Method and device for predicting user payment willingness, storage medium and electronic device

Abstract

The present disclosure belongs to the technical field of fare search, and provides a user payment willingness prediction method and device, a storage medium and an electronic device. The method comprises: generating corresponding session representations based on feature representations of a plurality of flight items; evolving different time-level preferences of a target user based on a session representation and a feature representation evolution network to obtain a preference feature representation of the target user; determining a payment willingness probability of the target user for a target flight in all flights based on the preference feature representation of the target user and a probability distribution; and predicting the payment willingness of the target user for the target flight based on the payment willingness probability of the target flight to obtain a prediction result. The prediction method introduces a feature evolution mechanism of the target user, so that the flight information interacting with the target user is retained, the preference feature of the target user obtained is more comprehensive, the final prediction result is more accurate, and the prediction result is more in line with the actual situation of the target user.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of fare search, and particularly relates to a user payment willingness prediction method and device, a storage medium and an electronic device. BACKGROUND

[0002] User payment willingness prediction is to predict future user flight demand and payment willingness based on different data processing points before flight departure, as a basis for flight seat quantity and price control. The concept of willingness to pay comes from the field of economics, and the meaning of willingness to pay is that the consumer individual is willing to pay the highest price for goods or services. In the field of revenue management, the most effective way to increase airline revenue is to set the willingness to pay price that different customers can accept.

[0003] Early research on willingness to pay mainly reflects on the customer's buy-up rate. Buy-up is defined as the behavior of customers transferring demand from a closed cabin to a higher price cabin, and researchers have improved the traditional EMSRb algorithm and proposed a new heuristic algorithm. Another study simulated the algorithm in the PODS (Passenger Origin-Destination Simulator) system, but did not provide a method to estimate the customer buy-up rate. Based on the concept of buy-up, a Q-forecasting prediction algorithm is proposed to estimate the demand for the lowest cabin (referred to as Q-class) and the probability of buying up to a higher cabin. Further research defines the two different demands as price-oriented demand and product-oriented demand, and designs an HF (Hybrid forecasting) method to estimate and predict the two demands respectively. There is also research to describe the customer's buy-up behavior through linear regression and to predict the booking demand of each cabin.

[0004] The WTP of these methods is mainly reflected in the demand function. For example, the demand function of a customer for a given price can be expressed as the probability that the customer's WTP is not less than the given price. To solve the problem of the estimation of demand parameters, some studies have introduced market share as an external variable to limit the degree of freedom between the arrival rate and the demand parameters of the MNL (Multinomial Logit) model. Another study divides the estimation task of the arrival rate and the demand parameters of the selection model into a two-step estimation task. Some studies have given the conditions for the uniqueness and identifiability of the parameter estimation of the MNL model, and proposed a MM (Minorization-maximization) demand estimation algorithm. The above studies are all based on the MLE (Maximum likelihood estimation) method to model the parameter estimation task. Some scholars have also chosen other parameter estimation methods to describe the demand parameter estimation task in different scenarios. For example, some studies have proposed a LM (Loss Minimization) algorithm based on MIP (Mixed-Integer Programming) to solve the arrival rate and the parameters of the MNL model simultaneously. Some studies have considered the endogeneity problem between the control behavior of the airline company and the observed sales data, and proposed a two-step GMM (Generalized Method of Moments) estimation method. Based on the EM (Expectation Maximization) algorithm, some studies have proposed an estimation method for Markov chain selection models. Another study describes the estimation task as a constrained convex optimization problem and applies the Frank-Wolfe algorithm to solve it. Some studies have proposed a new dynamic pricing strategy, which is different from other methods in that it decomposes demand into the level of individual customers, each customer has a different WTP, and the WTP of the customer group forms a customer group WTP distribution.

[0005] In recent years, domestic scholars have also begun to combine the actual operation data of airlines to predict the willingness to pay of users, and mainly reflect the application of user selection model to the dynamic pricing of airlines. Some studies use non-homogeneous Poisson process to simulate user arrival and use discrete choice model to quantify user purchase willingness and probability. Some studies use Logit model to predict user purchase willingness based on user arrival and user market segmentation. In addition, researchers use expected marginal benefit, equilibrium angle and customer perceived value to optimize the revenue of the original buy rate model. However, the key user selection model in the above studies is from the perspective of airlines. Some studies start from the user's perspective, use personalized recommendation algorithm to predict the user's willingness to pay and conduct dynamic pricing research according to user attributes and multi-behavior feedback.

[0006] The prediction of willingness to pay is from the perspective of airlines to control the number of cabins and prices in terms of the depiction of user demand function, the estimation of demand parameters or the construction of user selection model, which cannot decompose the willingness to pay to the level of each user. Although some studies decompose the demand to the level of single user, they still estimate from the perspective of all user behaviors. Some studies from the user's perspective use the selection behavior history of each user to estimate the willingness to pay of each user by using personalized recommendation, but do not consider the evolution of user preferences at different time levels. In practice, user preferences change over time, and each user has different preferences for time.

[0007] How to accurately predict the willingness to pay of users is a technical problem to be solved. SUMMARY

[0008] Therefore, it is necessary to provide a method, device, storage medium and electronic equipment for predicting user willingness to pay in view of the defects that the existing prediction method cannot accurately predict the willingness to pay of users.

[0009] In a first aspect, an embodiment of the present application provides a method for predicting user willingness to pay, the method comprising:

[0010] Based on the continuous numerical features and discrete category features in the flight feature information, the feature representation of the plurality of flight projects corresponding to the flight feature layer is generated;

[0011] Based on the feature representation of the plurality of flight projects, the corresponding session representation is generated;

[0012] Based on the session representation and feature representation evolution network, the preferences of the target user at different time levels are evolved to obtain the preference feature representation of the target user;

[0013] determine a payment willingness probability of the target user for the target flight among the total flights based on the preference feature representation of the target user and the probability distribution;

[0014] predict the payment willingness of the target user for the target flight based on the payment willingness probability of the target flight, to obtain a prediction result.

[0015] Optionally, the determining of the payment willingness probability of the target user for the target flight among the total flights based on the preference feature representation of the target user and the probability distribution comprises:

[0016] mapping, by a multinomial decoder, the preference feature representation of the target user to a payment willingness probability distribution of the target user for the total flights;

[0017] obtaining a probability distribution and taking the probability distribution as a target distribution approximated by the payment willingness probability distribution;

[0018] performing, by a KL divergence method, minimum optimization processing on the payment willingness probability distribution and the probability distribution, to obtain the payment willingness probability of the target flight.

