured to input the account basic data into an account behavior prediction model to obtain the probabilities that the target account executes at least two behaviors in a preset account behavior link, where the account behavior prediction model is trained based on the aforementioned account behavior prediction model training method.

[0025] According to one aspect of the embodiments of the present application, a computer device is provided. The computer device includes a processor and a memory. At least one instruction, at least one program, a code set, or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the aforementioned account behavior prediction model training method or account behavior prediction method.

[0026] According to one aspect of the embodiments of the present application, a computer-readable storage medium is provided. At least one instruction, at least one program, a code set, or an instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the aforementioned account behavior prediction model training method or account behavior prediction method.

[0027] According to one aspect of the embodiments of the present application, a computer program product is provided. The computer program product includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device executes to implement the aforementioned account behavior prediction model training method or account behavior prediction method.

[0028] The technical solution provided by the embodiments of the present application can bring the following beneficial effects:

[0029] The embodiments of the present application provide a method for training an account behavior prediction model. This method for training an account behavior prediction model can train the account behavior prediction model based on the consistency difference between the correlation relationship among the behavior prediction results corresponding to multiple samples and the correlation relationship among the behavior level labels of the end behaviors corresponding to the multiple samples in the account behavior link. Different from the related technologies that only predict a behavior in the account behavior link based on the label of a certain behavior itself, the embodiments of the present application not only need to apply the information of a certain behavior to model the corresponding behavior prediction result, but also need to quantify the consistency difference based on the behavior level labels at the end of the account behavior link. That is to say, at least the self-modeling result and related information of a certain behavior to be predicted in the account behavior link need to be considered, and the consistency with the behavior level labels of the end behaviors also needs to be considered. At least the behavior information of two nodes (behaviors) in the account behavior link is utilized, and the new dimension information of the behavior level labels is also utilized, and the behavior order of the account behavior link is considered, which fully improves the comprehensive utilization rate of the topological structure of the account behavior link and the information at different nodes, and improves the accuracy of account behavior prediction. The account behavior prediction model can output the predicted values corresponding to multiple behaviors in the account behavior link, and the accuracy of each predicted value is improved compared with the related technologies.

[0030] The embodiments of the present application also provide an account behavior prediction method, which uses the aforementioned account behavior prediction model to perform account behavior prediction. Description of the Drawings

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0032] Figure 1 It is a schematic diagram of chained account behaviors in a resource interaction scenario provided by an embodiment of the present application;

[0033] Figure 2 It is a schematic diagram of the application program running environment provided by an embodiment of the present application;

[0034] Figure 3 It is a flowchart of the method for training an account behavior prediction model provided by an embodiment of the present application;

[0035] Figure 4 It is a schematic diagram of the preset model structure provided by an embodiment of the present application;

[0036] Figure 5It is a schematic flowchart of a method for generating a first behavior prediction result set provided by an embodiment of the present application;

[0037] Figure 6 It is a schematic flowchart of an account behavior prediction method provided by an embodiment of the present application;

[0038] Figure 7 It is a block diagram of an account behavior prediction model training device provided by an embodiment of the present application;

[0039] Figure 8 It is a block diagram of an account behavior prediction device provided by an embodiment of the present application;

[0040] Figure 9 It is a structural block diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0041] Before introducing the method embodiments provided by the present application, relevant terms or nouns that may be involved in the method embodiments of the present application are briefly introduced first, so as to facilitate the understanding of those skilled in the art of the present application.

[0042] Artificial Intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, a theory, method, technology, and application system that perceives the environment, acquires knowledge, and uses knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence is also to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning, and decision-making.

[0043] Artificial intelligence technology is an interdisciplinary subject, involving a wide range of fields, including both hardware-level technologies and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, large account behavior prediction model training technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0044] Machine Learning (ML) is an interdisciplinary field that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0045] Deep learning: The concept of deep learning originated from the research of artificial neural networks. A multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning forms more abstract high-level representations of attribute categories or features by combining low-level features to discover the distributed feature representations of data. The structure of deep learning can be used to construct the preset model in the embodiments of this application.

[0046] Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to achieve data computing, storage, processing, and sharing. Cloud technology is the general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model, which can form a resource pool and be used on demand, flexibly and conveniently. Cloud computing technology will become an important support. The back-end services of technical network systems require a large amount of computing and storage resources, such as video websites, picture-based websites, and more portal websites. With the highly developed applications of the Internet industry, in the future, each item may have its own identification mark and needs to be transmitted to the back-end system for logical processing. Data at different levels will be processed separately, and various industry data requires the support of a powerful system back-end, which can only be achieved through cloud computing.

[0047] Media content: Media content refers to various forms of information and content presented on various media platforms. It includes: Text-based content: including short essays, long articles, columns, etc., which is the most traditional form of media content. Picture-based content: including pictures, posters, comics, etc., which can be published through social media, communities and other platforms, and is usually used to convey information or express emotions. Video-based content: including short videos, live broadcasts, etc., which can be published through short video platforms or live broadcast platforms. Audio-based content: including broadcasts, podcasts, etc., which are usually published through audio platforms and mainly target users' auditory needs. Interactive content: including Q&A, voting, surveys, etc., which can be published through Q&A platforms and can promote interaction and communication between the media and the audience. In addition, media content also includes various forms such as audio, video, animation, charts and images, and can present news reports, current affairs comments, entertainment programs, advertising and other content. Different media platforms and forms are suitable for different audience groups, and media content is also constantly changing and innovating.

[0048] Account credit-granting behavior: It refers to the first account providing resource support to the second account or guaranteeing the credit of the second account in a preset activity. There can be multiple levels of credit-granting in the credit-granting behavior. The higher the credit-granting level, usually the better the credit of the second account and the stronger the payment ability for resources.

[0049] Resource advance behavior: The second account advances resources from the first account.

