Information pushing method and device and computer readable storage medium

Abstract

The invention relates to an information pushing method and device and a computer readable storage medium, and relates to the technical field of computer software. The method comprises: for each user group, according to user behavior data of the user group, based on a general probability prediction function and a time decay function, determining a click probability and an order placing probability of the user group to each message channel combination; for each message channel combination of the user group, determining an expected income of the message channel combination of the user group according to a preset click value, a preset order placing value and the click probability, order placing probability and push cost of the message channel combination; and performing message pushing according to the message channel combination corresponding to the maximum expected income, collecting user behavior feedback information, updating the click and order placing probability of each user group to each message channel combination by using a Bayesian formula, and then determining the expected income of the message channel combination of the user group again. According to the invention, the information pushing accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer software technology, and in particular to an information push method, device and computer-readable storage medium. Background Art

[0002] At present, in the air travel service platform, pushing flight dynamics, special air tickets, hotel discounts and other information to users has become a core means to improve conversion rate and customer stickiness. Existing push technologies are mostly based on users' browsing history or ticket purchase behavior, using fixed rules or static models for message distribution, which is difficult to adapt to the diversity and time sensitivity of user needs. There is a great deal of uncertainty in the push effect, resulting in a waste of push resources or insufficient conversion.

[0003] Therefore, there is an urgent need for an information push method to improve the operational efficiency and user experience of the air travel platform. Summary of the invention

[0004] Based on this, it is necessary to provide an information push method, device and computer-readable storage medium to address the above technical problems.

[0005] In a first aspect, an information push method is provided, the method comprising: Obtain user feature data and user behavior data of the user to be pushed; According to the user characteristic data, the users to be pushed are divided into multiple user groups; Combining message types of various information to be pushed and various push channels to obtain multiple message channel combinations; For each of the user groups, according to the user behavior data of the user group, based on a general probability prediction function and a time decay function, determine the click probability and order probability of the user group for each of the message channel combinations; For each of the message channel combinations of the user group, the expected revenue of the message channel combination of the user group is determined according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination, and the preset push cost of the message channel combination; Push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group; Based on the user behavior feedback information, the Bayesian formula is used to update the click probability and order probability of each user group for each message channel combination, and then the message channel combination for the user group is executed to determine the expected benefit of the message channel combination for the user group according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination.

[0006] As an optional implementation, the user characteristic data includes the user's gender, age, geographic location, interest preferences and historical purchase data.

[0007] As an optional implementation manner, dividing the users to be pushed into multiple user groups according to the user characteristic data includes: Generating a user feature vector for each user to be pushed according to the user feature data; Cluster analysis is performed on the user feature vectors based on a preset clustering algorithm to obtain multiple user groups.

[0008] As an optional implementation, the user behavior data includes click behavior data and order behavior data.

[0009] As an optional implementation mode, for each user group, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, the formula for determining the click probability and order probability of the user group for each message channel combination is: P click (G i , M j , C k )= [ ∑ t W(t)x i,j,k , t+1 ] / [ ∑ t W(t)x i,j,k , t+2 ]; P order (G i , M j , C k )= [ ∑ t W(t)y i,j,k , t+1 ] / [ ∑ t W(t)y i,j,k , t+2 ]; in, Pclick (G i , M j , C k ) For user groups G i For message type M j and push channels C k The click probability of the message channel combination composed of W(t) is the time decay function, x i,j,k , t+1 is the number of clicks at the t+1th time step after time t, x i,j,k , t+2 is the number of clicks at the t+2th time step after time t, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of y i,j,k , t+1 is the number of t+1 The number of orders placed in a time step, y i,j,k , t+2 is the number of t+2 The number of orders placed in the time step.

[0010] As an optional implementation mode, for each message channel combination of the user group, according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination, the formula for determining the expected benefit of the message channel combination of the user group is: E i,j,k =P click (G i , M j , C k)×V click +P order (G i , M j , C k )×V order −Cost(M j ,C k ) ; Among them, E i,j,k For user groups G i For message type M j and push channels C k The expected return of the message channel combination composed of P click (G i , M j , C k ) For user groups G i For message type M j and push channels C k The click probability of the message channel combination composed of V click is the click value, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of V order is the order value, Cost(M j ,C k ) For message type M j And push channels Ck The push cost of the message channel combination.