[0019] Optionally, before the obtaining of the probability distribution, the method further comprises:

[0020] adopting a one-hot encoding method to convert the target flight into a flight number vector for the target flight in a preset dimension, and taking the flight number vector as the probability distribution.

[0021] Optionally, the evolving of the preference of the target user at different time levels based on the session representation and the feature representation evolution network comprises:

[0022] obtaining the session representation, the session representation being used to replace the interaction process between all flights in the session and the target user;

[0023] for the target user, expressing, by an update function of the feature representation evolution network, a feature evolution process of the target user and the kth session when the target user and the kth session interact, to obtain a feature representation generated by the kth interaction and a preference representation generated by the k-1th interaction, where u j is the target user, the feature evolution process is used to aggregate long-term features and short-term change features of the target user, and a neural network used to aggregate the long-term features and the short-term change features comprises an RNN, an LSTM, and a GRU.

[0024] Optionally, the generating of the corresponding session representation based on the feature representation of the plurality of flight items comprises:

[0025] obtaining a flight feature sequence;

[0026] sequentially arranging the flight feature sequence according to the time sequence of the time-labeled interaction events;

[0027] obtaining feature representations of the plurality of flight items;

[0028] mapping the feature representations of the plurality of flight items into corresponding session representations through a preset aggregation function, wherein a neural network used by the preset aggregation function is an LSTM.

[0029] Optionally, the generating of the feature representations of the plurality of flight items corresponding to the flight feature layer based on the continuous numerical features and the discrete category features in the flight feature information comprises:

[0030] obtaining flight feature information, wherein the flight feature information comprises information corresponding to the continuous numerical features and information corresponding to the discrete category features, the continuous numerical features are dense features, and the discrete category features are sparse features;

[0031] for the continuous numerical features, projecting the numerical features into first feature representations through a projection matrix, wherein the first feature representations comprise the projection matrix and a bias matrix;

[0032] for the discrete category features, performing numerical processing on the discrete category features to convert category identifiers into numbers;

[0033] performing conversion processing on the numerically processed discrete category features to obtain dense low-dimensional features, and obtaining second feature representations through a lookup table manner with the numbers as indexes;

[0034] obtaining the first feature representations and the second feature representations, and performing splicing processing on the first feature representations and the second feature representations to generate the feature representations of the plurality of flight items corresponding to the flight feature layer.

[0035] Optionally, before the generating of the feature representations of the plurality of flight items corresponding to the flight feature layer based on the continuous numerical features and the discrete category features in the flight feature information, the method further comprises:

[0036] obtaining an event sequence and dividing the event sequence into the session set;

[0037] for the session set, obtaining a number of sessions reserved by a target user, and obtaining a length of events reserved in one session;

[0038] obtaining a time window T and taking the time window T as a division mark of a session;

[0039] In a case that a time interval between two adjacent events in the event sequence is greater than the time window T, the two adjacent events are divided into different sessions to divide the sessions of the target user;

[0040] the kth session S k of the target user is divided into a plurality of time layers, and a time layer corresponding to a time interval between a first interaction event and a second interaction event in the kth session S k of the target user is set as a time layer of the kth session S

[0041] a time tag set of the session is set, and the time tag set includes time information of a time layer, time information of a week layer and time information of a month layer.

[0042] In a second aspect, an embodiment of the present application provides a device for predicting a payment willingness of a user, and the device comprises:

[0043] a first generation module configured to generate feature representations of a plurality of flight items corresponding to a flight feature layer based on continuous numerical features and discrete category features in flight feature information;

[0044] a second generation module configured to generate corresponding session representations based on the feature representations of the plurality of flight items;

[0045] an evolution module configured to evolve preferences of the target user in different time layers based on the session representations and the feature representations to obtain a preference feature representation of the target user;

[0046] a determination module configured to determine a payment willingness probability of the target user to a target flight in all flights based on the preference feature representation of the target user and a probability distribution;

[0047] a prediction module configured to predict a payment willingness of the target user to the target flight based on the payment willingness probability of the target flight to obtain a prediction result.

[0048] In a third aspect, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, and the computer program causes a computer to execute the method in the first aspect when the computer program is executed in the computer.

[0049] In a fourth aspect, an electronic device is provided, and the electronic device comprises a memory and a processor, the memory stores executable code, and the processor executes the executable code to implement the method in the first aspect.

[0050] In the embodiment of the present application, based on the continuous numerical features and discrete category features in the flight feature information, the feature representations of the multiple flight items corresponding to the flight feature layer are generated; the corresponding session representations are generated based on the feature representations of the multiple flight items; the preference of the target user at different time levels is evolved based on the session representations and the feature representation evolution network, to obtain the preference feature representation of the target user; the payment willingness probability of the target user to the target flight in all flights is determined based on the preference feature representation of the target user and the probability distribution; and the payment willingness of the target user to the target flight is predicted based on the payment willingness probability of the target flight. The prediction method of the user payment willingness provided in the embodiment of the present application introduces the feature evolution mechanism of the target user, so that the flight information interacting with the target user is retained, the preference feature of the target user obtained is more comprehensive, the final prediction result is more accurate, and the prediction result is more in line with the actual situation of the target user. BRIEF DESCRIPTION OF DRAWINGS

[0051] The exemplary embodiments of this application can be more fully understood with reference to the following drawings in which:

[0052] Figure 1 The flow chart of the prediction method of the user payment willingness provided according to an exemplary embodiment of the present application;

[0053] Figure 2 The flow chart of the generation of the user preference feature representation based on time evolution;

[0054] Figure 3 The time feature extraction and feature evolution model diagram;

[0055] Figure 4 The payment intention prediction model diagram;

[0056] Figure 5 The structural schematic diagram of the prediction device 500 of the user payment willingness provided according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0057] Exemplary embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings; however, these embodiments are not intended to limit the scope of the present disclosure, but rather, the present disclosure can be implemented in various forms. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0058] It should be noted that the technical terms or scientific terms used in the present application should be the general meanings understood by the skilled in the art of the present application, unless otherwise specified.

[0059] In addition, the terms "first" and "second" and the like are used to distinguish different objects, rather than to describe a specific order. Furthermore, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but can optionally further include steps or units not listed, or can optionally further include other steps or units inherent to such processes, methods, products, or apparatus.

[0060] Embodiments of the present application provide a user payment willingness prediction method and device, a computer readable medium, and an electronic device, which are described below with reference to the accompanying drawings.