[0050] Chain account behavior: Chain User Behavior refers to the situation where a certain account needs to continuously execute multiple operations or behaviors in the process of completing a certain task or achieving a certain goal. There is a certain logical relationship and dependency between these operations or behaviors, forming a chain structure. Chain account behavior exists in various application scenarios. For example, in the e-commerce interaction application scenario, the chain account behavior can be account browsing products - adding to the shopping cart - settlement - account payment. In the media content interaction scenario, the chain account behavior can be account browsing behavior - account liking behavior - account forwarding behavior. In the embodiments of the present application, taking the resource interaction scenario as an example, the chain account behavior is described in detail:

[0051] In the resource interaction scenario, the first account can deliver media content related to resource interaction to relevant accounts, promoting the generation of chain account behavior related to resource interaction between some of the relevant accounts and the first account. The accounts participating in this chain account behavior are the second accounts. This chain account behavior can be summarized as delivering media content to an account group - resource advance behavior - credit-granting behavior.

[0052] Please refer to Figure 1, which shows a schematic diagram of chained account behaviors in an exemplary resource interaction scenario proposed in an embodiment of the present application. In the resource interaction scenario, the group of each account that delivers media content can be understood as the overall market. After exposing the media content to the accounts in the overall market, some second accounts in the overall market can apply for resource advance from the first account, that is, a resource advance behavior occurs. On the premise of the occurrence of the resource advance behavior, the first account can grant credit to the second account according to the actual situation, and the second account accepts the credit result, that is, a credit granting behavior occurs. Specifically, the first account can determine different credit granting levels for second accounts with different payment capabilities and different credit capabilities according to its own situation. The default credit granting level of the second account that has not been granted credit is 0, and the lowest credit granting level among the second accounts that have been granted credit is 1.

[0053] Before specifically elaborating on the embodiments of the present application, the relevant technical background related to the embodiments of the present application is introduced to facilitate the understanding of those skilled in the art of the present application.

[0054] In the related art, the account behavior can be modeled based on artificial intelligence, and the trained account behavior prediction model is used to estimate the probability of a certain behavior occurring in the account behavior link, and the recommendation or delivery of relevant media content is guided based on this probability. Taking the resource interaction scenario as an example, usually the account behavior prediction model predicts the probability of the occurrence of the credit granting behavior of an account when the media content is delivered to the account, abbreviated as the credit granting probability. According to the size of the credit granting probability, it is decided which accounts to deliver the media content to. However, the technical solution of delivering media content only based on the credit granting probability does not consider the occurrence probabilities of the behaviors in each link of the complete behavior link of "delivering media content to the account group - resource advance behavior - credit granting behavior", that is, it does not consider the conversion situation of each link of the link conversion, but only considers the credit granting probability, and also cannot take into account the information of the credit granting level.

[0055] That is to say, the account behavior prediction model in the related art is difficult to comprehensively model the link conversion situation of the chained account behavior, and overly models the occurrence probability of a single behavior in the chained account behavior, resulting in the difficulty of quantifying the comprehensive conversion situation of the chained account behavior, thereby resulting in relatively low accuracy and completeness of the account behavior prediction. Further, when guiding the delivery of media content based on the account behavior prediction model, the delivery effect of the media content is also reduced. In the resource interaction scenario, the account behavior prediction model in the related art cannot model the credit granting level either, which further reduces the accuracy and completeness of the account behavior prediction and reduces the delivery effect of the media content.

[0056] An embodiment of the present application provides a method for training an account behavior prediction model. This method for training an account behavior prediction model can train the account behavior prediction model based on the consistency difference between the correlation relationship among the behavior prediction results corresponding to multiple samples and the correlation relationship among the behavior level labels of the end behaviors corresponding to the multiple samples in the account behavior link. Different from the related technologies that only predict a certain behavior in the account behavior link based on the label of that behavior itself, the embodiment of the present application not only needs to use the information of a certain behavior to model the corresponding behavior prediction result, but also needs to quantify the consistency difference based on the behavior level labels at the end of the account behavior link. That is to say, at least the self-modeling result and related information of a certain behavior to be predicted in the account behavior link need to be considered, and the consistency with the behavior level labels of the end behaviors also needs to be considered. At least the behavior information of two nodes (behaviors) in the account behavior link is utilized, and the new dimension information of the behavior level labels is also utilized, and the behavior order of the account behavior link is considered, which fully improves the comprehensive utilization rate of the topological structure of the account behavior link and the information at different nodes, and improves the accuracy of account behavior prediction. The account behavior prediction model can output the predicted values corresponding to multiple behaviors in the account behavior link, and the accuracy of each predicted value is improved compared with the related technologies.

[0057] An embodiment of the present application further provides an account behavior prediction method, which uses the aforementioned account behavior prediction model to perform account behavior prediction.

[0058] To make the objectives, technical solutions, and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0059] Please refer to Figure 2 , which shows a schematic diagram of an application program running environment provided by an embodiment of the present application. The application program running environment may include: a terminal 10 and a server 20.

[0060] The terminal 10 includes, but is not limited to, electronic devices such as mobile phones, computers, intelligent voice interaction devices, intelligent home appliances, vehicle-mounted terminals, game consoles, e-book readers, multimedia playback devices, and wearable devices. A client of the application program may be installed in the terminal 10.

[0061] In the embodiments of the present application, the above application program can be any application program that can provide account behavior prediction model training services or account behavior prediction services. Typically, the application program can be an application program of the account behavior service type or the media content recommendation type. Of course, in addition to the application programs of the account behavior service type or the media content recommendation type, services that rely on account behavior prediction model training services or account behavior prediction services can also be provided in other types of application programs. For example, news application programs, social application programs, interactive entertainment application programs, browser application programs, shopping application programs, content sharing application programs, virtual reality (VR) application programs, augmented reality (AR) application programs, etc., and the embodiments of the present application do not make any limitations in this regard. The embodiments of the present application do not make any limitations in this regard. Optionally, a client of the above application program runs in the terminal 10.

[0062] The server 20 is used to provide background services for the client of the application program in the terminal 10. For example, the server 20 can be the background server of the above application program. The server 20 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, 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, CDN (Content Delivery Network, content distribution network), and big data and artificial intelligence platforms. Optionally, the server 20 provides background services for the application programs in multiple terminals 10 at the same time.

[0063] Optionally, the terminal 10 and the server 20 can communicate with each other through the network 30. The terminal 10 and the server 20 can be directly or indirectly connected through wired or wireless communication methods, and the present application does not make any limitations in this regard.

