A multi-type behavior sequence prediction recommendation method and system, and a storage medium
By constructing a behavior encoder and establishing implicit and explicit dependencies, the problem of complex coupled dependencies in multi-behavior fusion is solved, thus improving the recommendation prediction effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF TECH
- Filing Date
- 2024-01-16
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies in e-commerce session recommendations based on multi-behavioral feature fusion fail to fully integrate various behavioral information and cannot effectively handle the complex coupling and dependency relationships between behaviors, resulting in poor recommendation performance.
By constructing a behavior encoder to encode shopping basket data corresponding to various behaviors, multiple fused behavior vector sequences are obtained, and implicit and explicit dependencies are established. Combining user feature information and product feature information, the probability of a product appearing in the predicted shopping basket for each behavior is calculated.
It achieves full integration of multi-behavioral information, resolves the complex coupling and dependency relationships between behaviors, and improves the recommendation prediction effect.
Smart Images
Figure CN117893283B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data mining, and more specifically, to a method, system, and storage medium for predicting and recommending multi-type behavior sequences. Background Technology
[0002] In recent years, research in the field of recommendation systems has developed rapidly. Its aim is to learn from users' historical behavioral data through modeling, and then recommend the next item / basket (a basket consists of a series of related items) that users are interested in. Classic research in this area focuses on predicting the next item / basket based on one type of target behavioral data and predicting the next item / basket based on multiple types of behavioral data. Work on predicting the next item / basket based on one type of target behavioral data only considers capturing the user's interests and preferences from one set of behavioral data. Recommendation work in this direction mainly uses Markov chain algorithms, neural network algorithms, and pooling / attention mechanisms. These algorithms can only capture the user's needs and preferences under that specific behavior. Due to the sparsity of the data, capturing user preference information from only one behavioral feedback is inaccurate.
[0003] Existing technology discloses an e-commerce conversation recommendation method and system based on multi-behavioral feature fusion. The method includes the following steps: acquiring four types of user conversation behavior data from an e-commerce database: clicks, favorites, purchases, and adding to cart; constructing an e-commerce conversation recommendation model based on multi-behavioral feature fusion, which includes a behavioral feature extraction module, a product feature extraction module, a high-order product feature extraction module, a conversation feature extraction module, and a product recommendation module; training the e-commerce conversation recommendation model using the four types of conversation behavior data from the e-commerce database; and using the trained e-commerce conversation recommendation model to recommend products to users in the conversation and output the recommendation results. However, this method does not fully integrate multi-behavioral information and cannot effectively handle the complex coupling dependencies between behaviors, resulting in poor final prediction and recommendation performance. Summary of the Invention
[0004] The purpose of this invention is to disclose a multi-type behavior sequence prediction and recommendation method, system, and storage medium that provides better prediction and recommendation performance.
[0005] To achieve the above objectives, the present invention provides a multi-type behavior sequence prediction and recommendation method, comprising:
[0006] S1: Obtain user characteristic information, product information, and shopping basket data corresponding to various behaviors; product information can be divided into product information that has been interacted with by the user and product information that has not been interacted with by the user, depending on whether it has been interacted with by the user.
[0007] S2: Construct a behavior encoder to encode each type of shopping basket data in the shopping basket data corresponding to multiple behaviors, and obtain multiple shopping basket vector sequences; encode user feature information to obtain user feature vectors; encode product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; encode product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user.
[0008] S3: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, obtain multiple fused behavior vector sequences;
[0009] S4: Establish implicit dependency relationships between each fusion behavior vector sequence of multiple final behavior vector sequences and other fusion behavior vector sequences to obtain multiple implicit behavior mixed feature vectors; establish explicit dependency relationships between each fusion behavior vector sequence of multiple fusion behavior vector sequences and other fusion behavior vector sequences to obtain multiple explicit behavior mixed feature vectors.
[0010] S5: Obtain the probability of all products appearing in each behavior prediction shopping basket based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user.
[0011] S6: Based on the predicted probability of all products appearing in the shopping basket for each behavior, complete the recommendation display of products.
[0012] Furthermore, in step S1, the user feature information integrates a user's long-term feature information, including a user's personalized preference information; the behavior includes click behavior, add-to-cart behavior, favorite behavior, and purchase behavior; and the shopping basket data is a collection of products over a period of time.
