Method for Predicting Consumer Behavior and Optimizing Decision-making in E-commerce Platform Based on Iterative Attention

Abstract

The present invention relates to the technical field of e-commerce management, and particularly to a method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on iterative attention; through steps such as data collection, preprocessing, and consumer classification, the method deeply analyzes the consumer behavior data on the e-commerce platform, vectorizes, classifies, and performs weighted summation on the consumer behavior data based on the attention mechanism, establishes a consumer behavior prediction model, and performs iterative calculations on the click log sequence through a recurrent neural network; this method can dynamically optimize the prediction model, improve the prediction accuracy of consumers' future behaviors, establish an internal connection between different click and purchase behaviors, assign different weights to each consumer behavior through the attention mechanism, accurately capture the purchase intention of users, and ultimately can achieve personalized recommendation, accurate prediction, and real-time decision-making optimization, thereby improving the operation efficiency, user experience, and commercial value of the e-commerce platform.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce management, and particularly to a method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on iterative attention. Background Art

[0002] With the rapid development of e-commerce, e-commerce platforms have accumulated a vast amount of user behavior data, including behaviors such as user browsing, clicking, searching, and shopping. These data provide valuable insights for the platform, enabling it to understand user needs, optimize product recommendations, and enhance the user experience. However, how to accurately predict consumer behavior from this vast amount of data and achieve personalized recommendations and decision-making optimization remains a major challenge for e-commerce platforms.

[0003] Most existing consumer behavior prediction methods rely on traditional statistical methods and rule-based models, such as Collaborative Filtering, Content-based Recommendation, and Matrix Factorization. These methods predict users' future preferences and needs by analyzing users' historical behaviors and the behaviors of other similar users.

[0004] However, these traditional methods have the following limitations:

[0005] 1. Data sparsity: Collaborative filtering methods perform poorly when there is less interaction data between users and items. For new users or cold-start items, these methods are difficult to give effective recommendations.

[0006] 2. Long-tail problem: Most traditional recommendation algorithms perform poorly in dealing with "long-tail items", that is, for those items with low view counts but potential markets, the recommendation accuracy of the recommendation system is relatively low.

[0007] 3. Lack of dynamics: Traditional methods usually cannot capture changes in user behavior in real time and cannot respond quickly according to short-term changes in users.

[0008] In recent years, with the rapid development of deep learning technology, consumer behavior prediction methods based on deep learning have been widely applied. Especially when dealing with large-scale and high-dimensional data, deep learning can effectively mine complex patterns in the data and overcome some limitations of traditional methods.

[0009] Common deep learning-based consumer behavior prediction methods include Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), etc. Although deep learning methods can significantly improve prediction accuracy, they still face challenges in processing multi-level and complex user behavior data. For example, how to effectively combine multi-dimensional data such as users' click behavior, purchase behavior, and product information to further enhance the prediction ability of the model has become a hot topic in current deep learning research.

[0010] To overcome the limitations of traditional deep learning methods in processing complex data, in recent years, the Attention Mechanism has been widely introduced into the field of consumer behavior prediction. The Attention Mechanism originated from the field of Natural Language Processing (NLP). Its core idea is to assign different weights to different parts of the input data, enabling the model to focus on the parts most relevant to the task.

[0011] In the prediction of consumer behavior on e-commerce platforms, the Attention Mechanism can help the model automatically identify which user behaviors have important impacts on predicting consumers' purchase behavior. For example, the click behavior of certain products may be more predictive than that of other products, and the purchase history of certain users may be more crucial for predicting future behavior. Through attention weighting, the model can more accurately find these "key" behaviors in the vast amount of data, thereby improving the accuracy of prediction.

[0012] Although the consumer behavior prediction methods based on deep learning and the Attention Mechanism are theoretically highly accurate, there are still many challenges in practical applications:

[0013] 1. Data quality issue: The data on e-commerce platforms often contains noise and missing values. How to achieve efficient prediction under the condition of low data quality is an urgent problem to be solved.

[0014] 2. Real-time issue: Consumer behavior is highly dynamic. How to capture users' behavior changes in real time and make rapid decisions is still a difficult problem for most e-commerce platforms.