[0061] Please refer to Figure 1 which shows a flowchart of a user payment willingness prediction method provided by some embodiments of the present application, as shown in Figure 1 The user payment willingness prediction method can include the following steps:

[0062] Step S101: Based on continuous numerical features and discrete category features in flight feature information, generate feature representations of multiple flight projects corresponding to the flight feature layer.

[0063] In an example, based on continuous numerical features and discrete category features in flight feature information, generating feature representations of multiple flight projects corresponding to the flight feature layer includes the following steps:

[0064] Obtain flight feature information, the flight feature information including information corresponding to continuous numerical features and information corresponding to discrete category features, the continuous numerical features being dense features, and the discrete category features being sparse features;

[0065] For continuous numerical features, project the numerical features into a first feature representation through a projection matrix, the first feature representation including the projection matrix and a bias matrix;

[0066] For discrete category features, perform numerical processing on the discrete category features to convert category identifiers into numbers;

[0067] For the numerically processed discrete category features, perform conversion processing to obtain dense low-dimensional features; using numbers as indexes, obtain a second feature representation through a lookup table;

[0068] The first feature representation and the second feature representation are obtained, and the first feature representation and the second feature representation are spliced to generate a feature representation of a plurality of flight items corresponding to a flight feature layer.

[0069] In an example, the feature information of the flight contains two types of features, |i| is the number of features of the flight, one is a continuous numerical feature It is a dense feature data; the other is a discrete category feature It is a sparse feature data. |i| d and |i| s are the number of dense features and sparse features respectively, and |i| d + |i| s = |i|.

[0070] For continuous features, use a projection matrix to project the numerical features into corresponding feature representations:

[0071] emb den = σ (I den W den + B den )

[0072] wherein, are trainable projection matrix and bias matrix respectively, and σ(.) is the activation function ReLU.

[0073] For discrete data features, first, the discrete features are numerically valued, and the category identifiers are converted into numbers. Then, it is converted into a dense low-dimensional feature Take the number as the index, and get its feature representation by looking up the table.

[0074] The feature representation of the flight item is composed of dense feature representation and sparse feature representation:

[0075]

[0076] wherein, is a splicing operation, h i is the dimension of the flight feature representation, h i = h d + h s , emb item is the feature representation of the flight item, emb dense is the dense feature representation, and emb sparse is the sparse feature representation.

[0077] In one example, before generating feature representations of multiple flight items corresponding to the flight feature layer based on the continuous numerical features and discrete categorical features in the flight feature information, the method for predicting user willingness to pay provided by an embodiment of the present invention may further include the following steps:

[0078] The target user's session is divided to obtain a session division layer.

[0079] In one example, dividing the target user's session includes the following steps:

[0080] Obtain an event sequence and divide the event sequence into a session set;

[0081] For the session collection, obtain the number of sessions retained by the target user; and obtain the length of events retained in a session;

[0082] Get the time window T and use it as the session division mark;

[0083] When the time interval between two adjacent events in the event sequence is greater than the time window T, the two adjacent events are divided into different sessions to divide the target user's session;

[0084] The kth session S of the target user k The time when the first interaction event in the set occurs is the session S k Time mark;

[0085] Set the time stamp set of the session, which includes time information at the hour level, week level, and month level.

[0086] In one example, for an event sequence Q = {e1, e2, ..., e n}, and divide it into a session set S = {S1, S2, ..., S |S|}, |S| is the number of sessions reserved for the target user. The kth session S of the target user u k Indicated as S k ={e k1 ,e k2 ,…,e k|I|}, where |I| is the length of events retained in a session. Set the time window T as the session division mark. When the time interval between two adjacent events in Q is greater than T, the two events are divided into different sessions. k The first interaction event e in the collection k1 The occurrence time t sess As a session S ktime information unit h ={1,2,…,24}, unit w ={1,2,…,7}, unit m ={1,2,…,12}.

[0087] In an example, the method for predicting user payment willingness provided by the embodiment of the application further comprises the following steps:

[0088] In the session graph, if the target user interacts with the flight in the session, it is determined that there is a common edge between the session node and the flight node.

[0089] In an example, in the session graph, if the target user interacts with the flight in the session, it is determined that there is a common edge between the session node and the flight node, comprising the following steps:

[0090] For the target user, a subset of the session set of the target user, a subset of the flight set of the target user, and an interaction event set of the target user are obtained;

[0091] Based on the subset of the session set of the target user, the subset of the flight set of the target user, and the subset of the flight set of the target user and the interaction event set of the target user, a session graph is constructed, and the interaction event set of the target user further comprises time information of each interaction event;

[0092] In the session graph, if the target user interacts with the flight in the session, it is determined that there is a common edge between the session node and the flight node.

[0093] In an example, for the target user u∈U, a subset S u ∈S, a subset F u ∈F of the flight set is associated. An interaction event set E u Records the occurrence of each interaction event. A session graph G u ={S u ,F u ,E u} is constructed.

[0094] In G u , if the target user u interacts with the flight in the session S i , there is an edge (S i , F j )∈E u between the S i node and the F j node, and E u also contains time information t of each interaction event.ij The interaction event set E describes a many-to-many relationship.

[0095] Step S102: generating a corresponding session representation based on the feature representations of the plurality of flight items.

[0096] In an example, generating a corresponding session representation based on the feature representations of the plurality of flight items comprises the following steps:

[0097] Obtaining a flight feature sequence;

[0098] Arranging the flight feature sequence in the time sequence of the time-labeled interaction events;

[0099] Obtaining the feature representations of the plurality of flight items;

[0100] Mapping the feature representations of the plurality of flight items to the corresponding session representation through a preset aggregation function, and the neural network used by the preset aggregation function is LSTM.

[0101] In an example, on the basis of obtaining the feature representations of the flight items, the flight feature sequence is arranged in the time sequence of the time-labeled interaction events, i.e., S k =(emb item,1 ,emb item,2 ,…,emb item,n ), n is the maximum number of flight items in a session, and the feature representations of the plurality of flight items are mapped to a session representation through an aggregation function:

[0102]

[0103] wherein, σ(.) is an activation function, is a trainable nonlinear projection matrix, is a trainable bias matrix. Agg sesS (.) is an aggregation function, the input vector dimension of which is the dimension size item of the flight item feature, and the output is the dimension h sess of the session representation. The aggregation function can have multiple choices such as RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), and multi-head attention network. Considering that the flight items in a session are arranged according to the interaction events t j,i , the long short-term memory network that can fuse context information is selected as the aggregation function.