[0011] As an optional implementation manner, the user behavior feedback information includes click rate, non-click rate, order rate and non-order rate. The formula for updating the click probability and order probability of each user group for each message channel combination using the Bayesian formula based on the user behavior feedback information is: P′ click (G i , M j , C k )= [ P click (G i , M j , C k )×P feedback|click ] / { P click (G i , M j , C k )× P feedback|click +[1-P click (G i , M j , C k )]×P feedback|unclick}; P′ order (G i , M j , C k )= [ P order (G i , M j , C k )×P feedback|order ] / { P order(G i , M j , C k )× P feedback|order +[1-P order (G i , M j , C k )]×P feedback|unorder}; in, P′ click (G i , M j , C k ) For the updated user group G i For message type M j and push channels C k The click probability of the message channel combination composed of P click (G i , M j , C k ) For the user group before the update G i For message type M j and push channels C k The click probability of the message channel combination composed of P feedback|click is the click rate, P feedback|unclick is the non-click rate, P′ order (G i , M j , C k ) For the updated user group G i For message type M j and push channels Ck The order probability of the message channel combination composed of P order (G i , M j , C k ) For the user group before the update G i For message type M j and push channels C k The order probability of the message channel combination composed of P feedback|order is the order rate, P feedback|unorder The rate of unordered orders.

[0012] In a second aspect, an information push device is provided, the device comprising: The collection module is used to obtain the user feature data and user behavior data of the user to be pushed; A classification module, used for classifying the users to be pushed into multiple user groups according to the user characteristic data; A combination module is used to combine the message types of various information to be pushed and various push channels to obtain multiple message channel combinations; A processing module, for determining, for each of the user groups, the click probability and order probability of the user group for each of the message channel combinations based on the user behavior data of the user group, based on a general probability prediction function and a time decay function; The processing module is further used to determine, for each message channel combination of the user group, the expected revenue of the message channel combination of the user group according to a preset click value, a preset order value, a click probability, an order probability of the message channel combination, and a preset push cost of the message channel combination; A push module, used to push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group; The processing module is also used to update the click probability and order probability of each user group for each message channel combination based on the user behavior feedback information using the Bayesian formula, and then execute the step of determining the expected benefit of the message channel combination for the user group based on the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination for the user group.

[0013] As an optional implementation manner, the classification module is further used to generate a user feature vector of each user to be pushed according to the user feature data; Cluster analysis is performed on the user feature vectors based on a preset clustering algorithm to obtain multiple user groups.

[0014] In a third aspect, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and when the processor executes the computer program, the method steps described in any one of the first aspects are implemented.

[0015] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method steps as described in any one of the first aspects are implemented.

[0016] The present application provides an information push method, device and computer-readable storage medium. The technical solution provided by the embodiments of the present application brings at least the following beneficial effects, the method comprising: obtaining user feature data and user behavior data of users to be pushed; dividing the users to be pushed into multiple user groups according to the user feature data; combining the message types of various types of information to be pushed and various push channels to obtain multiple message channel combinations; for each of the user groups, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, determining the click probability and order probability of the user group for each of the message channel combinations; for each of the message channel combinations of the user group, according to the preset click value, the preset order value and The expected benefit of the message channel combination for the user group is determined based on the click probability, order probability and preset push cost of the message channel combination of the message channel combination; the message is pushed to the user group according to the message channel combination corresponding to the maximum expected benefit, and the user behavior feedback information of each user group is collected; based on the user behavior feedback information, the click probability and order probability of each user group for each message channel combination are updated using the Bayesian formula, and then the steps of determining the expected benefit of the message channel combination for the user group are performed based on the preset click value, the preset order value and the click probability, order probability and preset push cost of the message channel combination for each message channel combination for the user group. The information push method provided in the present application proposes a complete closed-loop solution based on user clustering, probability modeling, value evaluation and feedback update mechanism to address the problems of low user response prediction accuracy, weak behavior change modeling capability, and low push resource utilization efficiency. Compared with the traditional static label grouping method, it can more accurately capture the nonlinear relationship between user features, thereby improving the accuracy of subsequent predictions. The time decay function is introduced to weight the historical behavior data, so that recent behavior has a greater influence weight in the modeling and better reflects the current interest status. At the same time, combined with the general probability prediction function, the user group's response tendency to specific push content under current conditions can be accurately modeled. The probability model is dynamically updated using the Bayesian formula, which has real-time adaptive learning capabilities and can continuously optimize the prediction accuracy of user group behavior, thereby further improving push accuracy and long-term effects. Therefore, this application effectively solves the problems of existing information push systems in terms of inaccurate group response modeling, insufficient resource utilization, and unstable push revenue, and significantly improves the information distribution efficiency and commercial conversion effects in scenarios such as air travel.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of an information push method provided in an embodiment of the present application; Figure 2 A flowchart of a method for dividing user groups provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an information push device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] The following will describe in detail an information push method provided by an embodiment of the present application in combination with a specific implementation method. Figure 1 A flowchart of an information push method provided in an embodiment of the present application, such as Figure 1 As shown, the specific steps are as follows: S101, obtaining user feature data and user behavior data of the user to be pushed.