[0013] Furthermore, in step S2, the behavior encoder includes pooling operations and attention operations.
[0014] Furthermore, in step S3, multiple fused behavior vector sequences are obtained based on user feature information, multiple behavior data, multiple behavior vector sequences, and user feature vectors;
[0015] include:
[0016] S3.1: Construct a relationship diagram of user characteristics and shopping basket data corresponding to various behaviors;
[0017] S3.2: Calculate the similarity between user feature vectors and multiple shopping basket vector sequences based on the relationship graph, and use this as the weight of each behavior;
[0018] S3.3: The shopping basket data corresponding to multiple behaviors are weighted and summed according to the weight of each behavior to obtain the preliminary basket representation of the sequence;
[0019] S3.4: The model performs a max pooling operation on the user feature vector and the shopping basket data corresponding to each behavior in the dimension to obtain multiple fused behavior vector sequences.
[0020] Further, in step S4, establishing implicit dependency relationships between each of the multiple fusion behavior vector sequences and other fusion behavior vector sequences to obtain multiple implicit behavior hybrid feature vectors includes: setting a behavior weight matrix for each behavior, which can learn the degree of correlation between the behavior and other behaviors; multiplying the concatenated fusion behavior vector sequence with the weight matrix under that behavior to obtain multiple implicit behavior hybrid feature vectors that capture the implicit dependency relationships between behaviors; specifically, let This represents the user's behavioral data within the nth time window. This represents the k-th action. For behavior The basket representation of the next sequence is given by denoted by , where each fused behavior vector sequence is concatenated with all behaviors. , ,Behavior The behavioral weight matrix below is denoted as Then for behavior The implicit basket mixture feature vector captures the implicit dependency associations between behaviors. , It can be represented as: ,in It is an activation function.
[0021] Further, in step S4, an explicit dependency relationship is established between each of the multiple fusion behavior vector sequences and the other fusion behavior vector sequences to obtain multiple explicit behavior hybrid feature vectors, including: Let Indicates a click action. This indicates the act of purchasing an additional car. Indicates the act of collecting. Indicates purchasing behavior. This indicates behavior that captures explicit dependencies. The explicit basket mixing feature vector under the following conditions, then It can be represented as: , , , .
[0022] Further, in step S5, obtaining the probability of all products appearing in the shopping basket corresponding to each behavior based on multiple implicit behavior mixed feature vectors, multiple explicit behavior mixed feature vectors, product feature vectors that have interacted with the user, and product feature vectors that have not interacted with the user includes: calculating the similarity between multiple implicit behavior mixed feature vectors and product feature vectors that have not interacted with the user to obtain the probability of products that have not interacted with the user appearing in each behavior's predicted shopping basket; calculating the similarity between explicit basket mixed feature vectors and product feature vectors that have interacted with the user to obtain the probability of products that have interacted with the user appearing in each behavior's predicted shopping basket; integrating the probabilities of products that have not interacted with the user and products that have interacted with the user to obtain the probability of all products appearing in each behavior's predicted shopping basket, specifically: Let Indicates behavior Predict the probability of products that have not been interacted with by the user appearing in the shopping basket for each behavior. Indicates behavior The probability of products interacted with by the user appearing in the shopping basket for each behavior is predicted; let This represents the set of product vectors that have not been interacted with by the user. Let the set of product vectors that have been interacted with by the user be... , By combining the probabilities of products that have not been interacted with by the user and the probabilities of products that have been interacted with by the user in each behavior prediction shopping basket, the probability of all products appearing in each behavior prediction shopping basket is obtained. , .
[0023] Furthermore, step S5 also includes: after obtaining the probability of all items appearing in each behavior's predicted shopping basket, calculating the negative log-likelihood loss function as the loss value for the probability of appearance in each behavior's predicted shopping basket, and performing training optimization; for the loss function , can be represented as ,in For the actual next basket The predicted probability set for each item in the set, for Its prediction probability .
[0024] Furthermore, this invention also provides a multi-type behavior sequence prediction and recommendation system, including:
[0025] Acquisition module: Acquires user characteristic information, product information, and shopping basket data corresponding to various behaviors; product information can be divided into product information that has been interacted with by the user and product information that has not been interacted with by the user, depending on whether it has been interacted with by the user.