[0015] 3. Computational complexity issue: Deep learning models, especially those with the Attention Mechanism and iterative optimization, have relatively high computational complexity. How to reduce the computational cost while ensuring the model performance is an aspect that needs to be optimized currently. Summary of the Invention

[0016] To address the above problems, the object of the present invention is to propose: An e-commerce platform consumer behavior prediction and decision-making optimization method based on attention iteration, including the following steps:

[0017] S1. Data collection: Collect e-commerce information from the e-commerce platform as required, including consumer behavior data, transaction data, and product data;

[0018] S2. Data preprocessing: Preprocess the e-commerce information according to the usage requirements to obtain a preprocessed data set;

[0019] S3. Consumer classification: Construct a vectorized data set, a click log sequence, and a purchase behavior data set based on the preprocessed data set, classify the vectorized data set and the initial weight and value vector set of the vectorization, and perform weighted summation and recursive iterative calculation on the classified vectorized data set based on the attention mechanism;

[0020] S4. Establish a consumer prediction model: Input the vectorized data set that has undergone weighted summation and recursive iterative calculation into a classifier to determine the consumers to be predicted;

[0021] S5. Behavior prediction: Predict the subsequent behaviors of the consumers to be predicted based on the consumer prediction model.

[0022] Further, step S3 is specifically as follows:

[0023] S31. Construct a vectorized data set:

[0024] Determine sequence data according to the preprocessed data set, and the sequence data satisfies that one purchase behavior corresponds to clicks, where L is a positive integer and satisfies

[0025] The vectorized data set is represented by , where represents the th click behavior of the th user on the e-commerce platform, and represents the purchase behavior completed by the th user on the e-commerce platform after their th click, where , , U and I respectively represent the total number of users and the total number of clicks;

[0026] S32. Determine the click log sequence: According to the vectorized data set, determine the vectorized data set that meets the condition T1 as the click log sequence, and the condition T1 is: at least consecutive clicks constitute a purchase;

[0027] S33. Determine the purchase behavior data set: Based on the vectorized data set , take at least one vectorized data set that meets All vectors form a set, and a purchase behavior data set is constructed. The purchase behavior data set Satisfies:

[0028]

[0029] Among them, To respectively represent the vectorized data sets of the first user to the last user.

[0030] Furthermore, in step S31, Satisfies Or Or , where Is a positive integer, Indicates that at least consecutive Clicks constitute a purchase. The acceptable error is 2 times, and the value of N is changed based on the need for accuracy.

[0031] Furthermore, in step S31, the click behavior And the purchase behavior Are both represented in vector form.

[0032] Furthermore, step S3 includes the following steps:

[0033] S34. Circular classification: Classify the click log sequence and the purchase behavior data set according to different users;

[0034] S35. Construct an initial weight vector and an initial value vector: According to the circular classification result, determine all click log sequences of the same user to construct an initial weight vector, and determine all subsequent purchase behaviors under the same click log sequence of the same user to construct an initial value vector;

[0035] S36. Attention weighting: Based on the attention mechanism of the neural network, perform weighted summation calculation on the initial value vector to find the target purchase behavior with the highest probability for all subsequent purchase behaviors relative to this click log sequence, construct a target value vector, and all target value vectors for this click log sequence form a target vector set;

[0036] S37. Iterative recursion: Use the recursive neural network method to perform iterative calculation on the click log sequence. The initial value vectors of two adjacent click log sequences jointly determine the target value vector of the first click log sequence as the output, determine the initial value vector of the second click log sequence as the output, and the input and output jointly serve as the input to recursively input the subsequent click log sequences. The neural network continuously iterates to optimize the weight vector of the neural network until convergence.

[0037] Further, in step S35, the initial weight vector is represented by , and the initial value vector is represented by . and respectively represent the initial weight and the initial value corresponding to the user in the th click log;

[0038] The target value vector is represented by to represent the mapping relationship;

[0039] The weighted target value vector is represented by . represents the initial weight vector with a certain initial value vector as a parameter;

[0040] The final delivery vector after recursive calculation is represented by . represents the maximum value selected from the different initial weight vectors obtained during the iteration process for the same initial value vector . Numerically, it is the maximum weight among the users of the th click log on the e-commerce platform, indicating that the th click log has the highest click probability among all users of the e-commerce platform, and the corresponding click vector represents the users most likely to convert into purchase behavior.

[0041] Further, in step S36, the initial value vector is calculated according to the following formula:

[0042]

[0043] where represents the multiplication of the initial weight vector and the initial value vector to weight out the target value vector . All target value vectors form a target vector set.