[0022] In implementation, the system can extract the user feature data and user behavior data of the user to be pushed from the user database. The user feature data may include static features such as gender, age, interest preferences, historical orders, and historical browsing. At the same time, the system can also combine the behavior log system to obtain the user's click records, order behavior, length of stay, and other user behavior data. For example, a user A has the characteristics of "male, 25 years old, travel enthusiast, and has purchased tickets twice in the past three months", as well as the behavior data of "clicked on the special ticket for area B three times and placed an order once".

[0023] S102: Divide the users to be pushed into multiple user groups according to user characteristic data.

[0024] During implementation, the system can divide the users to be pushed into several user groups according to user characteristic data, such as "student travel type", "price sensitive type", "high-frequency business type", etc.

[0025] As an optional implementation, Figure 2A flowchart of a method for dividing user groups provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the specific steps of dividing the users to be pushed into multiple user groups according to the user characteristic data in S102 are as follows: S201, generating a user feature vector of each user to be pushed according to user feature data.

[0026] In implementation, the system can construct a multi-dimensional numerical user feature vector according to a preset structure based on user feature data including but not limited to gender, age, geographic location, interest preferences, travel consumption history, device type, etc. Categorical features can be represented by one-hot encoding, frequency encoding or embedded vectors, and numerical features can be normalized.

[0027] S202: Perform cluster analysis on user feature vectors based on a preset clustering algorithm to obtain multiple user groups.

[0028] In implementation, the system can select clustering algorithms according to business scenarios, such as K-Means (fast and efficient for large samples), DBSCAN (suitable for discovering non-spherical clusters), GMM (Gaussian mixture model, which can output probabilistic group membership), etc. The feature similarity between users is automatically determined during the clustering process, and the feature vectors can be divided into several clusters by calculating the Euclidean distance or cosine similarity. For example, in an air travel scenario, after clustering 100,000 users to be pushed, five user groups are obtained, namely frequent travelers, students, price-sensitive users, high-end business travelers, and inactive users.

[0029] As an optional implementation method, the user characteristic data includes but is not limited to the user's gender, age, geographic location, interest preferences and historical purchase data.

[0030] S103, combining the message types of various types of information to be pushed and various push channels to obtain multiple message channel combinations.

[0031] In implementation, the system can combine the information to be pushed (such as "low-priced tickets for return trips after the holidays", "50% discount on high-end hotels", etc.) with available push channels (such as APP pop-up windows, SMS, and emails) to generate a message such as [(M 1 ,C 1 ),(M 1 ,C 2 ),(M 2 ,C 1 ),…] message channel combination.

[0032] S104, for each user group, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, determine the click probability and order probability of the user group for each message channel combination.

[0033] In implementation, based on historical behavior data, general probability prediction functions (such as logistic regression, collaborative filtering, and deep FM) can be used to predict the click probability and order probability of each combination of user groups. Considering that user behavior (such as clicks and orders) can change over time, adding a time decay function can make the user's recent behavior have a greater impact on the probability calculation, while the behavior in the past has a smaller impact. For example, the click probability predicted by the "price-sensitive group" for the message channel combination of "special tickets + APP notification" is 15%, and the order probability is 5%.