[0104] Step S103: Evolving the preferences of the target user at different time levels based on the session representation and the feature representation evolution network to obtain the preference feature representation of the target user.

[0105] In an example, evolving the preferences of the target user at different time levels based on the session representation and the feature representation evolution network comprises the following steps:

[0106] Obtaining the session representation, which is used to replace the interaction process of all flights within the session and the target user;

[0107] For the target user, using an update function of the feature representation evolution network to represent the feature evolution process of the target user and the kth session when the interaction occurs, to obtain the feature representation generated by the kth interaction and the preference representation generated by the k-1th interaction, where u j For the target user, using the feature evolution process to aggregate the long-term features and the short-term change features of the target user, and the neural network used to aggregate the long-term features and the short-term change features comprises: RNN, LSTM and GRU.

[0108] In an example, for the target user u j , an update function is used to represent the feature evolution process of the target user and the kth session when the interaction occurs:

[0109]

[0110] represents the feature representation of the target user u j after the kth interaction, represents the preference representation generated by the k-1th interaction. f(·) is an interaction change function used to calculate the new influence of the preference change of the target user. Agg(·) represents an aggregation function used to aggregate the historical preferences and the updated part of the preferences of the target user. The process is essentially to aggregate the long-term features and the short-term change features of the target user, and neural networks such as RNN, LSTM, and GRU (Gated Recurrent Unit) can be used to realize the aggregation process.

[0111] Step S104: Determining the willingness-to-pay probability of the target user for the target flight among all flights based on the preference feature representation and the probability distribution of the target user.

[0112] In an example, determining the willingness-to-pay probability of the target user for the target flight among all flights based on the preference feature representation and the probability distribution of the target user comprises the following steps:

[0113] Through the polynomial decoder, the target user's preference feature representation is mapped into the target user's willingness to pay probability distribution for all flights;

[0114] Obtain a probability distribution and use the probability distribution as the target distribution to approximate the probability distribution of willingness to pay;

[0115] The KL divergence method is used to minimize and optimize the probability distribution of willingness to pay and the probability distribution to obtain the probability of willingness to pay for the target flight.

[0116] In one example, a polynomial decoder is used to map the target user's preference feature representation to the target user's willingness to pay probability distribution for all flights through a nonlinear mapping:

[0117] π(emb u )=Softmax(σ(W d emb u +B d ))

[0118] Among them, W d is the trainable projection matrix, B d is the trainable bias matrix.

[0119] The target user’s final interactive flight is converted into a vector with a dimension equal to the number of flights using a unique hot encoding method, and used as a probability distribution Q to represent the target user’s willingness to pay for all flights. This is π(emb u ) The target distribution to be approximated is minimized by KL divergence. u ) and Q, and use them as optimization objectives:

[0120]

[0121] Among them, M is the number of target user tables, g is the number of flights, For target user u j The probability of willingness to pay for flight k.

[0122] In one example, before obtaining the probability distribution, the method for predicting user willingness to pay provided by an embodiment of the present invention may further include the following steps:

[0123] Using one-hot encoding, the target flight is converted into a flight number vector for the target flight in the preset dimension; and the flight number vector is used as a probability distribution.

[0124] Step S105: Based on the willingness-to-pay probability of the target flight, the target user's willingness-to-pay for the target flight is predicted to obtain a prediction result.

[0125] In an actual application scenario, if the payment willingness probability of the target flight is higher, the prediction result obtained is that the target user has a higher willingness to pay for the target flight, and the target flight can be recommended to the target user; otherwise, if the payment willingness probability of the target flight is lower, the prediction result obtained is that the target user has a lower willingness to pay for the target flight.

[0126] In an actual application scenario, the user's flight taking has very obvious time periodicity, and such time periodicity is embodied in multiple time layers. For example, from the perspective of flight passenger flow, the passenger flow in the early morning and at midnight is less than that in the morning and afternoon in a day; the passenger flow on weekdays is less than that near the weekend in a week; and the passenger flow during the golden week of festivals and during the summer and winter vacations is obviously more than that on other days in a year. Such changes in passenger flow directly lead to changes in flight ticket prices, and specifically, the ticket price of the same route during the peak period will be higher. Such time periodicity reflects that the user has different flight taking preferences at different times, and has respective characteristics in different time dimensions. For example, the user tends to choose a business class with high price for taking a flight due to business needs on weekdays, and chooses an economy class with lower price on non-working days; a part of users tend to choose cheap tickets at midnight, and another part of users are less sensitive to price and tend to choose to take a flight during the day.

[0127] Capturing the time periodicity of the flight taking preferences of the target user is crucial for predicting the willingness. Therefore, the characteristics of the target user at different time layers (day, week, and month) are discovered and evolved to obtain the dynamic preferences of the target user, and a personalized recommendation algorithm is designed to improve the accuracy of the prediction of the payment willingness.

[0128] In an example, the method for predicting the payment willingness of a user can include the following steps:

[0129] Step a1: generating a target user preference feature representation based on time evolution.

[0130] The process of generating a target user preference feature representation based on time evolution is shown in Figure 2 , features are extracted from the flight and user dimensions, aggregation and preference evolution are performed in different time dimensions, and the payment willingness of the user is predicted to obtain a prediction result.

[0131] As shown in Figure 2 , the process of generating a target user preference feature representation based on time evolution is specifically as follows:

[0132] Step a11: dividing the session of the target user behavior.

[0133] The preferences of the target user will change over time and with the occurrence of interaction events. Therefore, the interaction history of the target user is divided according to a certain time interval. Intra-session presents short-term preference consistency; inter-session presents dynamic difference of target user preference. Multiple session records of the target user not only have differences, but also have connections.

[0134] For a target user u e U, it is associated with a subset S of the session set u ∈S, a subset F of the flight set u ∈F. The interaction event set E u records the occurrence of each interaction event. The session graph G u = {S u ,F u ,E u} is constructed.

[0135] In G u , if the target user u interacts with the flight in the session S i , there is an edge (S i , F j ) e E i between the S j node and the F u node. E u also contains the time information t ij of each interaction event. The interaction event set E describes a many-to-many relationship.