[0044] Further, in step S36, the recurrence equation for the first click log sequence is:

[0045]

[0046] where represents the purchase probability that the current user progresses to the next click.

[0047] Further, in step S37, the recurrence equation for the second click log sequence is:

[0048]

[0049] where Represents the purchase probability of progressing to the next user at the current click volume.

[0050] Furthermore, in step S37, the click log sequence includes clicking on product promotion language, product pictures, and product titles.

[0051] Beneficial effects

[0052] This solution proposes an e-commerce platform consumer behavior prediction and decision optimization method based on attention iteration, which can significantly improve the e-commerce platform's ability in predicting consumer behavior, optimizing decisions, and enhancing user experience. Specifically, there are the following significant beneficial effects:

[0053] 1. Improve the accuracy of consumer behavior prediction:

[0054] By introducing the attention mechanism and iterative optimization algorithm, this solution can effectively identify and focus on the key factors in consumer behavior. The attention mechanism can automatically assign weights to different click behaviors, purchase behaviors, and other data, thereby accurately capturing which behaviors have a greater impact on predicting consumers' future behaviors. For example, certain clicks of users may directly affect the purchase decision, while other behaviors may only be noise during the browsing process. By dynamically adjusting the weights, the model can automatically optimize the prediction results according to different scenarios and data situations, thus improving the accuracy of the prediction.

[0055] 2. Achieve personalized recommendations and precise decisions:

[0056] The implementation of this solution can accurately predict the next actions of users based on their historical behavior data and real-time behavior, thereby providing personalized recommendations and decision optimization for the e-commerce platform. For example, when a user browses a certain category of products, the platform can adjust the product recommendations in real time according to the predicted purchase behavior and optimize the display strategy. Such personalized recommendations can not only enhance the user experience but also effectively improve the conversion rate and sales volume.

[0057] 3. Optimize the data processing process and improve the processing efficiency:

[0058] In traditional e-commerce platforms, user behavior data is usually huge and complex, and how to extract valuable information from the massive data has always been a technical problem. This solution converts the original data into a high-quality data set for model learning through steps such as data preprocessing, behavior classification, and vectorization. At the same time, through the iterative optimization of the recurrent neural network, it can process multi-level and multi-dimensional data more efficiently and continuously optimize the model, improving the efficiency and accuracy of data processing.

[0059] 4. Better adapt to the dynamic changes of user behavior:

[0060] Consumer behavior is highly dynamic. Over time, users' interests and needs may change significantly. Traditional models often fail to capture these changes in real time. However, this solution combines a recurrent neural network and an attention mechanism to dynamically adjust user behavior. In practical applications, when a user's behavior changes, the platform can quickly update the prediction model and make corresponding decisions in a timely manner, thereby improving the flexibility and real-time performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] To deepen the understanding of the present invention, the following will further elaborate on the present invention in conjunction with embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0063] Embodiment 1

[0064] According to Figure 1 shown, this embodiment provides an e-commerce platform consumer behavior prediction and decision optimization method based on attention iteration, including the following steps:

[0065] S1. Data collection: Collect e-commerce information of the e-commerce platform according to requirements, including consumer behavior data, transaction data, and product data;

[0066] S2. Data preprocessing: Preprocess the e-commerce information according to usage requirements to obtain a preprocessed data set;

[0067] S3. Consumer classification: Construct a vectorized data set, a click log sequence, and a purchase behavior data set based on the preprocessed data set, classify the vectorized data set and the vectorized initial weight and value vector set, and perform weighted summation and recursive iterative calculation on the classified vectorized data set based on the attention mechanism;

[0068] S4. Establish a consumer prediction model: Input the vectorized data set that has undergone weighted summation and recursive iterative calculation into a classifier to determine the consumer to be predicted;

[0069] S5. Behavior prediction: Predict the subsequent behavior of the consumer to be predicted based on the consumer prediction model.

[0070] Further, step S3 is specifically as follows:

[0071] S31. Construct a vectorized data set:

[0072] Determine sequence data according to the preprocessed data set, and the sequence data satisfies One click behavior corresponds to one purchase behavior, where L is a positive integer and satisfies ;

[0073] The vectorized data set is represented by , where represents the th click behavior of the th user on the e-commerce platform, represents the purchase behavior completed by the th user on the e-commerce platform after their th click, where , , U and I respectively represent the total number of users and the total number of clicks;

[0074] S32. Determine the click log sequence: According to the vectorized data set, determine the vectorized data set that satisfies condition T1 as the click log sequence, where condition T1 is: at least consecutive clicks constitute a purchase;

[0075] S33. Determine the purchase behavior data set: Based on the vectorized data set , form a set by combining all vectors that satisfy at least one vectorized data set to construct the purchase behavior data set, and the purchase behavior data set satisfies:

[0076]

[0077] where to respectively represent the vectorized data sets of the first user to the last user.

