Data recommendation method and device, equipment and medium
By using neural network models and attention mechanisms to process user interaction data and product characteristics, and generating personalized recommendation lists, the problem of low product recommendation accuracy caused by information explosion is solved, and the accuracy of recommendations and user satisfaction is improved.
Patent Information
- Application Number
- CN202411469703.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-21
- Publication Date
- 2025-02-11
AI Technical Summary
Due to the explosive growth in information acquisition and the limited ability of users to select information, the accuracy of product recommendations is low.
By obtaining user product information click data and converting it into product feature vectors, the product information relevance is calculated using preset neural network models, combining cross attention mechanisms and self-attention mechanisms, the interest weights of interactive data and product feature vectors are calculated, and personality is generated. Recommended product list.
It improves the accuracy of product recommendations, makes recommendations more in line with users' actual needs and preferences, and enhances user satisfaction and platform sales conversion rate.
Smart Images
Figure CN119417564A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a data recommendation method, device, equipment and medium. Background Art
[0002] Product recommendation algorithms are a very critical part of e-commerce platforms. They help users discover products that they may be interested in, while also helping merchants increase sales and customer satisfaction. Here are some common product recommendation algorithms: Content-based Filtering: This method recommends products with similar characteristics based on the user's past purchase or browsing history. It usually analyzes the attributes of the product, such as category, brand, price, user reviews, etc. Collaborative Filtering: Collaborative filtering is divided into two types: user-user collaborative filtering and item-item collaborative filtering. User-user collaborative filtering recommends products that are liked by other users who have similar purchase or rating behavior to the target user. Item-item collaborative filtering recommends other products that are similar to products that the user liked in the past. Model-based Recommendations: These algorithms use machine learning models to predict user preferences for products. Common models include matrix factorization (such as SVD) and deep learning models (such as neural collaborative filtering NCF). Hybrid Recommendation Systems: Combine multiple recommendation algorithms to improve the accuracy and coverage of recommendations. For example, combining content-based recommendations and collaborative filtering. Knowledge-based Recommendations: This method uses information about users (such as age, gender, and geographic location) and information about products to recommend products. Usually a clear user profile is required. Social Network Recommendations: Use the user's social network relationships to recommend products, such as products that friends or people they follow like. Context-aware Recommendations: Consider the user's current context, such as time, location, weather, etc., to recommend products. Rule-based Recommendations: Recommend products based on predefined rules, such as "Users who bought this product also bought...". Graph-based Recommendations: Use nodes and edges in graph theory to represent the relationship between users and products, and discover recommendations through graph algorithms. Deep Learning-based Recommendations: Use deep learning techniques, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to handle complex recommendation problems. However, each algorithm has its advantages and limitations. In practical applications, the appropriate algorithm or algorithm combination is usually selected based on business needs and data characteristics.
[0003] Especially in today's era of explosive growth in information, recording users' online behavior and mining their intentions and preferences can provide strong support for corporate operations and marketing. For example, the insurance industry has gradually become an industry where research on the characteristics of users' insurance behavior has received more attention from the industry. The business model of the insurance industry has generated a large amount of high-quality and valuable data, which has great mining and use value. On the one hand, insurance users can obtain explosive growth in information, and on the other hand, the limited ability of users to select information makes it necessary to solve the information overload phenomenon faced by insurance users through personalized recommendation technology. Therefore, when mining user behavior for intentions and preferences, the low accuracy of product recommendations due to the explosive growth of information and the limited ability of users to select information is a technical problem that needs to be solved urgently. Summary of the invention
[0004] The present invention provides a data recommendation method, apparatus, computer equipment and medium to solve the technical problem of low accuracy of product recommendations caused by the explosive growth of information acquisition and the limited ability of users to select information when mining the intentions and preferences of current user behaviors.
[0005] In a first aspect, a data recommendation method is provided, comprising:
[0006] Obtain user product information click data and convert it into product feature vector;
[0007] Inputting the product feature vector into a preset neural network model to calculate product information relevance, and extracting a first candidate product feature vector within a preset relevance sorting interval according to the product information relevance;
[0008] Acquire client interaction data, and associate the interaction data with the first candidate product feature vector using a cross-attention mechanism;
[0009] Calculating interest weights of the interaction data and the first candidate product feature vector, and outputting a corresponding second candidate product feature vector;
[0010] A self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list.