[0026] Encoding Module: Constructs a behavior encoder to encode each type of shopping basket data corresponding to multiple behaviors, obtaining multiple shopping basket vector sequences; encodes user feature information to obtain user feature vectors; encodes product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; and encodes product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user.
[0027] Fusion module: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, it obtains multiple fused behavior vector sequences;
[0028] Dependency Module: Establishes implicit dependency relationships between each fused behavior vector sequence of multiple final behavior vector sequences and other fused behavior vector sequences to obtain multiple implicit behavior mixed feature vectors; establishes explicit dependency relationships between each fused behavior vector sequence of multiple fused behavior vector sequences and other fused behavior vector sequences to obtain multiple explicit behavior mixed feature vectors.
[0029] The calculation module obtains the probability of each product appearing in the predicted shopping basket for each behavior based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user.
[0030] Recommendation display module: Based on the predicted probability of each product appearing in the shopping basket for each behavior, the module completes the recommendation display of products.
[0031] Furthermore, the present invention also provides a multi-type behavior sequence prediction and recommendation storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the multi-type behavior sequence prediction and recommendation method.
[0032] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:
[0033] This invention improves prediction and recommendation performance by establishing implicit and explicit dependencies between each fused behavior vector sequence and other fused behavior vector sequences, thereby fully integrating multi-behavior information and resolving the complex coupling dependencies between behaviors. Attached Figure Description
[0034] Figure 1 This is a flowchart of a multi-type behavior sequence prediction and recommendation method as described in Example 1;
[0035] Figure 2 This is a framework diagram of a multi-type behavior sequence prediction and recommendation system as described in Example 3;
[0036] Figure 3 A complete schematic diagram of the architecture of a multi-type behavior sequence prediction and recommendation model;
[0037] Figure 4 This is a schematic diagram of a multi-type behavior sequence prediction and recommendation method;
[0038] Figure 5 A schematic diagram illustrating an application scenario for a multi-type behavior sequence prediction and recommendation method; Detailed Implementation
[0039] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.
[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0041] Example 1:
[0042] Preferred embodiments of the present invention provide, as follows Figure 1 The multi-type behavior sequence prediction recommendation method shown includes:
[0043] S1: Obtain user characteristic information, product information, and shopping basket data corresponding to various behaviors; product information can be divided into product information that has been interacted with by the user and product information that has not been interacted with by the user, depending on whether it has been interacted with by the user.
[0044] S2: Construct a behavior encoder to encode each type of shopping basket data in the shopping basket data corresponding to multiple behaviors, and obtain multiple shopping basket vector sequences; encode user feature information to obtain user feature vectors; encode product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; encode product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user.
[0045] S3: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, obtain multiple fused behavior vector sequences;
[0046] S4: Establish implicit dependency relationships between each fusion behavior vector sequence of multiple final behavior vector sequences and other fusion behavior vector sequences to obtain multiple implicit behavior mixed feature vectors; establish explicit dependency relationships between each fusion behavior vector sequence of multiple fusion behavior vector sequences and other fusion behavior vector sequences to obtain multiple explicit behavior mixed feature vectors.
[0047] S5: Obtain the probability of all products appearing in each behavior prediction shopping basket based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user.
[0048] S6: Based on the predicted probability of all products appearing in the shopping basket for each behavior, complete the recommendation display of products.
[0049] This embodiment establishes implicit and explicit dependencies between each fused behavior vector sequence and other fused behavior vector sequences, respectively, to fully integrate multi-behavior information and resolve the complex coupling dependencies between behaviors, thereby improving the prediction and recommendation effect.
[0050] Example 2:
[0051] The preferred embodiments of the present invention are further disclosed based on Embodiment 1:
[0052] In step S1, the user feature information integrates a user's long-term feature information, including a user's personalized preference information; the behavior includes click behavior, add-to-cart behavior, favorite behavior, and purchase behavior; and the shopping basket data is a collection of products over a period of time.
[0053] In step S2, the behavior encoder includes pooling operations and attention operations.