[0078] Further, in step S31, satisfies or or , where is a positive integer, represents that at least consecutive clicks constitute a purchase, and the acceptable error is 2 times. The value of N is changed based on the need for accuracy.

[0079] Further, in step S31, both the click behavior and the purchase behavior are represented in vector form.

[0080] Further, step S3 includes the following steps:

[0081] S34. Circular Classification: Classify the click log sequence and the purchase behavior data set according to different users;

[0082] S35. Construct Initial Weight Vector and Initial Value Vector: According to the circular classification result, determine all click log sequences of the same user to construct the initial weight vector, and determine all subsequent purchase behaviors under the same click log sequence of the same user to construct the initial value vector;

[0083] S36. Attention Weighting: Based on the attention mechanism of the neural network, perform weighted summation calculation on the initial value vector to find the target purchase behavior with the highest probability for all subsequent purchase behaviors relative to this click log sequence, construct the target value vector, and all target value vectors for this click log sequence form the target vector set;

[0084] S37. Iterative Recursion: Use the recursive neural network method to perform iterative calculation on the click log sequence. The initial value vectors of two adjacent click log sequences jointly determine the target value vector of the first click log sequence as the output, determine the initial value vector of the second click log sequence as the output, and the input and output jointly serve as the input to recursively follow up the input of the click log sequence. The neural network continuously iterates to optimize the weight vector of the neural network until convergence.

[0085] Further, in step S35, the initial weight vector is represented by and the initial value vector is represented by . and respectively represent the initial weight and initial value corresponding to user in the th click log;

[0086] The target value vector is represented by to represent the mapping relationship;

[0087] The weighted target value vector is represented by . represents the initial weight vector with a certain initial value vector as a parameter;

[0088] The final delivery vector after recursive calculation is represented by . represents the maximum value selected from different initial weight vectors obtained during the iteration process for the same initial value vector . Numerically, it is the maximum weight of the th click log among the users on the e-commerce platform, indicating that the th click log has the highest probability of being clicked among all users on the e-commerce platform, and the corresponding click vector represents the user most likely to convert into a purchase behavior.

[0089] Further, in step S36, the initial value vector is calculated according to the following formula:

[0090]

[0091] wherein, represents the initial weight vector multiplies with the initial value vector to weight out the target value vector , and all the target value vectors form a target vector set.

[0092] Further, in step S36, the recurrence equation of the first click log sequence is:

[0093]

[0094] wherein, represents the purchase probability that the current user progresses to the next click.

[0095] Further, in step S37, the recurrence equation of the second click log sequence is:

[0096]

[0097] wherein, represents the purchase probability that progresses to the next user at the current click volume.

[0098] Further, in step S37, the click log sequence includes clicking on product promotion language, product pictures, and product titles. Specific Embodiment 2

[0100] Suppose an e-commerce platform hopes to optimize product recommendation and advertising placement strategies through in-depth analysis of consumer behavior. The platform has a large amount of user behavior data, including user click logs and purchase records. To improve the operation efficiency of the platform, the platform hopes to accurately predict the future purchase behavior of each user and make personalized recommendations based on these predictions.

[0101] Step Description

[0102] S1. Data Collection

[0103] The e-commerce platform collects the behavior data of users on the platform, including user click logs (such as information on clicking product pictures, titles, promotion language, etc.), transaction data (such as detailed information on purchased products), and product data (such as product categories, prices, inventory, etc.). Based on this data, the platform establishes a complete database containing users, products, and transactions.

[0104] S2. Data Preprocessing

[0105] After data collection, the platform cleaned and preprocessed these raw data, removed invalid or missing data, and performed time serialization on the user behavior data. The click behaviors of each user were sorted in chronological order to generate the user's behavior sequence. At the same time, the product information and transaction records were combined with the user behavior data to form a comprehensive preprocessed data set.