[0011] In a second aspect, a data recommendation device is provided, comprising:
[0012] The acquisition module is used to obtain the user's product information click data and convert it into a product feature vector;
[0013] A first candidate product feature vector module, used to input the product feature vector into a preset neural network model to calculate product information relevance, and extract the first candidate product feature vector within a preset relevance sorting interval according to the product information relevance;
[0014] A cross-attention mechanism module, used for obtaining the interaction data of the client, and associating the interaction data with the first candidate product feature vector by using the cross-attention mechanism;
[0015] A second candidate product feature vector module, used to calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector;
[0016] The product recommendation module is used to calculate the interaction data between the second candidate product feature vector and the client by using a self-attention mechanism, output a behavior interest weight vector, and generate an associated product recommendation list.
[0017] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned data recommendation method when executing the computer program.
[0018] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned data recommendation method are implemented.
[0019] In the scheme implemented by the above-mentioned data recommendation method, device, computer equipment and storage medium, it is possible to obtain user product information click data and convert it into a product feature vector; input the product feature vector into a preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector within the preset relevance sorting interval based on the product information relevance; obtain the client's interaction data, and use a cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weights of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use a self-attention mechanism to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, which solves the current technical problem of low product recommendation accuracy due to the explosive growth of information acquisition and the limited ability of users to select information when mining user behavior to mine their intentions and preferences.
[0020] At the same time, the client's interaction data can be distinguished and extracted effective contextual information by adopting preset neural network models, cross-attention mechanisms, and self-attention mechanisms. It not only considers the direct interaction between users and products, but also considers the user's environment, historical behavior patterns, and other factors that may affect user decisions, which helps to improve the effectiveness of the recommendation system and make recommendations more in line with users' actual needs and preferences.
[0021] Furthermore, a preset neural network model is used to adaptively align the feature vector of the second candidate product, which can automatically learn the optimal mapping of input features to output space. This means that interest representations can be adaptively converted without manually designing feature conversion logic, effectively processing user interest representations, and generating more personalized and accurate recommendations by learning the complex interactions between users and products. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0023] Figure 1 is a schematic diagram of an application environment of a data recommendation method in an embodiment of the present invention;
[0024] Figure 2 is a flow chart of a data recommendation method in one embodiment of the present invention;
[0025] Figure 3 is a structural diagram of a data recommendation device in one embodiment of the present invention;
[0026] Figure 4 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;
[0027] Figure 5 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] The data recommendation method provided by the embodiment of the present invention can be applied in Figure 1 In the application environment, the client communicates with the server through the network. The server can obtain the user product information click data through the client and convert it into a product feature vector; input the product feature vector to the preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector in the preset relevance sorting interval according to the product information relevance; obtain the client's interaction data, and use the cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use the self-attention mechanism to calculate the interaction data of the second candidate product feature vector and the client, output the behavior interest weight vector, and generate an associated product recommendation list, and return it to the client. In the present invention, for some common click-through rate prediction models, the context information of user decisions is not fully utilized in the recommendation or advertising system, so it cannot achieve good performance, and can simultaneously extract explicit and implicit information in the product. Among them, the client can be but not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.
[0030] See also Figure 2 As shown, Figure 2 A flow chart of a data recommendation method provided by an embodiment of the present invention includes the following steps:
[0031] S10: Obtain user product information click data and convert it into a product feature vector;
[0032] In this embodiment, the acquisition of user product information click data and conversion into a product feature vector includes:
[0033] Obtain user product information click data, and parse out product click sequence and product information data;
[0034] Filter the product information data according to the product click order to obtain related product information click data;
[0035] The related product information click data is converted into the product feature vector.
[0036] It can be understood that, in this embodiment, it is composed of a plurality of CIU modules (Context Interaction Unit) and one AIAU module (Adaptive Interest Aggregation Unit).
[0037] In one embodiment, for example, each page and each candidate product's information (ID, price, page position, etc.) is converted into a vector and encoded as input, and the following features are defined:
[0038] The user's click sequence: xc = {xc1, xc2, ..., xc s}
[0039] The product information on the page is represented as: i ={p i,1 ,p i,2 ,...,p i,p}
[0040] The candidate product information is represented as: xs = {xs1, xs2, ..., xs C}
[0041] For example, a user browses products on an e-commerce platform and is particularly interested in a newly released smartphone, and clicks to view the details. Product classification: The ISU module in the CIU module analyzes the smartphone page viewed by the user and divides the other products on the page into two categories: products related to smartphones (such as phone cases, headphones, and chargers) and unrelated products (such as kitchen utensils and clothing). The click data of the related product information is converted into the product feature vector.