[0136] For the event sequence Q = {w1, e2, …, e n}, it is divided into the session set S = {S1, S2, …, S |S|}, |S| is the number of sessions reserved for the target user. The kth session S k of the target user u is represented as S k = {e k1 , e k2 , …, e k|I|}, where |I| is the length of the event reserved in a session. Set the time window T as the division mark of the session, when the time interval of two adjacent events in Q is greater than T, the two events are divided into different sessions. The occurrence time t k of the first interaction event e k1 in the S sess set is set as the time mark of the session S k . Set the time mark set UnitSet of the session, which contains three levels of time information unit h = {1, 2, …, 24}, unit w = {1, 2, …, 7}, unitm = {1, 2,..., 12}.

[0137] Step a12: The feature representation of the flight item is composed of the dense feature representation and the sparse feature representation.

[0138] Feature information of the flight Two types of features are included, |i| is the number of features of the flight, one is the continuous numerical feature It is a dense feature data; the other is the discrete category feature It is a sparse feature data. |i| d and |i| s are the number of features of the dense feature and the sparse feature respectively, and |i| d + |i| s = |i|.

[0139] For continuous features, use the projection matrix to project the numerical features into the corresponding feature representation:

[0140] emb den = σ (I den W den + B den )

[0141] wherein, are the trainable projection matrix and bias matrix respectively, and σ(.) is the activation function ReLU.

[0142] For discrete data features, first, the discrete features are numerized, and the category identifier is converted into a number. Then it is converted into a dense low-dimensional feature Take the number as the index, and get its feature representation by looking up the table.

[0143] The feature representation of the flight item is composed of the dense feature representation and the sparse feature representation:

[0144]

[0145] wherein, is the concatenation operation, h i is the dimension of the flight feature representation, h i = h d + h s , emb item is the feature representation of the flight item, emb dense is the dense feature representation, and emb sparse is the sparse feature representation.

[0146] Step a13: Generate the preference feature representation of the target user at different time levels.

[0147] On the basis of obtaining the feature representation of the flight item, the flight feature sequence is arranged in the time sequence of the time marked interaction event, that is, S k =(emb item,1 ,emb item,2 ,…,emb item,n ), n is the maximum value of the number of flight items in a session, and a plurality of feature representations of the flight items are mapped into a session representation by an aggregation function:

[0148]

[0149] Wherein, the σ(.) activation function, is a trainable nonlinear projection matrix, is a trainable bias matrix. Agg sess (.) aggregation function, the input vector dimension of which is the dimension size item of the flight item feature, and the output is the session representation dimension h sess , the aggregation function can have multiple choices such as RNN (Recurrent Neural Network, recurrent neural network), LSTM (Long Short-Term Memory, long short-term memory), multi-head attention network, etc. Considering that the flight items in the session are arranged according to the interaction event t j,i , the long short-term memory network which can fuse the context information is selected as the aggregation function.

[0150] Step a14: evolving the preference feature representation of the target user at different time levels to generate the target user preference feature representation based on time evolution.

[0151] The preference of the target user is not always constant, and as the interaction between the target user and the item occurs, the target user often generates new preference changes as the interaction event occurs. In order to capture this change, a feature representation evolution network is used, as shown in Figure 3 As shown in Figure 3 , the feature representation of each link can be aggregated and evolved using GRU (Gated Recurrent Unit, Gated Recurrent Unit), RNN, etc. The session representation is used instead of all the flights in the session to interact with the target user. In this way, on the one hand, compared with the interaction evolution between the target user and all the flights, the calculation amount of the model can be reduced and the calculation speed can be accelerated; on the other hand, the browsing event and the purchase event of the target user are integrated into one, solving the data cold start problem caused by the sparsity of the purchase event.

[0152] For the target user u j, an update function is used to express the target user and the kth session The process of feature evolution when interaction occurs:

[0153]

[0154] represents the target user u j The feature representation after the k-th interaction occurs, represents the preference expression generated in the k-1th interaction. f(·) is the interaction change function, which is used to calculate The new impact of changes in the target user's preferences. Agg(·) represents an aggregation function used to aggregate the target user's historical preferences and updated preferences. This process essentially aggregates the target user's long-term characteristics and short-term change characteristics. This aggregation process can be implemented using neural networks such as RNN, LSTM, and GRU.

[0155] Step a2: After obtaining the time-evolved representation of the target user's preference characteristics, predict the user's willingness to pay. The process of predicting the user's willingness to pay is as follows:

[0156] The target user's preference characteristics are mapped into the target user's willingness to pay probability distribution for all flights through a polynomial decoder through a nonlinear mapping:

[0157] π(emb u )=Softmax(σ(W d emb u +B d ))

[0158] Among them, W d is the trainable projection matrix, B d is the trainable bias matrix.

[0159] The target user’s final interactive flight is converted into a vector with a dimension equal to the number of flights using a unique hot encoding method, and used as a probability distribution Q to represent the target user’s willingness to pay for all flights. This is π(emb u ) The target distribution to be approximated is minimized by KL divergence. u ) and Q, and use them as optimization objectives:

[0160]

[0161] Among them, M is the number of target user tables, g is the number of flights, For target user u j For the willingness to pay probability of flight k, the specific process is as follows Figure 4 As shown.Figure 4 The preference feature of the target user and the clustered target user are combined, and the payment willingness of the target user to different discount flights is predicted.

[0162] The user payment willingness prediction method provided by the embodiment of the application captures the user preferences and features of the user in different time period layers by using the search and interaction history of the user, and evolves the user preferences and user features by using two neural networks to obtain the evolved user preferences of the user and predict the payment willingness. Compared with the traditional method, the prediction accuracy can be improved, and the different flight preferences of the user in different time periods can be effectively met.

[0163] In another example, the user payment willingness prediction method can include the following steps:

[0164] Step b1: obtaining usage data and obtaining data format.

[0165] Step b11: obtaining usage data. The behavior data of about 28,000 users of a certain domestic airline on 202 routes from September 1, 2020 to November 30, 2020 is obtained, a total of 1,048,000 behavior events. The difference is that, in order to capture the feature preferences of the user in different time layers, the time of the user interaction event is added, and the training set and the test set are divided according to the time as the boundary. The user interaction data from September 1, 2020 to November 15, 2020 is used as the training set data; the user interaction data from November 16, 2020 to November 30, 2020 is used as the test set data.

[0166] Step b12: obtaining data format.

[0167] Training data table (SX_train_dy.txt): each row represents a user-item interaction record, and the column names are: uid (user ID), iid (item ID), t (time), and label (event type).

[0168] Test data table (SX_test_dy.txt): column names: uid (user ID), iid (item ID), t (time), and label (event type); each row represents a user-item interaction record.