[0106] S3. Consumer Classification

[0107] In this step, the platform first constructed a vectorized data set that included the click behavior vectors and purchase behavior vectors of each user. Taking user A as an example, the click log of user A might contain the following data: the first click on product A, the second click on product B, and the third click on product C. Based on certain rules, the platform determined whether these click behaviors would be converted into purchase behaviors. For example, if user A finally purchased product C after clicking on product A and product B, this behavior was regarded as a purchase behavior.

[0108] The platform used these vectorized data sets to perform weighted summation and recursive iterative calculations based on the attention mechanism to identify which behavior sequences were the most important for predicting future purchase behaviors. In this way, the platform could efficiently capture the key click log sequences that influenced the purchase decision and weight them.

[0109] S4. Establish a Consumer Prediction Model

[0110] Next, the platform input the vectorized data set after weighted summation into a classifier and trained a consumer behavior prediction model through machine learning algorithms. This model could predict the future purchase behaviors of users based on their historical behaviors. For example, by analyzing the historical click records and purchase behaviors of user A, the model could predict the types of products that user A might purchase next or the possible responses to certain promotional activities.

[0111] S5. Behavior Prediction

[0112] Based on the above - constructed consumer prediction model, the platform could predict the subsequent behaviors of each user. Taking user B as an example, the prediction model could analyze that user B might click on a certain product or participate in a certain promotional activity in the future. The prediction results not only helped the platform understand user needs but also provided a decision - making basis for the personalized recommendation system, thereby improving the conversion rate and sales volume of the platform.

[0113] Detailed Description of Specific Steps

[0114] S31. Construct a Vectorized Data Set

[0115] Assume that the platform has 100 users, and the behavior sequences of each user have been converted into vectors. For example, the vectorized data set of user A may contain multiple click behaviors and subsequent purchase behaviors. Each click behavior and purchase behavior are represented in vector form as follows:

[0116] Represents the th click behavior of the user;

[0117] Represents the th purchase behavior after the click of the user.

[0118] Based on this data, the platform can construct a complete set of user behavior vectors.

[0119] S32. Determine the click log sequence

[0120] Based on the preprocessed data, the platform extracts the click log sequences that meet condition T1. Condition T1 requires at least L consecutive clicks to constitute a purchase behavior. For example, if a user clicks on product A, product B, and product C consecutively and finally purchases product C, then this behavior sequence will be considered to meet condition T1.

[0121] S33. Determine the set of purchase behavior data

[0122] The platform further constructs a set of purchase behavior data, which includes the click logs and purchase behaviors of all users. For each user, the platform combines their click behaviors and purchase behaviors into a set of vectors in sequence to form the set of purchase behavior data. This set can help the platform understand the purchase patterns of different users on the platform and thus guide subsequent recommendations and advertising placements.

[0123] S34. Loop classification

[0124] In this step, the platform classifies the click log sequences and the set of purchase behavior data of all users. The behavior sequences of each user are processed independently to ensure that the behavior data of each user can accurately reflect their unique consumption habits.

[0125] S35. Construct the initial weight vector and the initial value vector

[0126] Based on the classification results of each user, the platform assigns an initial weight vector and an initial value vector to the click log sequence of each user. The initial weight vector reflects the attention of the user to each click behavior, while the initial value vector represents the possible purchase probability of the user after this behavior.

[0127] S36. Attention weighting

[0128] With the help of the attention mechanism, the platform performs a weighted sum on the initial value vectors to calculate the target value vector for each click log sequence that is most likely to be converted into a purchase behavior. For example, the platform can identify that the second click behavior of user A (clicking on product B) is most likely to trigger a purchase behavior, so a higher weight is assigned to this behavior.

[0129] S37. Iterative recursion

[0130] Finally, the platform performs iterative calculations on the click log sequences through a recurrent neural network. The output of each click log sequence will be used as the input for the next sequence until the model converges. This iterative process helps the platform continuously optimize the prediction results and improve the accuracy of consumer behavior prediction.

[0131] Through the above steps, the e-commerce platform can efficiently predict users' purchase behaviors using the attention iteration mechanism and optimize corresponding decisions. The platform can not only accurately identify which behaviors have a significant impact on purchase decisions, but also provide personalized product recommendation and advertising placement strategies based on the prediction results, thereby improving the conversion rate and user satisfaction of the platform.