[0034] As an optional implementation, user behavior data includes but is not limited to click behavior data and order behavior data.

[0035] As an optional implementation, in S104, for each user group, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, the formula for determining the click probability and order probability of the user group for each message channel combination is: P click (G i , M j , C k )= [ ∑ t W(t)x i,j,k , t+1 ] / [ ∑ t W(t)x i,j,k , t+2 ].

[0036] P order (G i , M j , C k )= [ ∑ t W(t)y i,j,k , t+1 ] / [ ∑ t W(t)y i,j,k , t+2 ].

[0037] in,P click (G i , M j , C k ) For user groups G i For message type M j and push channels C k The click probability of the message channel combination composed of W(t) is the time decay function, x i,j,k , t+1 is the number of clicks at the t+1th time step after time t, x i,j,k , t+2 is the number of clicks at the t+2th time step after time t, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of y i,j,k , t+1 is the number of t+1 The number of orders placed in a time step, y i,j,k , t+2 is the number of t+2 The number of orders placed in the time step.

[0038] In implementation, the system can collect G i Click behavior data and order behavior data in different time periods, combined with time decay function W(t) , for a user group to respond to a specific message type in a future time window M j and push channels C k The response behavior of is weighted probability estimated. Among them, the time decay function W(t) The form can be 1 / (t+1) or e -λt, which is used to reduce the impact of past behavior on the calculation results and enhance the model's ability to respond to recent trends. In the formula, the numerator is used to weight the aggregation of recent click / order behaviors, and the denominator is the behavior in the next stage to form a normalized probability estimate. For example: Taking the "price-sensitive user group" as an example, G 1 ", for example, the travel platform hopes to predict the user group's interest in "special price tickets for region B" ( M 2 ) Through the "APP pop-up window" ( C 1 ) The click probability and order probability of the push. Table 1 is a user behavior data table provided by the embodiment of the present application, as shown in Table 1, and the details are as follows (unit: times): Table 1

[0039] Assume that the time decay function W(t) = 1 / (t+1) ,but: P click (G 1 , M 2 , C 1 ) = (1×120+0.5×90+0.333×60) / (1×200+0.5×150+0.333×100)≈0.6.

[0040] P order (G 1 , M 2 , C 1 ) = (1×15+0.5×12+0.333×8) / (1×30+0.5×24+0.333×16)≈0.5.

[0041] By introducing time-decay weighted processing for click / order behaviors, we can not only improve the modeling capabilities of current interest states, but also effectively filter out early behavioral noise, and achieve accurate prediction of the responsiveness of different user groups and message channel combinations.

[0042] S105, for each message channel combination of the user group, determine the expected benefit of the message channel combination of the user group according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination.

[0043] During implementation, the system can calculate the expected profit of each combination based on the pre-set click value (such as one click can bring a yuan in profit), order value (such as average profit of b yuan per order), and push cost (such as c yuan / message for SMS and d yuan for pop-up window).

[0044] As an optional implementation, in S105, for each message channel combination of the user group, according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination, and the preset push cost of the message channel combination, the formula for determining the expected benefit of the message channel combination of the user group is: E i,j,k =P click (G i , M j , C k )×V click +P order (G i , M j , C k )×V order −Cost(M j ,C k ) .

[0045] Among them, E i,j,k For user groups G i For message type M j and push channels C k The expected return of the message channel combination composed of P click (G i , M j , C k ) For user groups Gi For message type M j and push channels C k The click probability of the message channel combination composed of V click is the click value, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of V order is the order value, Cost(M j ,C k ) For message type M j And push channels C k The push cost of the message channel combination.