[0169] User attribute table (attr_dy.txt): column names: uid (user ID), attr1, attr2, attr3, attr4, attr5, and attr6 (user attributes); each row represents the attribute information of the user.

[0170] Step b2: loading training data.

[0171] Input parameters: file: data file path, i.e.,. / data / SX_train_dy.txt.

[0172] Data loading: Use the pd.read_csv function to load the data file, with tab character ('\t') as the delimiter, and specify column names as ['uid', 'iid', 't', 'label']. Here, the column names represent user ID, item ID, timestamp, and label. Time window organization: Use groupby('t') to group data by timestamp (t). This step divides the data into multiple groups according to the timestamp, with each group representing a time window.

[0173] Organize data: By iterating through the time window groups (antDict), extract user ID, item ID, and label within each time window as user, item, and lab lists, respectively. Store these lists in antlist, forming a three-layer nested list.

[0174] Return value: antlist: a list containing three sublists, storing user ID, item ID, and label. Each sublist represents the users, items, and labels within a time window.

[0175] Return result example:

[0176] For a certain time window, the structure of antlist may be as follows: [

[0178] [u1, u2,...], # User ID list

[0179] [i1, i2,...], # Item ID list

[0180] [l1, l2,...] # Label list ]

[0182] Where u1, i1, l1 represent a user interaction event, u2, i2, l2 represent another user interaction event, and so on.

[0183] Step b3: Preprocess the data.

[0184] Read two data files named user_record_discount.csv and user_info_discount.csv, which contain user records and user attribute information.

[0185] Convert the time field in the record to a timestamp, and divide the data into training and test sets according to the timestamp. The timestamp threshold is '2020 / 11 / 16 00:00:00'.

[0186] Traverse the user records in the training set, extract the user's attribute information, including sparse features and dense features. At the same time, group the user's events in chronological order to form sessions, and generate user session lists and session time lists.

[0187] In the formed session, find the last shopping event according to the event flag (1 represents shopping event) as the candidate item of each user in the training set, and record the time of the candidate item.

[0188] Convert the time information in the training set to a list of month, day of the week and hour.

[0189] Record the session length of each user in the training set.

[0190] Similarly, traverse the user records in the test set, extract the user's attribute information, generate user session lists and session time lists. Find the candidate item of each user in the test set, and record the time of the candidate item.

[0191] Convert the time information in the test set to a list of month, day of the week and hour.

[0192] Record the session length of each user in the test set.

[0193] Return all generated information.

[0194] Step b4: Obtain the preprocessed data.

[0195] In one example, the user-related data specifically includes:

[0196] 1. user_dict_train, user_dict_test: dictionary, containing information such as attributes, target items, user sessions, session times and session lengths of training and test users.

[0197] {user_id: [user_attr, target_item, user_sessions, session_times, session_lengths]}

[0198] 2. user_setime_train, user_setime_test: user session time information.

[0199] {user_id: [session1_time, session2_time,...]}

[0200] 3. user_sesslength_train, user_sesslength_test: user's session length information.

[0201] {user_id: [session1_length, session2_length,...]}

[0202] In an example, the candidate item related data specifically includes:

[0203] train_can, test_can: dictionaries containing the candidate items of the training and test users.

[0204] {user_id: candidate_item}.

[0205] Step b5: Generate user embeddings at different time levels.

[0206] Step b51: Generate user and item representations.

[0207] Index the user through user_embeds to obtain the embedding representation of the user, stored in user_embed. Index the candidate item through item_embed to obtain the embedding representation of the candidate item, stored in candidate_embed.

[0208] Step b52: Generate session representation.

[0209] Obtain the embedding representation of the item in input_sess through item_embed, stored in sess_embed. Convert sess_embed to session embedding using item2sess, stored in sess_embed.

[0210] Step b53: Extract time features.

[0211] Extract time features for three time levels (hour, week, month) respectively through get_timeembed function, stored in time_feas.

[0212] Step b54: Update user representation.

[0213] Update the representation of the corresponding user in user_embed_update to time_feas.

[0214] Step b55: Activation and merging.

[0215] The time feature time_feas and the session embedding sess_embed are concatenated.

[0216] Step b56: Calculate the score.

[0217] The score is calculated through the fully connected layers (score_fc1 and score_fc2).

[0218] Step b57: Calculate the Loss.

[0219] In the training phase, the loss is calculated using the LogSoftmax and the Negative Log Likelihood (NLL).

[0220] Negative Sampling: In the training phase, negative sampling is performed to calculate the score of negative samples.

[0221] Output: Return the calculated loss and the predicted probability or score for the given user and candidate item.

[0222] Step b6: Evolve the preference feature representation of the user at different time levels.

[0223] Input session representation emb sess , user feature representation emb user , time level time_level, set of time level units UnitSet, session time t sess ;

[0224] The session sequence is divided into different time units, and the session sequence representation is aggregated to evolve the preference feature representation of the user at different time levels using the GRUCELL.

[0225] Step b7: Load test data.

[0226] Input parameters: file: data file path, i.e.. / data / SX_test_dy.txt.

[0227] Data loading: Use the pd.read_csv function to load the data file with tab character ('\t') as the delimiter and specify the column names as ['uid', 'iid', 't', 'label']. Here, the column names represent user ID, item ID, timestamp, and label.

[0228] Organization of time window:

[0229] Use groupby('t') to group the data by timestamp (t). This step divides the data into multiple groups according to the timestamp, and each group represents a time window.

[0230] Organize data:

[0231] By iterating through the time window group (testDict), extract user IDs, item IDs, and labels within each time window into user, item, and lab lists respectively. Store these lists into testlist, forming a three-level nested list.

[0232] Return value:

[0233] testlist: A list containing three sublists, storing user IDs, item IDs, and labels. Each sublist represents users, items, and labels within a time window.

[0234] Example of returned result:

[0235] For a certain time window, the structure of testlist might look like this: [

[0237] [u1, u2,...], # User ID list

[0238] [i1, i2,...], # Item ID list

[0239] [l1, l2,...] # Label list

[0240] ] Where u1, i1, l1 represent one user interaction event, u2, i2, l2 represent another user interaction event, and so on.

[0241] In summary, the main function of this function is to load the test data file, organize data by time window, and return a three-level nested list, where each sublist contains user, item, and label information within a time window.

[0242] Step b8: Predict user payment willingness.