[0064] Please refer to Figure 3 , which shows a flowchart of an account behavior prediction model training method provided by an embodiment of the present application. This method can be applied to a computer device, and the above computer device refers to an electronic device with data calculation and processing capabilities. For example, the execution subject of each step can be Figure 2 the server 20 in the application program running environment shown. This method can include the following steps:

[0065] S301. Obtain a first sample and a second sample. Both the first sample and the second sample include account basic data and a behavior level label corresponding to the end behavior of the account behavior link.

[0066] The embodiments of the present application do not limit the account basic data, which can be understood as the general term for the information required for account behavior prediction. In the embodiments of the present application, the training method is described in detail by taking "media content delivery to account groups - resource advance behavior - credit granting behavior" as an example. Of course, this example does not limit the embodiments of the present application, and the technical solutions in the embodiments of the present application can be used in any scenario where there is a chain of account behaviors with behavior levels at the end.

[0067] The account basic data may include account behavior historical data and / or account attribute data. Account behavior historical data refers to the historical behavior of a certain account. For example, whether a link related to resource interaction has been clicked, whether resource interaction behavior has been participated in, whether resource advance has been applied for, whether credit has been granted by a relevant account, etc. Account attribute data refers to the inherent static attributes of the account, or the account profile. For example, the age, gender, native place, educational background, personal interests, behavior habits, preference tendencies, financial situation, credit data, etc. of the user corresponding to the account. Account behavior historical data and account attribute data can be regarded as the basic information required for performing account behavior prediction, and the embodiments of the present application do not limit their specific contents.

[0068] The account basic data can be used to predict the occurrence probability of each behavior in the account behavior link, and the account basic data also includes the true values (labels) corresponding to each behavior. The above-mentioned account behavior link includes adjacent first behavior and second behavior. The above-mentioned first sample includes a behavior label set, and the behavior label set includes behavior labels corresponding to various behaviors that occur sequentially in the above-mentioned account behavior link. Specifically, the behavior label set includes a first behavior label corresponding to the first behavior and a second behavior label corresponding to the second behavior. Taking the account behavior link of "media content delivery to account groups - resource advance behavior - credit granting behavior" as an example, the behavior level label of the end behavior of the account behavior link refers to the credit level corresponding to the credit granting behavior. The resource advance behavior of the account after media content is delivered to the account group is the first behavior, and the credit granting behavior of the account after the resource advance behavior occurs is the second behavior. Whether the resource advance behavior occurs in the account after media content is delivered to the account group is recorded in the first behavior label corresponding to the first behavior, and whether the credit granting behavior occurs in the account after the resource advance behavior occurs is recorded in the second behavior label corresponding to the second behavior. Based on the account basic data, the probability that the resource advance behavior occurs in the account after media content is delivered to the account group can be predicted, and the probability that the credit granting behavior occurs in the account after the resource advance behavior occurs can also be predicted.

[0069] Both the first sample and the second sample include account basic data and behavior level labels, and the first sample and the second sample are different samples. Taking the account behavior link of "delivering media content to an account group - resource advance behavior - credit granting behavior" as an example, both the first sample and the second sample have the following data content:

[0070] label wanjian ={1,0}

[0071] label shouxin ={1,0}

[0072] label souxin-level ={0,1,2,3,4,5}

[0073] label wanjian being 1 indicates that resource advance behavior occurred to the account after delivering media content to the account group, and label wanjian being 0 indicates that no resource advance behavior occurred to the account after delivering media content to the account group. label shouxin being 1 indicates that credit granting behavior occurred to the account after the resource advance behavior, and label shouxin being 0 indicates that no credit granting behavior occurred to the account after the resource advance behavior. label souxin-level indicates that there are five levels of credit granting, 1, 2, 3, 4, 5, where 0 indicates not granted.

[0074] In the embodiments of the present application, two samples can be randomly selected from the sample set used for account behavior prediction training as the first sample and the second sample. For each sample combination formed by the first sample and the second sample, the method in the embodiments of the present application can be used to adjust the parameters of the account behavior prediction model. In other words, the step of training the account behavior prediction model based on the first sample and the second sample can be understood as the step in the loop, and this loop can be executed repeatedly to improve the training effect.

[0075] S302. Determine the relative constraint label, and the relative constraint label is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample.

[0076] Simply put, the relative constraint label indicates the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample. Still taking the account behavior link of "delivering media content to an account group - resource advance behavior - credit granting behavior" as an example, if the credit granting level of the first sample is greater than that of the second sample, the value is 1, otherwise the value is 0. That is

[0077] labelshouxin-level-cmp Indicates a relative constraint label, label1 shouxin-level and label2 shouxin-level respectively represent the behavior level labels corresponding to the first sample and the second sample respectively.

[0078] S303. Input the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set. The first prediction result set includes predicted values corresponding to multiple behaviors in the above account behavior link; input the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set.

[0079] The first prediction result set may include prediction results for multiple behaviors, and the second prediction result set may also include prediction results for multiple behaviors. Specifically, the first prediction result set includes a first prediction value for the first behavior, a first prediction value for the second behavior, and a first prediction value for the associated behavior. The first prediction value for the first behavior indicates the probability of the first behavior occurring in the first prediction result set. The first prediction value for the second behavior indicates the probability of the second behavior occurring in the first prediction result set. The first prediction value for the associated behavior indicates the probability of the associated behavior occurring in the first prediction result set. The probability of the associated behavior occurring represents the probability of the first behavior and the second behavior occurring in sequence. Correspondingly, the second prediction result set includes a second prediction value for the first behavior, a second prediction value for the second behavior, and a second prediction value for the associated behavior.

[0080] Still taking the account behavior link of "delivering media content to an account group - resource advance behavior - credit granting behavior" as an example, the first prediction value for the first behavior refers to the probability of the resource advance behavior of the account occurring after delivering media content to the account group obtained by predicting the first sample. The first prediction value for the second behavior refers to the probability of the credit granting behavior occurring after the resource advance behavior obtained by predicting the first sample. The first prediction value for the associated behavior refers to the probability of the credit granting behavior occurring after delivering media content to the account group obtained by predicting the first sample. The second prediction value for the first behavior refers to the probability of the resource advance behavior of the account occurring after delivering media content to the account group obtained by predicting the second sample. The second prediction value for the second behavior refers to the probability of the credit granting behavior occurring after the resource advance behavior obtained by predicting the second sample. The second prediction value for the associated behavior refers to the probability of the credit granting behavior occurring after delivering media content to the account group obtained by predicting the second sample.