[0046] In implementation, the system can be based on the click probability obtained in the previous stage. P click (G i , M j , C k ) And the probability of placing an order P order (G i , M j , C k ) , combined with the value that each click behavior preset by the platform can bring ( V click ) and the actual revenue from each order ( V order ), minus the channel cost of each push operation Cost(M j ,C k ), to calculate the expected benefit of each message channel combination under a specific user group E i,j,k For example, the platform is evaluating the following combination in the "price-sensitive user group ( G 1 )”, the expected return under the message type M 2 Special air tickets for region B, push channels C 1 Pop-up window for APP, P click (G 1 , M 2 , C 1 )= 0.6, P order (G 1 , M 2 , C 1 )= 0.5, V click is 0.2, V order is 100, Cost(M 2 ,C 1 ) is 0.05, substitute it into the formula to calculate E 1,2,1 = 0.6×0.2+0.5×100−0.05=50.07. For example, the user group G 1 For the SMS push channel C 2 The response is: P click (G 1 , M 2 , C 2 )= 0.3, P order (G 1 , M 2 , C 2 )= 0.6, Cost(M 2 ,C2 ) is 0.1, substitute it into the formula to calculate E 1,2,2 = 0.3×0.2+0.6×100−0.1=59.96. At this time, the system can choose a combination with a higher expected return, that is, choose to push through the SMS channel. The expected return evaluation mechanism provided in the embodiment of the present application comprehensively considers the user behavior probability (click probability and order probability) and the platform revenue elements (click revenue, order profit, and push cost), which can quantify the benefits of different message channel combinations and realize revenue-driven push decisions, which is significantly better than traditional optimization strategies based only on click probability. At the same time, the mechanism supports dynamic adjustment of parameters, such as improving the click / order value valuation during peak hours to adapt to the flexibility of business strategies.

[0047] S106, pushing messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collecting user behavior feedback information of each user group.

[0048] In implementation, the system can select the message channel combination with the highest expected return for each user group to push, and can collect user subsequent behaviors through embedding points, such as whether to click, whether to place an order, length of stay, etc. For example, the "price-sensitive group" received an APP pop-up push of "special tickets for area B", with an actual click rate of 18% and an order rate of 4%.

[0049] S107, based on the user behavior feedback information, using the Bayesian formula to update the click probability and order probability of each user group for each message channel combination, and then executing S105.

[0050] In implementation, user behavior feedback information (such as observed clicks / non-clicks, whether an order is placed) is used as posterior data, combined with the previously predicted prior probability, and the click probability and order probability of each user group for each message channel combination are dynamically adjusted through the Bayesian formula, and then the expected return calculation (S105) is re-executed to optimize the subsequent push strategy. For example, if the feedback finds that the response of the "business user group" to SMS push is far lower than expected, the corresponding click probability and order probability are adjusted, so that the system is more likely to choose APP channel push next time.

[0051] As an optional implementation, the user behavior feedback information in S105 includes click rate, non-click rate, order rate and non-order rate. Based on the user behavior feedback information, the formula for updating the click probability and order probability of each user group for each message channel combination using the Bayesian formula is: P′ click (G i , Mj , C k )= [ P click (G i , M j , C k )×P feedback|click ] / { P click (G i , M j , C k )× P feedback|click +[1-P click (G i , M j , C k )]×P feedback|unclick}。

[0052] P′ order (G i , M j , C k )= [ P order (G i , M j , C k )×P feedback|order ] / { P order (G i , M j , C k )× P feedback|order +[1-P order (G i , M j , C k)]×P feedback|unorder}.

[0053] in, P′ click (G i , M j , C k ) For the updated user group G i For message type M j and push channels C k The click probability of the message channel combination composed of P click (G i , M j , C k ) For the user group before the update G i For message type M j and push channels C k The click probability of the message channel combination composed of P feedback|click is the click rate, P feedback|unclick is the non-click rate, P′ order (G i , M j , C k ) For the updated user group G i For message type M j and push channels C k The order probability of the message channel combination composed of P order (G i , M j , C k ) For the user group before the update G i For message type M j and push channelsC k The order probability of the message channel combination composed of P feedback|order is the order rate, P feedback|unorder The rate of unordered orders.