[0042] S11, inputting the product feature vector into a preset neural network model to calculate product information relevance, and extracting a first candidate product feature vector within a preset relevance sorting interval according to the product information relevance;
[0043] This step may include:
[0044] Input the product feature vector into a preset neural network model, and calculate the relevance of at least two pieces of product information corresponding to the click data of the associated product information;
[0045] At least two of the product feature vectors corresponding to the preset relevance ranking interval are determined as the first candidate product feature vectors for listing.
[0046] The ISU module (Irrelevance Suppression Unit) divides other products selected by users on the page into two categories: related and unrelated. i The correlation can be obtained:
[0047] f i,j =MLP([xc i ,p i,j ,xc i -p i,i ,xc i⊙p i,j ]), where xc i The feature vector of the product browsed by the user, p i,j The feature vector of the kth product on the page, xc i -p i,j , the difference between the feature vectors of the product the user browses and the jth product on the page, xc i ⊙p i,j The element-wise product of the feature vector of the product browsed by the user and the jjth product on the page. The ISU module is based on the relevance score f i,j The top-k most relevant products are retained and their relevance scores are set to positive numbers. The relevance scores of other products are set to negative infinity (-∞), indicating that they are irrelevant to the products browsed by the user.
[0048] In this way, the ISU module can filter out content that is not relevant to the user's interests, thereby providing more personalized and accurate recommendations, increasing user satisfaction and the platform's sales conversion rate. This recommendation system based on user behavior and product relevance can significantly improve the user's shopping experience and the platform's commercial effectiveness.
[0049] The aforementioned preset neural network model is a multilayer perceptron (MLP), which is a basic artificial neural network model with a multilayer structure consisting of multiple neurons. MLP is a feedforward neural network, which is usually used to solve classification and regression problems. The basic structure of MLP includes an input layer, an output layer, and at least one or more hidden layers. Each layer is composed of multiple neurons, and each neuron generates an output by weighted summing the input values and passing through an activation function. In MLP, except for the input layer, neurons in each layer use nonlinear activation functions, which enables MLP to learn and simulate complex nonlinear relationships. The introduction of activation functions is the main difference between MLP and linear models, which allows MLP to capture the complex interactions between input data, and these neurons depend on the value of each input.
[0050] S12, obtaining client interaction data, and using a cross-attention mechanism to associate the interaction data with the first candidate product feature vector;
[0051] It can be understood that the client's interaction data is obtained, a behavioral interest model is constructed, and converted into an interaction data feature vector, and a cross-attention mechanism is used to use the interaction data feature vector as a query sequence and the first candidate product feature vector as a key-value sequence and associate them.
[0052] Specifically, the RIU module (Relevance Interaction Unit) of the CIU module will enhance the user's behavioral interest through other related products on the page. When the ISU module restricts non-related products on the page, other product information can be associated by building a cross-attention mechanism, which can be expressed as: Q i =xc i W Q , where xc i The feature vector of the product browsed by the user, W Q is the weight matrix associated with the query vector, which is used to transform the user interest vector into the query space; K i =p i W k , where p i The feature vector for each relevant product on the page, W k is the bond weight matrix; V i,j =p i W V , where p i The feature vector for each relevant product on the page, W V is the key weight matrix; where Q, K, and V represent query, key, and value, respectively.
[0053] The aforementioned acquisition of the client's interaction data builds a behavioral interest model, which aims to capture and represent the user's interests and preferences. This is usually achieved by analyzing the user's historical interaction data (such as browsing, clicking, purchasing, etc.). The interest model can help the recommendation system understand the product types or topics that the user may be interested in, and the implementation can be:
[0054] The interest model first needs to collect user interaction data, which may include: browsing history, click records, purchase behavior, search queries, user ratings and comments. Feature extraction extracts features from the client's interaction data. These features may include: the user's click frequency on a specific product category; the price range of the products purchased by the user; the user's preference for a specific brand; the time and location of the user's activeness
[0055] Model building, using machine learning or deep learning techniques to build interest models. Common methods include: Collaborative filtering: Recommendations based on user or product similarities. Content-based recommendations: Recommendations based on product features and user preferences. Deep learning models: For example, using neural networks to learn complex patterns of user interests.
[0056] Interest representation, the interest model generates an interest representation for each user, which may be a vector or a set of features to describe the user's interests. This representation can capture the user's preferences for different product features, such as preference for a certain price range, preference for a certain brand or type, and curiosity about new things.