[0081] In the embodiments of the present application, the first row being the first prediction value, the second row being the first prediction value and the associated behavior first prediction value can be respectively represented as pctr1, pcvr1, and pctcvr1. The first row being the second prediction value, the second row being the second prediction value and the associated behavior second prediction value can be respectively represented as pctr2, pcvr2, and pctcvr2. Wherein, pctcvr1 = pctr1 * pcvr1; pctcvr2 = pctr2 * pcvr2.

[0082] S304. Determine at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the prediction values with corresponding relationships in the first prediction result set and the second prediction result set.

[0083] Based on the foregoing, determining at least one behavior difference indication parameter includes at least one of the following: a first difference indication parameter determined according to the magnitude relationship between the first behavior first prediction value and the first behavior second prediction value; a second difference indication parameter determined according to the magnitude relationship between the second behavior first prediction value and the second behavior second prediction value; a third difference indication parameter determined according to the magnitude relationship between the associated behavior first prediction value and the associated behavior second prediction value.

[0084] Exemplarily, the second difference indication parameter can be represented as The third difference indication parameter can be represented as

[0085] S305. Adjust the parameters of the preset model according to the consistency difference between each of the behavior difference indication parameters and the relative constraint label to obtain an account behavior prediction model.

[0086] Based on the foregoing, the consistency difference includes at least one of the following: a first difference generated when the magnitude relationship indicated by the first difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label; a second difference generated when the magnitude relationship indicated by the second difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label; a third difference generated when the magnitude relationship indicated by the third difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label.

[0087] Exemplarily, the second difference can be quantified as The third difference can be quantified as Among them, CrossEntropy is the English expression of cross entropy, which is an important loss function in machine learning and deep learning, especially in classification problems. Cross entropy measures the difference between the actual output (probability) and the expected output (label). During the training process, the model optimizes its parameters by minimizing the cross entropy to make the model's predictions closer to the true labels. Of course, the present application does not limit the quantification methods of the first difference, the second difference or the third difference. Cross entropy can be used or other methods can be used.

[0088] The present application can adjust the parameters of the above-mentioned preset model based on the above-mentioned consistency difference to obtain an account behavior prediction model. In some embodiments, the consistency difference can be used only as one of the bases for model parameter tuning, and the model can also be tuned based on the total loss composed of the consistency difference and other losses. In an exemplary embodiment, for any predicted value in the above-mentioned first prediction result set, a target behavior label related to the above-mentioned predicted value can be determined in the above-mentioned behavior label set, and the true value corresponding to the above-mentioned predicted value can be determined according to the above-mentioned target behavior label; according to the difference between the above-mentioned true value and the above-mentioned predicted value, a loss term can be determined; according to each of the above-mentioned loss terms and the above-mentioned consistency difference, the parameters of the above-mentioned preset model can be adjusted to obtain the above-mentioned account behavior prediction model.

[0089] Of course, in some embodiments, multiple loss terms can be obtained not only based on the first prediction result set. For the same reason, multiple loss terms can also be obtained based on the second prediction result set, and the model can be tuned according to the total loss composed of all the obtained loss terms and the consistency difference. Taking obtaining multiple loss terms based on the first prediction result set as an example, this step can include at least one of the following: determining the loss term corresponding to the above-mentioned first behavior according to the difference between the above-mentioned first behavior label and the first predicted value of the above-mentioned first behavior; determining the loss term corresponding to the above-mentioned second behavior according to the difference between the above-mentioned second behavior label and the first predicted value of the above-mentioned second behavior; determining the loss term corresponding to the above-mentioned associated behavior according to the difference between the true value of the associated behavior and the first predicted value of the associated behavior, and the true value of the associated behavior is determined based on the above-mentioned first behavior label and the above-mentioned second behavior label; among them, each of the above-mentioned loss terms is used to adjust the parameters of the above-mentioned preset model.

[0090] The embodiments of the present application do not limit the method of adjusting the parameters of the above-mentioned preset model based on each loss term and the consistency difference. For example, a weighted sum operation can be performed on each loss term and the consistency difference to obtain a total loss, and then the gradient descent method can be used for parameter tuning based on the total loss. Of course, the weights are not limited and can be set according to the actual situation. Iterating the entire preset model together based on the total loss can significantly improve the accuracy of the account behavior prediction model for predicting each behavior.

[0091] Please refer toFigure 4 , which shows a schematic diagram of the preset model structure provided by the embodiments of the present application. The above-mentioned preset model includes a first behavior prediction layer and a second behavior prediction layer, and both the first behavior prediction layer and the second behavior prediction layer are connected to the feature extraction layer; now, each technical term in this preset model will be explained:

[0092] Input: The input of the preset model;

[0093] DEEP FM: The DEEP FM model is short for Deep Factorization Machine, which is a machine learning model used in recommendation systems. The DEEP FM model can introduce cross-term features by performing second-order combinations between features, thereby improving the prediction ability of the preset model. In the case of highly sparse data, the DEEP FM model can better estimate the relationships between features. Compared with traditional linear models, the DEEP FM model has an additional part for feature combination at the back. The DEEP FM model introduces cross-term features by combining pairwise features, thereby improving the model score. DEEP FM can be used as the feature extraction layer to perform feature extraction operations.

[0094] DNN model: The DNN neural network is a deep neural network model designed to solve non-linear regression problems. It has the advantages of depth and capacity and can adapt to various complex non-linear models. DNN can be used as the skeletal structure of the first behavior prediction layer and the second behavior prediction layer.

[0095] Market - Complete Part Network: Corresponding to the first behavior prediction layer, it is used to predict the probability of account resource pre-advance behavior occurring when the media content is exposed.

[0096] Complete Part - Credit Granting Network: Corresponding to the second behavior prediction layer, it is used to predict the probability of credit granting behavior occurring when the account resource pre-advance behavior occurs.

[0097] Please refer to Figure 5 , which shows a schematic diagram of the process for generating the first behavior prediction result set of the embodiments of the present application. The above-mentioned account basic data of the first sample is input into the preset model for account behavior prediction to obtain the first prediction result set, including;

[0098] S501. Input the account basic data of the first sample into the above-mentioned feature extraction layer for feature extraction to obtain basic data features.