[0054] In practice, after completing the message push, the system can collect user behavior feedback information of each user group on different message types and push channel combinations in real time, and use the Bayesian formula to dynamically modify the click probability and order probability, so that the model can adaptively reflect the changes in user interests or the fluctuations in message effects, thereby improving the prediction accuracy. For example: the current system's initial prediction for a certain combination (special price ticket + APP notification) under the "price-sensitive user group" is: click probability (prior) P click (G i , M j , C k ) =0.6, order probability (prior) P order (G i , M j , C k ) =0.5. After push, the system calculates the user behavior feedback information of this message channel combination as follows: click rate P feedback|click =0.65, no click rate P feedback|unclick =0.35, order rate P feedback|order =0.48, unordered rate P feedback|unorder =0.52. Substitute the Bayesian formula into the update, and the updated click probability is P′ click (G i , M j , C k ) =0.6×0.65 / (0.6×0.65+0.4×0.35)≈0.7358. Updated order probability P′ order (G i , M j , Ck )= 0.5×0.48 / (0.5×0.48+0.5×0.52)≈0.48.

[0055] The embodiment of the present application provides an information push method, the method comprising: obtaining user feature data and user behavior data of the user to be pushed. According to the user feature data, the users to be pushed are divided into multiple user groups. The message types of various types of information to be pushed and various push channels are combined to obtain multiple message channel combinations. For each user group, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, the click probability and order probability of the user group for each message channel combination are determined. For each message channel combination of the user group, according to the preset click value, the preset order value, the click probability of the message channel combination, the order probability and the preset push cost of the message channel combination, the expected benefit of the message channel combination of the user group is determined. According to the message channel combination corresponding to the maximum expected benefit, the message is pushed to the user group, and the user behavior feedback information of each user group is collected. Based on user behavior feedback information, the Bayesian formula is used to update the click probability and order probability of each user group for each message channel combination, and then the steps of determining the expected benefits of the message channel combination of the user group are performed according to the preset click value, the preset order value, the click probability of the message channel combination, the order probability and the preset push cost of the message channel combination for each message channel combination for the user group. The information push method provided in the embodiment of the present application proposes a complete closed-loop solution based on user clustering, probability modeling, value assessment and feedback update mechanism to address the problems of low user response prediction accuracy, weak behavior change modeling ability, and low push resource utilization efficiency in the prior art. Compared with the traditional static label grouping method, it can more accurately capture the nonlinear relationship between user features, thereby improving the accuracy of subsequent predictions. The time decay function is introduced to weight the historical behavior data, so that recent behavior has a greater influence weight in the modeling and better reflects the current interest state. At the same time, combined with the general probability prediction function, the response tendency of the user group to specific push content under current conditions can be accurately modeled. The Bayesian formula is used to dynamically update the probability model, which has real-time adaptive learning capabilities and can continuously optimize the prediction accuracy of user group behavior, thereby further improving the push accuracy and long-term effect. Therefore, this application effectively solves the problems of the existing information push system in terms of inaccurate group response modeling, insufficient resource utilization, and unstable push revenue, and significantly improves the information distribution efficiency and commercial conversion effect in scenarios such as air travel.

[0056] It should be understood that although Figure 1 to Figure 2The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 to Figure 2 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0057] It can be understood that the same / similar parts between the various embodiments of the above method in this specification can refer to each other, and each embodiment focuses on the differences from other embodiments. For related points, please refer to the description of other method embodiments.

[0058] The present application also provides an information push device, such as Figure 3 As shown, the device comprises: The collection module 310 is used to obtain user feature data and user behavior data of the user to be pushed.

[0059] The classification module 320 is used to classify the users to be pushed into multiple user groups according to the user characteristic data.

[0060] The combination module 330 is used to combine the message types of various types of information to be pushed and various push channels to obtain multiple message channel combinations.

[0061] The processing module 340 is used to determine, for each user group, the click probability and order probability of the user group for each message channel combination according to the user behavior data of the user group, based on the general probability prediction function and the time decay function.

[0062] The processing module 340 is also used to determine the expected benefit of the message channel combination for the user group based on the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination of the user group.

[0063] The push module 350 is used to push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group.

[0064] The processing module 340 is also used to update the click probability and order probability of each user group for each message channel combination based on the user behavior feedback information using the Bayesian formula, and then execute the steps of determining the expected benefit of the message channel combination for the user group based on the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination for the user group.