[0054] In step S3, multiple fused behavior vector sequences are obtained based on user feature information, multiple behavioral data, multiple behavior vector sequences, and user feature vectors;
[0055] include:
[0056] S3.1: Construct a relationship diagram of user characteristics and shopping basket data corresponding to various behaviors;
[0057] S3.2: Calculate the similarity between user feature vectors and multiple shopping basket vector sequences based on the relationship graph, and use this as the weight of each behavior;
[0058] S3.3: The shopping basket data corresponding to multiple behaviors are weighted and summed according to the weight of each behavior to obtain the preliminary basket representation of the sequence;
[0059] S3.4: The model performs a max pooling operation on the user feature vector and the shopping basket data corresponding to each behavior in the dimension to obtain multiple fused behavior vector sequences.
[0060] In step S4, establishing implicit dependency relationships between each of the multiple fusion behavior vector sequences and other fusion behavior vector sequences to obtain multiple implicit behavior hybrid feature vectors includes: setting a behavior weight matrix for each behavior, which can learn the degree of correlation between the behavior and other behaviors; multiplying the concatenated fusion behavior vector sequence with the weight matrix under that behavior to obtain multiple implicit behavior hybrid feature vectors that capture the implicit dependency relationships between behaviors; specifically, let This represents the user's behavioral data within the nth time window. This represents the k-th action. For behavior The basket representation of the next sequence is given by denoted by , where each fused behavior vector sequence is concatenated with all behaviors. , ,Behavior The behavioral weight matrix below is denoted as Then for behavior The implicit basket mixture feature vector captures the implicit dependency associations between behaviors. , It can be represented as: ,in It is an activation function.
[0061] In step S4, an explicit dependency relationship is established between each of the multiple fusion behavior vector sequences and the other fusion behavior vector sequences to obtain multiple explicit behavior hybrid feature vectors, including: Let Indicates a click action. This indicates the act of purchasing an additional car. Indicates the act of collecting. Indicates purchasing behavior. This indicates behavior that captures explicit dependencies. The explicit basket mixing feature vector under the following conditions, then It can be represented as: , , , .
[0062] In step S5, obtaining the probability of all items appearing in the shopping basket corresponding to each behavior based on multiple implicit behavior mixed feature vectors, multiple explicit behavior mixed feature vectors, feature vectors of items that have interacted with the user, and feature vectors of items that have not interacted with the user includes: calculating the similarity between multiple implicit behavior mixed feature vectors and feature vectors of items that have not interacted with the user to obtain the probability of items that have not interacted with the user appearing in the predicted shopping basket for each behavior; calculating the similarity between explicit basket mixed feature vectors and feature vectors of items that have interacted with the user to obtain the probability of items that have interacted with the user appearing in the predicted shopping basket for each behavior; and integrating the probabilities of items that have not interacted with the user and items that have interacted with the user to obtain the probability of all items appearing in the predicted shopping basket for each behavior, specifically: Let Indicates behavior Predict the probability of products that have not been interacted with by the user appearing in the shopping basket for each behavior. Indicates behavior The probability of products interacted with by the user appearing in the shopping basket for each behavior is predicted; let This represents the set of product vectors that have not been interacted with by the user. Let the set of product vectors that have been interacted with by the user be... , By combining the probabilities of products that have not been interacted with by the user and the probabilities of products that have been interacted with by the user in each behavior prediction shopping basket, the probability of all products appearing in each behavior prediction shopping basket is obtained. , .
[0063] Step S5 further includes: after obtaining the probability of all items appearing in each behavior's predicted shopping basket, calculating the negative log-likelihood loss function as the loss value for the probability of each item appearing in each behavior's predicted shopping basket, and performing training optimization; for the loss function , can be represented as ,in For the actual next basket The predicted probability set for each item in the set, for Its prediction probability .
[0064] This embodiment uses max pooling to extract the most salient features of users' long-term and short-term interests, fully capturing the long-term and short-term preferences of individual users. By establishing implicit and explicit dependencies between each fused behavior vector sequence and other fused behavior vector sequences, the multi-behavior information can be fully integrated, solving the complex coupling dependencies between behaviors, thereby improving the prediction and recommendation effect.