[0099] The basic data features can be understood as the input for the first behavior prediction layer and the second behavior prediction layer. Of course, in another embodiment, the account basic data of the first sample can be input into the feature extraction layer for feature extraction to obtain the basic data features corresponding to the first sample. The account basic data of the second sample can also be input into the feature extraction layer for feature extraction to obtain the basic data features corresponding to the second sample.

[0100] S502. Input the above basic data features into the first behavior prediction layer and the second behavior prediction layer respectively, and obtain the first prediction value of the first behavior and the first prediction value of the second behavior respectively.

[0101] According to the foregoing description, both the first behavior prediction layer and the second behavior prediction layer take the basic data features corresponding to the first sample as input, and can correspondingly output pctr1 and pcvr1. Of course, if both the first behavior prediction layer and the second behavior prediction layer take the basic data features corresponding to the second sample as input, they can correspondingly output pctr2 and pcvr2.

[0102] S503. Based on the product of the first prediction value of the first behavior and the first prediction value of the second behavior, obtain the first prediction value of the associated behavior.

[0103] According to the foregoing description, the first prediction value of the first behavior, the first prediction value of the second behavior, and the first prediction value of the associated behavior are respectively represented as pctr1, pcvr1, and pctcvr1. The second prediction value of the first behavior, the second prediction value of the second behavior, and the second prediction value of the associated behavior are respectively represented as pctr2, pcvr2, and pctcvr2. Among them, pctcvr1 = pctr1 * pcvr1; pctcvr2 = pctr2 * pcvr2.

[0104] If considering various loss terms and consistency differences generated by the first sample, several difference indication parameters can be determined first. Exemplarily, the following difference indication parameters can be obtained:

[0105]

[0106]

[0107] Then determine the relative constraint label, that is

[0108] Then the total loss obtained from various loss terms and consistency differences generated by the first sample Among them,

[0109] loss pctr1= crossEntropy(pctr1, label wanjian )

[0110] loss pcvr1 = crossEntropy(pcvr1, label wanjian-shouxin )

[0111] loss pctcvr1 = crossEntropy(pctcvr1, label shouxin )

[0112]

[0113]

[0114] where label wanjian-shouxin is the label determined based on label wanjian and label shouxin in the account behavior link of "delivering media content to the account group - resource advance behavior - credit granting behavior", and identifies the label wanjian when label shouxin = 1.

[0115] In the embodiments of the present application, they are all consistency differences, and the quantization target is the relative relationship of the behavior prediction values and whether it is consistent with the relative relationship of the credit rating. In this way, not only the purpose of multi-behavior node modeling, or full-link modeling, is achieved, but also the credit rating relationship is taken into account.

[0116] Please refer to Figure 6 , which shows a schematic flow diagram of the account behavior prediction method provided by the embodiments of the present application. An account behavior prediction method, the above method includes:

[0117] S601. Obtain the account basic data of the target account;

[0118] The meaning of the account basic data is similar to that in the previous text and will not be elaborated here. And the target account and its account basic data can come from the validation set or from the application stage of the account behavior prediction model.

[0119] S602. Input the above account basic data into the account behavior prediction model to obtain the probabilities that the above target account executes at least two behaviors in the preset account behavior link respectively, and the above account behavior prediction model is trained based on the aforementioned account behavior prediction model training method.

[0120] In the account behavior prediction model trained by the foregoing account behavior prediction model training method, the prediction accuracy of each behavior in the account behavior link has been improved. Still taking the account behavior link of "delivering media content to an account group - resource advance behavior - credit granting behavior" as an example, the probability of the resource advance behavior occurring after delivering a certain media content, that is, pctr3, and the probability of the account credit granting behavior occurring after the resource advance behavior occurs, that is, pcvr3, can be predicted. Further, the probability of the credit granting behavior occurring after delivering a certain media content, pctcvr3 = pctr3 * pcvr3, can also be predicted. In many of the current related technologies, only pctcvr3 or only pctr3, or only pcvr3 can be predicted, that is, only the probability of a certain behavior occurring can be predicted. However, in the embodiments of the present application, the probabilities of multiple behaviors occurring can be predicted. For example, pctr3, pcvr3, and pctcvr3 can be predicted simultaneously, and the accuracies of pctr3, pcvr3, and pctcvr3 are all higher than those of the related technologies.

[0121] In some feasible embodiments, several accounts with high pctcvr3 can be selected as the delivery accounts of the media content, so as to ensure a good media content delivery effect and maximize the occurrence of the resource advance behavior and the credit granting behavior.

[0122] The following is an embodiment of the device of the present application, which can be used to execute the method embodiment of the present application. For the details not disclosed in the device embodiment of the present application, please refer to the method embodiment of the present application.

[0123] Please refer to Figure 7 , which shows a block diagram of an account behavior prediction model training device provided by an embodiment of the present application. The device has the function of implementing the foregoing account behavior prediction model training method. The foregoing function can be implemented by hardware or by hardware executing corresponding software. The device can be a computer device or can be set in a computer device. The device may include:

[0124] A sample acquisition module 701, configured to acquire a first sample and a second sample. Both the first sample and the second sample include account basic data and a behavior level label corresponding to the last behavior in the account behavior link;

[0125] A training module 702, configured to perform the following operations:

[0126] Determine a relative constraint label, where the relative constraint label is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample;

[0127] Input the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes predicted values corresponding to multiple behaviors in the account behavior link;

[0128] Input the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set;

[0129] Determine at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the predicted values with corresponding relationships in the first prediction result set and the second prediction result set;

[0130] Adjust the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

[0131] In one embodiment, the first sample includes a behavior label set, and the behavior label set includes behavior labels corresponding to various behaviors that occur sequentially in the account behavior link. The training module 702 is configured to perform the following operations:

[0132] For any predicted value in the first prediction result set, determine a target behavior label related to the predicted value in the behavior label set, and determine the true value corresponding to the predicted value according to the target behavior label; determine a loss term according to the difference between the true value and the predicted value;

[0133] The adjusting the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model includes:

[0134] Adjust the parameters of the preset model according to each loss term and the consistency difference to obtain the account behavior prediction model.