[0132] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An e-commerce platform consumer behavior prediction and decision optimization method based on attention iteration, characterized in that: The following steps are involved: S1. Data collection: collect e-commerce information from e-commerce platforms as needed, including consumer behavior data, transaction data, and product data; S2. Data preprocessing: preprocessing the e-commerce information according to usage requirements to obtain a preprocessed data set; Preprocessing removes invalid or missing data, and performs time serialization on user behavior data, sorting each user's click behavior in chronological order to generate a user behavior sequence; S3, consumer classification: construct a vectorized data set, a click log sequence and a purchase behavior data set based on the preprocessed data set, classify the vectorized data set and the vectorized initial weight and value vector set, and perform weighted summation and recursive iterative calculation on the classified vectorized data set based on the attention mechanism; specifically: S31. Build vectorized data set: Determine sequence data according to the preprocessed data set, the sequence data satisfies Each click behavior corresponds to one purchase behavior, and L is a positive integer and satisfies ; The vectorized data set is It indicates that, Indicates the first User's Click behavior, Indicates the first Users in their Purchases completed after clicks; , , U and I represent the total number of users and the total number of clicks, respectively; S32, determine the click log sequence: according to the vectorized data set, determine the vectorized data set that meets condition T1 as the click log sequence, and the condition T1 is: at least continuous Clicks constitute a purchase; S33. Determine the purchase behavior data set: vectorize the data set Based on at least one vectorized data set All vectors of form a set, construct a purchase behavior data set, purchase behavior data set satisfy: ； in, to Respectively represent the vectorized data sets from the first user to the last user; S34, cyclic classification: classifying the click log sequence and the purchase behavior data set according to different users; S35, constructing an initial weight vector and an initial value vector: according to the cyclic classification result, determining all click log sequences of the same user to construct an initial weight vector, and determining all subsequent purchase behaviors under the same click log sequence of the same user to construct an initial value vector; S36, attention weighting: performing weighted sum calculation on the initial value vector based on the attention mechanism of the neural network, finding the target purchase behavior with the greatest probability for all subsequent purchase behaviors relative to the click log sequence, and constructing a target value vector. All target value vectors for the click log sequence constitute a target vector set; S37, iterative recursion: adopt recursive neural network to perform iterative calculation on the click log sequence, the initial value vectors of two adjacent click log sequences jointly determine the target value vector of the first click log sequence as the output, determine the initial value vector of the second click log sequence as the output, the input and output are jointly used as the input of the subsequent click log sequence recursively, and the neural network optimizes the weight vector of the neural network through continuous iteration until convergence; S4. Establishing a consumer prediction model: inputting the vectorized data set after weighted summation and recursive iterative calculation into the classifier to determine the consumers to be predicted; S5. Behavior prediction: predicting the subsequent behavior of the consumer to be predicted based on the consumer prediction model.

2. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 1, characterized in that: In step S31, satisfy or or ,in, is a positive integer, Indicates at least continuous Clicks constitute a purchase, and the acceptable error is 2 times. The value of N is changed based on the need for accuracy.

3. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 2, characterized in that: In step S31, the click behavior and the purchase behavior are expressed in vector form.

4. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 3 is characterized by: In step S35, the initial weight vector is The initial value vector is represented by express, and Respectively expressed in Click logs for users The corresponding initial weights and initial values; The target value vector is Indicates a mapping relationship; The weighted target value vector is express, represents the initial weight vector with a certain initial value vector as a parameter; The final delivery vector after recursive calculation is express, Indicates that in the iteration process, for the same initial value vector The different initial weight vectors obtained The maximum value selected from The maximum weight of the click log among the users of the e-commerce platform indicates The click log has the highest probability of being clicked among all e-commerce platform users, and the corresponding click vector represents the user who is most likely to convert into purchasing behavior.

5. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 4, characterized in that: In step S36, the initial value vector is calculated as follows: ； in, represents the initial weight vector With the initial value vector Multiply and weight the target value vector , all target value vectors constitute the target vector set.

6. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 5, characterized in that: In step S36, the recursive equation of the first click log sequence is: ； in, Indicates the purchase probability of the current user progressing to the next click.

7. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 6, characterized in that: In step S37, the recursive equation of the second click log sequence is: ； in, Indicates the purchase probability of the next user at the current click volume.

8. The method for predicting consumer behavior and optimizing decision-making on an e-commerce platform based on attention iteration according to claim 7, characterized in that: In step S37, the click log sequence includes product promotion words, product pictures and product titles.

Images