[0243] Input the target user representation into the decoder, treat the neural network as a multinomial likelihood function, and adjust the parameters of the neural network by maximizing the likelihood estimate. Through a multinomial decoder, the preference feature representation of the target user is mapped to the probability distribution of the target user's preference for all flights through a nonlinear mapping, as follows:

[0244] π(emb u )=Softmax(σ(W d emb u +B d ))

[0245] Where W d is a trainable projection matrix, Bd is a trainable bias matrix.

[0246] Further calculate the target user's preference probability distribution probability distribution Q for all flights, minimize π(emb u ) and Q by KL divergence, as the optimization objective, as shown in the formula:

[0247]

[0248] Wherein, M is the number of target users, g is the number of flights, is the target user u j Preference probability for flight k.

[0249] The user payment willingness prediction method provided by the embodiment of the present application introduces the feature evolution mechanism of the target user, so that the flight information interacting with the target user is retained, the preference features of the target user obtained are more comprehensive, the final prediction result is more accurate, and the prediction result is more in line with the actual situation of the target user. In addition, the prediction method provided by the embodiment of the present application also considers the feature difference of the user at multiple time levels, and can construct the time sequence features of the target user from multiple angles.

[0250] In the above embodiment, a user payment willingness prediction method is provided, and the present application also provides a user payment willingness prediction device corresponding thereto. The user payment willingness prediction device provided by the embodiment of the present application can implement the user payment willingness prediction method described above, and the user payment willingness prediction device can be realized by software, hardware or a combination of software and hardware. For example, the user payment willingness prediction device can include integrated or separate functional modules or units to perform the corresponding steps in the above methods.

[0251] Please refer to Figure 5 , which shows a schematic diagram of a user payment willingness prediction device provided by some embodiments of the present application. Since the device embodiment is basically similar to the method embodiment, it is described more simply, and the related parts refer to the part of the method embodiment. The device embodiment described below is only illustrative.

[0252] As shown in Figure 5 , the user payment willingness prediction device 500 can include:

[0253] The first generation module 501 is configured to generate feature representations of a plurality of flight items corresponding to the flight feature layer based on continuous numerical features and discrete category features in the flight feature information;

[0254] The second generation module 502 is configured to generate corresponding session representations based on the feature representations of the plurality of flight items;

[0255] an evolution module 503, configured to evolve preferences of different time levels of the target user based on the session representation and the feature representation evolution network, to obtain a preference feature representation of the target user;

[0256] a determination module 504, configured to determine a payment willingness probability of the target user to a target flight in all flights based on the preference feature representation of the target user and a probability distribution;

[0257] a prediction module 505, configured to predict a payment willingness of the target user to the target flight based on the payment willingness probability of the target flight, to obtain a prediction result.

[0258] In some embodiments of the present application, the determination module 504 is specifically configured to:

[0259] map the preference feature representation of the target user to a payment willingness probability distribution of the target user to all flights through a polynomial decoder;

[0260] obtain the probability distribution, and take the probability distribution as a target distribution for approximation of the payment willingness probability distribution;

[0261] perform minimization optimization processing on the payment willingness probability distribution and the probability distribution in a KL divergence manner, to obtain the payment willingness probability of the target flight.

[0262] In some embodiments of the present application, the prediction device 500 of the payment willingness of the user provided by the present application can further include:

[0263] a probability distribution determination module (not shown in Figure 5 ), configured to, before obtaining the probability distribution, convert the target flight into a flight number vector for the target flight in a preset dimension in a one-hot encoding manner, and take the flight number vector as the probability distribution.

[0264] In some embodiments of the present application, the evolution module 503 is specifically configured to:

[0265] obtain the session representation, the session representation being used to replace an interaction process of all flights in a session and the target user;

[0266] for the target user, express a feature evolution process of the target user and the kth session when the target user and the kth session interact through an update function of the feature representation evolution network, to obtain a feature representation generated by the kth interaction and a preference representation generated by the k-1th interaction, wherein u jFor the target user, the feature evolution process is used to aggregate the long-term features and short-term change features of the target user. The neural networks used to aggregate the long-term features and short-term change features include: RNN, LSTM and GRU.

[0267] In some implementations of the embodiments of the present invention, the second generating module 502 is specifically configured to:

[0268] Get flight feature sequence;

[0269] Arrange the flight feature sequences in order according to the chronological order of the time-stamped interaction events;

[0270] Obtain feature representations of multiple flight items;

[0271] Through a preset aggregation function, the feature representations of multiple flight items are mapped to corresponding session representations. The neural network used by the preset aggregation function is LSTM.

[0272] In some implementations of the embodiments of the present invention, the first generating module 501 is specifically configured to:

[0273] Obtain flight feature information, which includes information corresponding to continuous numerical features and information corresponding to discrete categorical features. Continuous numerical features are dense features, while discrete categorical features are sparse features.

[0274] For continuous numerical features, the numerical feature projection is converted into the first feature representation through the projection matrix. The first feature representation includes the projection matrix and the bias matrix.

[0275] For discrete category features, perform numerical processing on the discrete category features to convert the category identifier into a number;

[0276] The discrete category features that have been numerically processed are converted to obtain dense low-dimensional features; using numbers as indexes, the second feature representation is obtained by table lookup;

[0277] The first feature representation and the second feature representation are obtained, and the first feature representation and the second feature representation are concatenated to generate feature representations of multiple flight items corresponding to the flight feature layer.

[0278] In some implementations of the embodiments of the present invention, the user willingness to pay prediction device 500 provided by the embodiments of the present invention may further include:

[0279] Session partitioning module (in Figure 5 ), specifically used for:

[0280] Before generating the feature representation of the plurality of flight items corresponding to the flight feature layer based on the continuous numerical features and the discrete category features in the flight feature information, an event sequence is acquired, and the event sequence is divided into a session set;

[0281] For the session set, the number of sessions reserved by the target user is acquired, and the length of events reserved in one session is acquired;

[0282] A time window T is acquired, and the time window T is taken as a division mark of the session;

[0283] In a case where the time interval between two adjacent events in the event sequence is greater than the time window T, the two adjacent events are divided into different sessions, so as to divide the sessions of the target user;

[0284] The kth session S k of the target user is taken as the time mark of the session S k ;

[0285] A time mark set of the session is set, and the time mark set includes time information of a time layer, time information of a week layer and time information of a month layer.

[0286] In some embodiments of the embodiments of the present application, the user payment willingness prediction device 500 provided by the embodiments of the present application has the same inventive concept and beneficial effects as the user payment willingness prediction method provided by the foregoing embodiments of the present application.