[0135] In one embodiment, the account behavior link includes adjacent first behavior and second behavior. The first prediction result set includes a first predicted value of the first behavior, a first predicted value of the second behavior, and a first predicted value of an associated behavior. The first predicted value of the first behavior indicates the occurrence probability of the first behavior in the first prediction result set. The first predicted value of the second behavior indicates the occurrence probability of the second behavior in the first prediction result set. The first predicted value of the associated behavior indicates the occurrence probability of the associated behavior in the first prediction result set. The occurrence probability of the associated behavior characterizes the probability of sequentially occurring the first behavior and the second behavior;

[0136] Correspondingly, the second prediction result set includes a second predicted value of the first behavior, a second predicted value of the second behavior, and a second predicted value of the associated behavior.

[0137] In one embodiment, the above-mentioned training module 702 is configured to perform the following operations:

[0138] A first difference indication parameter determined according to the magnitude relationship between the first prediction value of the above first row and the second prediction value of the above first row;

[0139] A second difference indication parameter determined according to the magnitude relationship between the first prediction value of the above second row and the second prediction value of the above second row;

[0140] A third difference indication parameter determined according to the magnitude relationship between the first prediction value of the above associated behavior and the second prediction value of the above associated behavior.

[0141] In one embodiment, the above-mentioned consistency difference includes at least one of the following:

[0142] A first difference generated when the magnitude relationship indicated by the above first difference indication parameter is inconsistent with the magnitude relationship indicated by the above relative constraint label;

[0143] A second difference generated when the magnitude relationship indicated by the above second difference indication parameter is inconsistent with the magnitude relationship indicated by the above relative constraint label;

[0144] A third difference generated when the magnitude relationship indicated by the above third difference indication parameter is inconsistent with the magnitude relationship indicated by the above relative constraint label.

[0145] In one embodiment, the above first sample includes a behavior label set, the behavior label set includes a first behavior label corresponding to the above first behavior, and a second behavior label corresponding to the above second behavior, and the training module 702 is configured to perform the following operations:

[0146] Determine a loss term corresponding to the above first behavior according to the difference between the above first behavior label and the first prediction value of the above first behavior;

[0147] Or,

[0148] Determine a loss term corresponding to the above second behavior according to the difference between the above second behavior label and the first prediction value of the above second behavior;

[0149] Or,

[0150] Determine a loss term corresponding to the above associated behavior according to the difference between the true value of the associated behavior and the first prediction value of the above associated behavior, and the true value of the associated behavior is determined based on the above first behavior label and the above second behavior label;

[0151] Wherein, each of the above loss terms is used to adjust the parameters of the above preset model.

[0152] In one embodiment, the above-mentioned preset model includes a first behavior prediction layer and a second behavior prediction layer, and both the first behavior prediction layer and the second behavior prediction layer are connected to the feature extraction layer; the training module 702 is used to perform the following operations:

[0153] Input the account basic data of the first sample into the feature extraction layer for feature extraction to obtain basic data features;

[0154] Input the basic data features into the first behavior prediction layer and the second behavior prediction layer respectively to obtain the first prediction value of the first behavior and the first prediction value of the second behavior respectively;

[0155] Based on the product of the first prediction value of the first behavior and the first prediction value of the second behavior, obtain the first prediction value of the associated behavior.

[0156] In one embodiment, the above-mentioned account behavior link is the account behavior link in the resource interaction scenario. The account behavior link includes adjacent first behavior and second behavior. The first behavior is a resource advance behavior, and the second behavior is a credit-granting behavior. The grade label is used to indicate the credit-granting grade.

[0157] It should be noted that for the device provided in the above-mentioned embodiment, when realizing its functions, only the above-mentioned division of each functional module is used for illustration. In actual application, the above-mentioned functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above-mentioned embodiment and the method embodiment belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be repeated here.

[0158] Please refer to Figure 8 , which shows the block diagram of an account behavior prediction device provided by an embodiment of the present application. The device has the function of implementing the above-mentioned account behavior prediction method. The above-mentioned function can be realized by hardware or by hardware executing corresponding software. The device can be a computer device or can be set in a computer device. The device may include:

[0159] A data acquisition module 801, configured to acquire the account basic data of the target account;

[0160] A behavior prediction module 802, configured to input the account basic data into the account behavior prediction model to obtain the probabilities that the target account executes at least two behaviors in the preset account behavior link respectively. The account behavior prediction model is trained based on the aforementioned account behavior prediction model training method.

[0161] Please refer to Figure 9, which shows the structural block diagram of a computer device provided by an embodiment of the present application. The computer device can be a server for executing the above-mentioned account behavior prediction model training method or account behavior prediction method. Specifically:

[0162] The computer device 1000 includes a central processing unit (CPU) 1001, a system memory 1004 including a random access memory (RAM) 1002 and a read-only memory (ROM) 1003, and a system bus 1005 connecting the system memory 1004 and the central processing unit 1001. The computer device 1000 also includes a basic input / output system (I / O (Input / Output) system) 1006 for facilitating the transfer of information between various components within the computer, and a mass storage device 1007 for storing an operating system 1013, application programs 1014, and other program modules 1015.

[0163] The basic input / output system 1006 includes a display 1008 for displaying information and input devices 1009 such as a mouse, keyboard, etc. for user input. Both the display 1008 and the input devices 1009 are connected to the central processing unit 1001 through an input / output controller 1010 connected to the system bus 1005. The basic input / output system 1006 may also include an input / output controller 1010 for receiving and processing inputs from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1010 also provides output to a display screen, printer, or other types of output devices.

[0164] The mass storage device 1007 is connected to the central processing unit 1001 through a mass storage controller (not shown) connected to the system bus 1005. The mass storage device 1007 and its associated computer-readable medium provide non-volatile storage for the computer device 1000. That is to say, the mass storage device 1007 may include a computer-readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0165] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage technologies, CD-ROM, DVD (Digital Video Disc), or other optical storage, magnetic tape cartridges, tapes, disk storage, or other magnetic storage devices. Of course, those skilled in the art will know that computer storage media is not limited to the above several types. The above-mentioned system memory 1004 and mass storage device 1007 can be collectively referred to as memory.

[0166] According to various embodiments of the present application, the computer device 1000 can also run on a remote computer on the network through a network such as the Internet. That is, the computer device 1000 can be connected to the network 1012 through the network interface unit 1011 connected to the system bus 1005, or in other words, the network interface unit 1011 can also be used to connect to other types of networks or remote computer systems (not shown).