[0065] Example 3:
[0066] Preferred embodiments of the present invention provide, as follows Figure 2 The multi-type behavior sequence prediction recommendation system shown includes:
[0067] Acquisition module: Acquires user characteristic information, product information, and shopping basket data corresponding to various behaviors; product information can be divided into product information that has been interacted with by the user and product information that has not been interacted with by the user, depending on whether it has been interacted with by the user.
[0068] Encoding Module: Constructs a behavior encoder to encode each type of shopping basket data corresponding to multiple behaviors, obtaining multiple shopping basket vector sequences; encodes user feature information to obtain user feature vectors; encodes product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; and encodes product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user.
[0069] Fusion module: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, it obtains multiple fused behavior vector sequences;
[0070] Dependency Module: Establishes implicit dependency relationships between each fused behavior vector sequence of multiple final behavior vector sequences and other fused behavior vector sequences to obtain multiple implicit behavior mixed feature vectors; establishes explicit dependency relationships between each fused behavior vector sequence of multiple fused behavior vector sequences and other fused behavior vector sequences to obtain multiple explicit behavior mixed feature vectors.
[0071] The calculation module obtains the probability of each product appearing in the predicted shopping basket for each behavior based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user.
[0072] Recommendation display module: Based on the predicted probability of each product appearing in the shopping basket for each behavior, the module completes the recommendation display of products.
[0073] This embodiment establishes implicit and explicit dependencies between each fused behavior vector sequence and other fused behavior vector sequences, respectively, to fully integrate multi-behavior information and resolve the complex coupling dependencies between behaviors, thereby improving the prediction and recommendation effect.
[0074] Example 4:
[0075] A preferred embodiment of the present invention provides a multi-type behavior sequence prediction and recommendation storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the multi-type behavior sequence prediction and recommendation method.
[0076] In summary, the preferred embodiment of this invention provides a method, system, and storage medium for predicting and recommending multiple types of behavior sequences. The method includes: acquiring user feature information, product information, and shopping basket data corresponding to multiple behaviors; then encoding these data; obtaining multiple fused behavior vector sequences based on the encoded vectors; establishing implicit and explicit dependency relationships between each fused behavior vector sequence and other fused behavior vector sequences to obtain multiple implicit and explicit behavior mixed feature vectors; calculating the probability of each product appearing in each behavior's predicted shopping basket; and finally, displaying the recommended products. This invention, by establishing implicit and explicit dependency relationships respectively, can fully fuse multiple behavior information and resolve the complex coupling dependencies between behaviors, thereby improving the prediction and recommendation effect.
[0077] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A method for predicting and recommending multi-type behavior sequences, characterized in that, Includes the following steps: S1: Obtain user characteristic information, product information, and shopping basket data corresponding to various behaviors; product information is divided into product information that has been interacted with by the user and product information that has not been interacted with by the user, depending on whether it has been interacted with by the user's behavior. S2: Construct a behavior encoder to encode each type of shopping basket data in the shopping basket data corresponding to multiple behaviors, and obtain multiple shopping basket vector sequences; encode user feature information to obtain user feature vectors; encode product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; encode product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user. S3: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, obtain multiple fused behavior vector sequences; S4: Establish implicit dependency relationships between each fusion behavior vector sequence of multiple final behavior vector sequences and other fusion behavior vector sequences to obtain multiple implicit behavior hybrid feature vectors; including: setting a behavior weight matrix for each behavior, which learns the degree of correlation between the behavior and other behaviors, multiplying the concatenated fusion behavior vector sequence with the weight matrix under the behavior to obtain multiple implicit behavior hybrid feature vectors that capture the implicit dependency relationships between behaviors; By establishing explicit dependencies between each of the multiple fusion behavior vector sequences and the other fusion behavior vector sequences, multiple explicit behavior hybrid feature vectors are obtained; including: Let Indicates a click action. This indicates the act of purchasing an additional car. Indicates the act of collecting. Indicates purchasing behavior. This indicates behavior that captures explicit dependencies. The explicit basket mixing feature vector under the following conditions, then Represented as: , , , ; S5: Obtain the probability of all products appearing in each behavior prediction shopping basket based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user. S6: Based on the predicted probability of all products appearing in the shopping basket for each behavior, complete the recommendation display of products.