[0287] According to another aspect of the embodiments, a computer readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method described in connection with Figure 1 is also provided.

[0288] According to still another aspect of the embodiments, an electronic device is also provided, comprising a memory and a processor, the memory having stored therein executable code which, when executed by the processor, implements the method described in connection with Figure 1 .

[0289] Those skilled in the art should be aware that the functions described in the above one or more examples can be implemented in hardware, software, firmware or any combination thereof. When implemented in software, the functions can be stored in a computer readable medium or transmitted as one or more instructions or codes on a computer readable medium.

[0290] The above detailed description of the specific embodiments of the present application has been given to illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solutions of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting user willingness to pay, characterized in that: The method comprises: Based on the continuous numerical features and discrete categorical features in the flight feature information, generate feature representations of multiple flight items corresponding to the flight feature layer; generating corresponding conversation representations based on the feature representations of the plurality of flight items; Based on the session representation and feature representation evolution network, evolving the preferences of the target user at different time levels to obtain the preference feature representation of the target user; The evolving network based on the session representation and feature representation, evolving preferences of target users at different time levels, includes: Obtaining the session representation, where the session representation is used to represent the interaction process between all flights in the session and the target user; For the target user, the update function of the feature representation evolution network is used to express the target user and the kth session The feature evolution process when the interaction occurs is to obtain the feature representation generated by the kth interaction and the preference representation generated by the k-1th interaction, wherein the kth session in For a target user, the long-term features and short-term change features of the target user are aggregated through the feature evolution process, and the neural networks used to aggregate the long-term features and the short-term change features include: RNN, LSTM and GRU; Determining the target user's willingness to pay probability for a target flight among all flights based on the target user's preference characteristic representation and probability distribution; The determining, based on the target user's preference characteristic representation and probability distribution, the target user's willingness to pay probability for the target flight among all the flights includes: Mapping the target user's preference feature representation into a probability distribution of the target user's willingness to pay for all flights through a polynomial decoder; Obtaining a probability distribution, and using the probability distribution as a target distribution to approximate the willingness-to-pay probability distribution; By using the KL divergence method, the willingness to pay probability distribution and the probability distribution are minimized and optimized to obtain the willingness to pay probability of the target flight; Based on the willingness-to-pay probability of the target flight, the willingness-to-pay of the target user for the target flight is predicted to obtain a prediction result.

2. The prediction method according to claim 1, characterized in that Before obtaining the probability distribution, the method further includes: The target flight is converted into a flight number vector for the target flight in a preset dimension by using a one-hot encoding method; and the flight number vector is used as the probability distribution.

3. The prediction method according to claim 1, wherein: Generating corresponding conversation representations based on the feature representations of the plurality of flight items includes: Get flight feature sequence; Arranging the flight feature sequences in sequence according to the chronological order of the time-stamped interaction events; Obtaining feature representations of the plurality of flight items; The feature representations of the plurality of flight items are mapped to corresponding session representations through a preset aggregation function, and the neural network adopted by the preset aggregation function is LSTM.

4. The prediction method according to claim 1, wherein: The step of generating feature representations of multiple flight items corresponding to the flight feature layer based on the continuous numerical features and discrete categorical features in the flight feature information includes: Acquire flight feature information, where the flight feature information includes information corresponding to the continuous numerical feature and information corresponding to the discrete categorical feature, where the continuous numerical feature is a dense feature and the discrete categorical feature is a sparse feature; For continuous numerical features, the numerical features are projected and converted into a first feature representation through a projection matrix, where the first feature representation includes a projection matrix and a bias matrix; For discrete category features, performing numerical processing on the discrete category features to convert category identifiers into numbers; The discrete category features processed by numerical processing are converted to obtain dense low-dimensional features; the second feature representation is obtained by using the number as an index through table lookup; The first feature representation and the second feature representation are obtained, and the first feature representation and the second feature representation are concatenated to generate feature representations of a plurality of flight items corresponding to the flight feature layer.

5. The prediction method according to claim 1, wherein: Before generating feature representations of a plurality of flight items corresponding to the flight feature layer based on the continuous numerical features and discrete categorical features in the flight feature information, the method further includes: Acquire an event sequence, and divide the event sequence into the session set; For the session set, obtaining the number of sessions retained by the target user; and obtaining the length of events retained in one session; Get the time window T and use it as the session division mark; When the time interval between two adjacent events in the event sequence is greater than the time window T, the two adjacent events are divided into different sessions to divide the session of the target user; The kth session of the target user The time when the first interaction event in the set occurs is the session Time mark; A time stamp set of the session is set, wherein the time stamp set includes time information at the hour level, time information at the week level, and time information at the month level.

6. A device for predicting user willingness to pay, characterized in that: The device comprises: A first generation module is configured to generate feature representations of multiple flight items corresponding to the flight feature layer based on the continuous numerical features and discrete categorical features in the flight feature information; A second generating module, configured to generate corresponding conversation representations based on the feature representations of the plurality of flight items; An evolution module, configured to evolve the preferences of the target user at different time levels based on the session representation and feature representation evolution network to obtain a feature representation of the target user's preferences; The evolution module is specifically used for: Obtaining the session representation, where the session representation is used to represent the interaction process between all flights in the session and the target user; For the target user, the update function of the feature representation evolution network is used to express the target user and the kth session The feature evolution process when the interaction occurs is to obtain the feature representation generated by the kth interaction and the preference representation generated by the k-1th interaction, wherein the kth session in For a target user, the long-term features and short-term change features of the target user are aggregated through the feature evolution process, and the neural networks used to aggregate the long-term features and the short-term change features include: RNN, LSTM and GRU; a determination module, configured to determine the target user's willingness to pay probability for a target flight among all flights based on the target user's preference characteristic representation and probability distribution; The determining module is specifically configured to: Mapping the target user's preference feature representation into a probability distribution of the target user's willingness to pay for all flights through a polynomial decoder; Obtaining a probability distribution, and using the probability distribution as a target distribution to approximate the willingness-to-pay probability distribution; By using the KL divergence method, the willingness to pay probability distribution and the probability distribution are minimized and optimized to obtain the willingness to pay probability of the target flight; The prediction module is used to predict the payment willingness of the target user for the target flight based on the payment willingness probability of the target flight to obtain a prediction result.

7. A computer-readable storage medium, characterized in that A computer program is stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method according to any one of claims 1 to 5 is implemented.

Images