[0167] The above-mentioned memory further includes a computer program, which is stored in the memory and is configured to be executed by one or more processors to implement the above-mentioned account behavior prediction model training method or account behavior prediction method.

[0168] In an exemplary embodiment, a computer-readable storage medium is also provided. At least one instruction, at least one segment of program, code set, or instruction set is stored in the above-mentioned storage medium. When the at least one instruction, the at least one segment of program, the code set, or the instruction set is executed by a processor, the above-mentioned account behavior prediction model training method is implemented.

[0169] Specifically, the account behavior prediction model training method includes:

[0170] Obtain a first sample and a second sample. Both the first sample and the second sample include account basic data and a behavior level label corresponding to the end behavior of the account behavior link;

[0171] Determine a relative constraint label, which is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample;

[0172] Input the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes prediction values corresponding to multiple behaviors in the account behavior link;

[0173] Input the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set;

[0174] Determine at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the prediction values with corresponding relationships in the first prediction result set and the second prediction result set;

[0175] Adjust the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

[0176] In one embodiment, the first sample includes a behavior label set, and the behavior label set includes behavior labels corresponding to various behaviors that occur sequentially in the account behavior link. The method further includes:

[0177] For any prediction value in the first prediction result set, determine a target behavior label related to the prediction value in the behavior label set, and determine the true value corresponding to the prediction value according to the target behavior label; determine a loss term according to the difference between the true value and the prediction value;

[0178] The step of adjusting the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model includes:

[0179] Adjust the parameters of the preset model according to the loss terms and the consistency difference to obtain the account behavior prediction model.

[0180] In one embodiment, the account behavior link includes an adjacent first behavior and a second behavior. The first prediction result set includes a first prediction value of the first behavior, a first prediction value of the second behavior, and a first prediction value of an associated behavior. The first prediction value of the first behavior indicates the occurrence probability of the first behavior in the first prediction result set, the first prediction value of the second behavior indicates the occurrence probability of the second behavior in the first prediction result set, and the first prediction value of the associated behavior indicates the occurrence probability of the associated behavior in the first prediction result set. The occurrence probability of the associated behavior characterizes the probability of the sequential occurrence of the first behavior and the second behavior;

[0181] Correspondingly, the above second prediction result set includes a first row with a second prediction value, a second row with a second prediction value, and a second prediction value for the associated behavior.

[0182] In one embodiment, determining at least one behavior difference indication parameter includes at least one of the following:

[0183] A first difference indication parameter determined according to the magnitude relationship between the first prediction value of the first row and the second prediction value of the first row;

[0184] A second difference indication parameter determined according to the magnitude relationship between the first prediction value of the second row and the second prediction value of the second row;

[0185] A third difference indication parameter determined according to the magnitude relationship between the first prediction value of the associated behavior and the second prediction value of the associated behavior.

[0186] In one embodiment, the above consistency difference includes at least one of the following:

[0187] A first difference generated when the magnitude relationship indicated by the first difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label;

[0188] A second difference generated when the magnitude relationship indicated by the second difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label;

[0189] A third difference generated when the magnitude relationship indicated by the third difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label.

[0190] In one embodiment, the above first sample includes a behavior label set, the behavior label set includes a first behavior label corresponding to the first behavior and a second behavior label corresponding to the second behavior, and the method further includes:

[0191] Determining a loss term corresponding to the first behavior according to the difference between the first behavior label and the first prediction value of the first behavior;

[0192] Or,

[0193] Determining a loss term corresponding to the second behavior according to the difference between the second behavior label and the first prediction value of the second behavior;

[0194] Or,

[0195] Determining a loss term corresponding to the associated behavior according to the difference between the true value of the associated behavior and the first prediction value of the associated behavior, where the true value of the associated behavior is determined based on the first behavior label and the second behavior label;

[0196] Among them, each of the above loss terms is used to adjust the parameters of the above preset model.

[0197] In one embodiment, the above preset model includes a first behavior prediction layer and a second behavior prediction layer, and both the first behavior prediction layer and the second behavior prediction layer are connected to the feature extraction layer; inputting the account basic data of the first sample into the preset model for account behavior prediction to obtain a first prediction result set, including:

[0198] Input the account basic data of the first sample into the feature extraction layer for feature extraction to obtain basic data features;

[0199] Input the basic data features into the first behavior prediction layer and the second behavior prediction layer respectively to obtain a first prediction value of the first behavior and a first prediction value of the second behavior;

[0200] Based on the product of the first prediction value of the first behavior and the first prediction value of the second behavior, obtain the first prediction value of the associated behavior.

[0201] In one embodiment, the above account behavior link is an account behavior link in a resource interaction scenario, the above account behavior link includes adjacent first behavior and second behavior, the first behavior is a resource advance behavior, the second behavior is a credit granting behavior, and the above level label is used to indicate the credit level.

[0202] In an exemplary embodiment, there is also provided a computer-readable storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored, and when the at least one instruction, the at least one program, the code set or the instruction set is executed by a processor, the above account behavior prediction method is implemented.

[0203] Specifically, the account behavior prediction method includes:

[0204] Obtain the account basic data of the target account;

[0205] Input the above account basic data into the account behavior prediction model to obtain the probabilities that the target account executes at least two behaviors in the preset account behavior link respectively, and the above account behavior prediction model is trained based on the foregoing account behavior prediction model training method.

[0206] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives), or optical discs, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0207] In an exemplary embodiment, there is also provided a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned account behavior prediction model training method or account behavior prediction method.

[0208] It should be understood that the term "plurality" mentioned herein refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. In addition, the step numbers described in this article only exemplarily show a possible execution sequence between steps. In some other embodiments, the above steps may not be executed in the order of the numbers. For example, two steps with different numbers are executed simultaneously, or two steps with different numbers are executed in the reverse order of the illustration. The embodiments of the present application do not limit this.

[0209] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the function of the module or unit.