2. The multi-type behavior sequence prediction and recommendation method according to claim 1, characterized in that, In step S1, the user feature information integrates a user's long-term feature information, including a user's personalized preference information; the behavior includes click behavior, add-to-cart behavior, favorite behavior, and purchase behavior; the shopping basket data is a collection of products over a period of time.
3. The multi-type behavior sequence prediction and recommendation method according to claim 1, characterized in that, In step S2, the behavior encoder includes pooling operations and attention operations.
4. The multi-type behavior sequence prediction and recommendation method according to claim 1, characterized in that, In step S3, multiple fused behavior vector sequences are obtained based on user feature information, multiple behavioral data, multiple behavior vector sequences, and user feature vectors; include: S3.1: Construct a relationship diagram of user characteristics and shopping basket data corresponding to various behaviors; S3.2: Calculate the similarity between user feature vectors and multiple shopping basket vector sequences based on the relationship graph, and use this as the weight of each behavior; S3.3: The shopping basket data corresponding to multiple behaviors are weighted and summed according to the weight of each behavior to obtain the preliminary basket representation of the sequence; S3.4: The model performs a max pooling operation on the user feature vector and the shopping basket data corresponding to each behavior in the dimension to obtain multiple fused behavior vector sequences.
5. The multi-type behavior sequence prediction and recommendation method according to claim 1, characterized in that, In step S4, a behavior weight matrix is set for each behavior. This weight matrix learns the correlation between the behavior and other behaviors. The concatenated sequence of fused behavior vectors is multiplied by the weight matrix under that behavior to obtain multiple implicit behavior hybrid feature vectors that capture the implicit dependencies between behaviors; specifically: Let This represents the user's behavioral data within the nth time window. This represents the k-th action. For behavior The basket representation of the next sequence is given by denoted by , where each fused behavior vector sequence is concatenated with all behaviors. , ,Behavior The behavioral weight matrix below is denoted as Then for behavior The implicit basket mixture feature vector captures the implicit dependency associations between behaviors. , Represented as: ,in It is an activation function.
6. A multi-type behavior sequence prediction and recommendation system, characterized in that, include: Acquisition module: Acquires user characteristic information, product information, and shopping basket data corresponding to various behaviors; Product information is categorized into product information that has interacted with the user and product information that has not interacted with the user, based on whether or not it has been interacted with by the user. Encoding Module: Constructs a behavior encoder to encode each type of shopping basket data corresponding to multiple behaviors, obtaining multiple shopping basket vector sequences; encodes user feature information to obtain user feature vectors; encodes product information that has been interacted with by the user to obtain product feature vectors that have been interacted with by the user; and encodes product information that has not been interacted with by the user to obtain product feature vectors that have not been interacted with by the user. Fusion module: Based on user feature information, shopping basket data corresponding to multiple behaviors, multiple shopping basket vector sequences, and user feature vectors, it obtains multiple fused behavior vector sequences; Dependency Module: Establishes implicit dependency relationships between each fused behavior vector sequence of multiple final behavior vector sequences and other fused behavior vector sequences to obtain multiple implicit behavior hybrid feature vectors; including: setting a behavior weight matrix for each behavior, which learns the degree of correlation between the behavior and other behaviors, and multiplying the concatenated fused behavior vector sequence with the weight matrix under the behavior to obtain multiple implicit behavior hybrid feature vectors that capture the implicit dependency relationships between behaviors; By establishing explicit dependencies between each of the multiple fusion behavior vector sequences and the other fusion behavior vector sequences, multiple explicit behavior hybrid feature vectors are obtained; including: Let Indicates a click action. This indicates the act of purchasing an additional car. Indicates the act of collecting. Indicates purchasing behavior. This indicates behavior that captures explicit dependencies. The explicit basket mixing feature vector under the following conditions, then Represented as: , , , ; The calculation module obtains the probability of each product appearing in the predicted shopping basket for each behavior based on the mixed feature vectors of multiple implicit behaviors, the mixed feature vectors of multiple explicit behaviors, the feature vectors of products that have interacted with the user, and the feature vectors of products that have not interacted with the user. Recommendation display module: Based on the predicted probability of each product appearing in the shopping basket for each behavior, the module completes the recommendation display of products.
7. A multi-type behavior sequence prediction and recommendation storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the multi-type behavior sequence prediction and recommendation method according to any one of claims 1 to 5.