[0065] As an optional implementation, the classification module 320 is specifically configured to generate a user feature vector of each user to be pushed according to the user feature data.

[0066] Based on a preset clustering algorithm, cluster analysis is performed on user feature vectors to obtain multiple user groups.

[0067] The embodiment of the present application provides an information push device, which includes: a collection module 310, which is used to obtain user feature data and user behavior data of users to be pushed. A classification module 320, which is used to divide users to be pushed into multiple user groups according to user feature data. A combination module 330, which is used to combine message types and various push channels of various types of information to be pushed to obtain multiple message channel combinations. A processing module 340, which is used to determine the click probability and order probability of each message channel combination of the user group based on the user behavior data of the user group, based on the general probability prediction function and the time decay function. The processing module 340 is also used to determine the expected benefit of the message channel combination of the user group according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination of the user group. A push module 350, which is used to push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group. The processing module 340 is also used to update the click probability and order probability of each user group for each message channel combination based on the user behavior feedback information using the Bayesian formula, and then execute the steps of determining the expected benefits of the message channel combination of the user group according to the preset click value, the preset order value, the click probability of the message channel combination, the order probability and the preset push cost of the message channel combination for each message channel combination of the user group. The information push device provided in the embodiment of the present application proposes a complete closed-loop solution based on user clustering, probability modeling, value assessment and feedback update mechanism to address the problems of low user response prediction accuracy, weak behavior change modeling ability, and low push resource utilization efficiency in the prior art. Compared with the traditional static label grouping method, it can more accurately capture the nonlinear relationship between user features, thereby improving the accuracy of subsequent predictions. The time decay function is introduced to weight the historical behavior data, so that the recent behavior has a greater influence weight in the modeling and better reflects the current interest state. At the same time, combined with the general probability prediction function, the response tendency of the user group to specific push content under current conditions can be accurately modeled. The Bayesian formula is used to dynamically update the probability model, which has real-time adaptive learning capabilities and can continuously optimize the prediction accuracy of user group behavior, thereby further improving the push accuracy and long-term effect. Therefore, this application effectively solves the problems of the existing information push system in terms of inaccurate group response modeling, insufficient resource utilization, and unstable push revenue, and significantly improves the information distribution efficiency and commercial conversion effect in scenarios such as air travel.

[0068] For the specific definition of the information push device, please refer to the definition of the information push method above, which will not be repeated here. Each module in the above-mentioned information push device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0069] In one embodiment, a computer-readable storage medium stores a computer program, which implements the steps of the above-mentioned information push method when executed by a processor.

[0070] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0072] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0073] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0074] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0075] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. An information push method, characterized in that: The method comprises: Obtain user feature data and user behavior data of the user to be pushed; According to the user characteristic data, the users to be pushed are divided into multiple user groups; Combining message types of various information to be pushed and various push channels to obtain multiple message channel combinations; For each of the user groups, according to the user behavior data of the user group, based on a general probability prediction function and a time decay function, determine the click probability and order probability of the user group for each of the message channel combinations; For each of the message channel combinations of the user group, the expected revenue of the message channel combination of the user group is determined according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination, and the preset push cost of the message channel combination; Push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group; Based on the user behavior feedback information, the Bayesian formula is used to update the click probability and order probability of each user group for each message channel combination, and then the message channel combination for the user group is executed to determine the expected benefit of the message channel combination for the user group according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination.

2. The method according to claim 1, characterized in that: The user characteristic data includes the user's gender, age, geographic location, interest preferences and historical purchase data.

3. The method according to claim 1, characterized in that The dividing the users to be pushed into multiple user groups according to the user characteristic data includes: Generating a user feature vector for each user to be pushed according to the user feature data; The user feature vectors are clustered and analyzed based on a preset clustering algorithm to obtain multiple user groups.

4. The method according to claim 1, characterized in that The user behavior data includes click behavior data and order behavior data.