[0210] In addition, in the specific implementation manner of the present application, when it comes to data related to user information, etc., when the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0211] The above are only exemplary embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A method for training an account behavior prediction model, characterized in that, The method includes: Obtaining a first sample and a second sample, where both the first sample and the second sample include account basic data and a behavior level label corresponding to the last behavior in the account behavior link; Determining a relative constraint label, where the relative constraint label is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample; Inputting the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes predicted values corresponding to multiple behaviors in the account behavior link; Inputting the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set; Determining at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the predicted values with corresponding relationships in the first prediction result set and the second prediction result set; Adjusting the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

2. The method according to claim 1, characterized in that, The first sample includes a behavior label set, where the behavior label set includes behavior labels corresponding to multiple behaviors occurring in sequence in the account behavior link. The method further includes: For any predicted value in the first prediction result set, determining a target behavior label related to the predicted value in the behavior label set, and determining the true value corresponding to the predicted value according to the target behavior label; determining a loss term according to the difference between the true value and the predicted value; The adjusting the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model includes: Adjusting the parameters of the preset model according to each loss term and the consistency difference to obtain the account behavior prediction model.

3. The method according to claim 1 or 2, characterized in that, The account behavior link includes adjacent first behavior and second behavior. The first prediction result set includes a first predicted value of the first behavior, a first predicted value of the second behavior, and a first predicted value of an associated behavior. The first predicted value of the first behavior indicates the occurrence probability of the first behavior in the first prediction result set. The first predicted value of the second behavior indicates the occurrence probability of the second behavior in the first prediction result set. The first predicted value of the associated behavior indicates the occurrence probability of the associated behavior in the first prediction result set. The occurrence probability of the associated behavior characterizes the probability of the first behavior and the second behavior occurring in sequence; Correspondingly, the second prediction result set includes a second predicted value of the first behavior, a second predicted value of the second behavior, and a second predicted value of the associated behavior.

4. The method according to claim 3, characterized in that, The determining at least one behavior difference indication parameter includes at least one of the following: A first difference indication parameter determined according to the magnitude relationship between the first predicted value of the first behavior and the first predicted value of the first behavior; A second difference indication parameter determined according to the magnitude relationship between the first predicted value of the second behavior and the first predicted value of the second behavior; A third difference indication parameter determined according to the magnitude relationship between the first prediction value of the associated behavior and the second prediction value of the associated behavior.

5. The method according to claim 4, characterized in that, The consistency difference includes at least one of the following: A first difference generated when the magnitude relationship indicated by the first difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label; A second difference generated when the magnitude relationship indicated by the second difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label; A third difference generated when the magnitude relationship indicated by the third difference indication parameter is inconsistent with the magnitude relationship indicated by the relative constraint label.

6. The method according to claim 5, characterized in that, The first sample includes a behavior label set, the behavior label set includes a first behavior label corresponding to the first behavior and a second behavior label corresponding to the second behavior, and the method further includes: Determining a loss term corresponding to the first behavior according to the difference between the first behavior label and the first prediction value of the first behavior; Or, Determining a loss term corresponding to the second behavior according to the difference between the second behavior label and the first prediction value of the second behavior; Or, Determining a loss term corresponding to the associated behavior according to the difference between the true value of the associated behavior and the first prediction value of the associated behavior, where the true value of the associated behavior is determined based on the first behavior label and the second behavior label; Wherein, each of the loss terms is used to adjust the parameters of the preset model.

7. The method according to claim 3, characterized in that, The preset model includes a first behavior prediction layer and a second behavior prediction layer, and both the first behavior prediction layer and the second behavior prediction layer are connected to the feature extraction layer; The inputting the account basic data of the first sample into the preset model for account behavior prediction to obtain a first prediction result set includes: Inputting the account basic data of the first sample into the feature extraction layer for feature extraction to obtain basic data features; Inputting the basic data features into the first behavior prediction layer and the second behavior prediction layer respectively to obtain the first prediction value of the first behavior and the first prediction value of the second behavior respectively; Based on the product of the first prediction value of the first behavior and the first prediction value of the second behavior, obtaining the first prediction value of the associated behavior.

8. The method according to claim 1, wherein The account behavior link is an account behavior link in a resource interaction scenario, the account behavior link includes adjacent first behavior and second behavior, the first behavior is a resource advance behavior, the second behavior is a credit-granting behavior, and the level label is used to indicate the credit-granting level.

9. An account behavior prediction method, characterized in that The method includes: Obtaining the account basic data of the target account; Inputting the account basic data into an account behavior prediction model to obtain the probabilities of the target account executing at least two behaviors in a preset account behavior link respectively, where the account behavior prediction model is trained by the account behavior prediction model training method according to any one of claims 1 to 8.

10. An account behavior prediction model training device, characterized in that The device includes: A sample acquisition module, configured to acquire a first sample and a second sample, where both the first sample and the second sample include account basic data and a behavior level label corresponding to the end behavior of the account behavior link; A training module, configured to perform the following operations: Determine a relative constraint label, which is used to indicate the relative magnitude relationship between the behavior level label corresponding to the first sample and the behavior level label corresponding to the second sample; Input the account basic data of the first sample into a preset model for account behavior prediction to obtain a first prediction result set, where the first prediction result set includes predicted values corresponding to multiple behaviors in the account behavior link; Input the account basic data of the second sample into the preset model for account behavior prediction to obtain a second prediction result set; Determine at least one behavior difference indication parameter, where the behavior difference indication parameter characterizes the relative magnitude relationship between the predicted values with corresponding relationships in the first prediction result set and the second prediction result set; Adjust the parameters of the preset model according to the consistency difference between each behavior difference indication parameter and the relative constraint label to obtain an account behavior prediction model.

11. An account behavior prediction device, characterized in that The device includes: A data acquisition module, configured to acquire the account basic data of a target account; A behavior prediction module, configured to input the account basic data into an account behavior prediction model to obtain the probabilities of the target account executing at least two behaviors in a preset account behavior link, where the account behavior prediction model is trained based on the account behavior prediction model training method described in any one of claims 1 to 8.

12. A computer device, characterized in that The computer device includes a processor and a memory, and at least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the account behavior prediction model training method described in any one of claims 1 to 8, or the account behavior prediction method described in claim 9.

13. A computer-readable storage medium, characterized in that At least one instruction, at least one program, a code set or an instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the account behavior prediction model training method described in any one of claims 1 to 8, or the account behavior prediction method described in claim 9.

14. A computer program product, characterized in that The computer program product includes computer instructions, the computer instructions are stored in a computer-readable storage medium, a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to execute to implement the account behavior prediction model training method described in any one of claims 1 to 8, or the account behavior prediction method described in claim 9.

Images