5. The method according to claim 1, characterized in that For each user group, according to the user behavior data of the user group, based on the general probability prediction function and the time decay function, the formula for determining the click probability and order probability of the user group for each message channel combination is: P click (G i , M j , C k )= [ ∑ t W(t)x i,j,k , t+1 ] / [ ∑ t W(t)x i,j,k , t+2 ]； P order (G i , M j , C k )= [ ∑ t W(t)y i,j,k , t+1 ] / [ ∑ t W(t)y i,j,k , t+2 ]； in, P click (G i , M j , C k ) For user groups G i For message type M j and push channels C k The click probability of the message channel combination composed of W(t) is the time decay function, x i,j,k , t+1 is the number of clicks at the t+1th time step after time t, x i,j,k , t+2 is the number of clicks at the t+2th time step after time t, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of y i,j,k , t+1 is the number of t+1 The number of orders placed in a time step, y i,j,k , t+2 is the number of t+2 The number of orders placed in the time step.

6. The method according to claim 1, characterized in that The formula for determining the expected revenue of the message channel combination for the user group according to the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination for the user group is: E i,j,k =P click (G i , M j , C k )×V click +P order (G i , M j , C k )×V order −Cost(M j ,C k ) ； Among them, E i,j,k For user groups G i For message type M j and push channels C k The expected return of the message channel combination composed of P click (G i , M j , C k ) For user groups G i For message type M j and push channels C k The click probability of the message channel combination composed of V click is the click value, P order (G i , M j , C k ) For user groups G i For message type M j and push channels C k The order probability of the message channel combination composed of V order is the order value, Cost(M j ,C k ) Message Type M j And push channels C k The push cost of the message channel combination.

7. The method according to claim 1, characterized in that The user behavior feedback information includes click rate, non-click rate, order rate and non-order rate. The formula for updating the click probability and order probability of each user group for each message channel combination using the Bayesian formula based on the user behavior feedback information is: P′ click (G i , M j , C k )= [ P click (G i , M j , C k )×P feedback|click ] / { P click (G i , M j , C k )×P feedback|click + [1-P click (G i , M j , C k )]×P feedback|unclick}； P′ order (G i , M j , C k )= [ P order (G i , M j , C k )×P feedback|order ] / { P order (G i , M j , C k )×P feedback|order + [1-P order (G i , M j , C k )]×P feedback|unorder}； in, P′ click (G i , M j , C k ) For the updated user group G i For message type M j and push channels C k The click probability of the message channel combination composed of P click (G i , M j , C k ) For the user group before the update G i For message type M j and push channels C k The click probability of the message channel combination composed of P feedback|click is the click rate, P feedback|unclick is the non-click rate, P′ order (G i , M j , C k ) For the updated user group G i For message type M j and push channels C k The order probability of the message channel combination composed of P order (G i , M j , C k ) For the user group before the update G i For message type M j and push channels C k The order probability of the message channel combination composed of P feedback|order is the order rate, P feedback|unorder The rate of unordered orders.

8. An information push device, characterized in that: The device comprises: The collection module is used to obtain the user feature data and user behavior data of the user to be pushed; A classification module, used for classifying the users to be pushed into multiple user groups according to the user characteristic data; A combination module is used to combine the message types of various information to be pushed and various push channels to obtain multiple message channel combinations; A processing module, for determining, for each of the user groups, the click probability and order probability of the user group for each of the message channel combinations based on the user behavior data of the user group, based on a general probability prediction function and a time decay function; The processing module is further used to determine, for each message channel combination of the user group, the expected revenue of the message channel combination of the user group according to a preset click value, a preset order value, a click probability, an order probability of the message channel combination, and a preset push cost of the message channel combination; A push module, used to push messages to the user group according to the message channel combination corresponding to the maximum expected benefit, and collect user behavior feedback information of each user group; The processing module is also used to update the click probability and order probability of each user group for each message channel combination based on the user behavior feedback information using the Bayesian formula, and then execute the step of determining the expected benefit of the message channel combination for the user group based on the preset click value, the preset order value, the click probability, the order probability of the message channel combination and the preset push cost of the message channel combination for each message channel combination for the user group.

9. The device according to claim 8, characterized in that The classification module is further used to generate a user feature vector for each user to be pushed according to the user feature data; The user feature vectors are clustered and analyzed based on a preset clustering algorithm to obtain multiple user groups.

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

Images