[0057] Model Application,Interest models can be applied to different recommendation scenarios, including:,Personalized recommendations: recommend products that may be of interest to each user.,New user recommendations: provide a preliminary list of recommendations for new users until,enough data is collected to optimize the recommendations.,Dynamic recommendations: dynamically adjust the recommendation list based on the user’s,latest behavior.
[0058] As mentioned above, the workflow of the RIU module is to enhance user interest: The RIU module enhances user interest in smartphones through a cross-attention mechanism. It uses query (Q), key (K), and value (V) to associate Li Hua’s behavior with the information of other related products on the page.
[0059] S13, calculating the interest weights of the interaction data and the first candidate product feature vector, and outputting the corresponding second candidate product feature vector;
[0060] It should be noted that the softmax function is used to calculate the interest weights of the interaction data and the first candidate product feature vector;
[0061] The interest weight is weighted and summed with the first candidate product feature vector to obtain the second candidate product feature vector.
[0062] The behavioral interest constructed by cross-attention can be expressed as:
[0063] where f i is the product information relevance score, D s is the dimension parameter, W O is the output weight matrix, xc i It is the feature vector of the user browsing the product, representing the original user interest vector, which may be constructed based on the user's historical interaction data (such as browsing, clicking, purchasing, etc.). i The query vector is the representation of the user’s interest in the query space. is the representation of product features in the key space, V i It is the representation of product features in the value space. aug: is the abbreviation of “augmented”, which means “enhanced” or “expanded”. The user interest vector xCi is combined with the product feature vector Vi through the cross attention mechanism, while taking into account the relevance score f obtained from the ISU module. i, to generate a user interest vector containing contextual information.
[0064] The purpose of this process is to capture the user's interests in a specific context. For example, if the user is viewing outdoor sports equipment, the system will not only consider the user's general interest in this type of product, but also consider the information of other related products on the page to provide more personalized recommendations. In this way, the recommendation system can more accurately understand the user's needs and preferences in a specific context.
[0065] For example, building cross-attention: the RIU module calculates cross-attention weights and uses these weights to enhance the user’s interest representation on their smartphone.
[0066] The aforementioned cross-attention mechanism: The cross-attention mechanism is used to handle the dependency between two different sequences, especially in multimodal tasks, such as combining the user's behavior sequence and the product's feature sequence. It obtains the product information most relevant to the user's interest by calculating the attention weight between the query sequence (user interest) and the key-value sequence (product feature).
[0067] The above-mentioned step S12 and this step are specifically applied to realize the process:
[0068] 1. Client interaction data collection: Collect user interaction data on the platform, such as browsing history, click records, purchase behavior, etc.
[0069] 2. Interest model construction: Use the client's interaction data to build an interest model. This may involve using machine learning algorithms, such as clustering to analyze user preferences, or deep learning models, such as neural networks, to learn high-dimensional representations of user interests.
[0070] 3. Product feature extraction: For products on the platform, extract relevant features such as category, price, brand, user reviews, etc.
[0071] 4. Application of cross-attention mechanism: Introduce the cross-attention mechanism in the recommendation system, take the user interest model as the query sequence (Q), and the product features as the key value sequence (K,
[0072] V). By calculating the attention weights, the model is able to identify the product features that are most relevant to the user’s interests.
[0073] 5. Recommendation list generation: Generate a recommendation list based on the output of the cross-attention mechanism. Product features with higher weights will make the corresponding products rank higher in the recommendation list.
[0074] 6. Model training and optimization: through actual user feedback (such as clicks, purchases)
[0075] To train and optimize interest models and cross-attention mechanisms to improve recommendation accuracy and user satisfaction.
[0076] 7. Real-time recommendation: When users interact with the system, interest models and cross-attention mechanisms are applied in real time to dynamically generate personalized recommendations.
[0077] By combining the interest model and the cross-attention mechanism, the recommendation system can more accurately capture the user's dynamic interests and provide more personalized recommendation results.
[0078] S14. Use a self-attention mechanism to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list.
[0079] In this embodiment, a self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, and output a behavior interest weight vector;
[0080] Input the behavior interest weight vector to the average pooling layer for aggregation to obtain a product recommendation vector;
[0081] A product recommendation list associated with the product recommendation vector is generated.
[0082] It should be further explained that the self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, and the behavior interest weight vector is output, including:
[0083] Adopting a preset neural network model to perform adaptive alignment processing on the second candidate product feature vector;
[0084] The second candidate product feature vector after adaptive alignment is combined with the behavioral interest weight vector corresponding to the client's interaction data using a self-attention mechanism.
[0085] This embodiment is implemented through an AIAU module. The AIAU module (Adaptive Interest Aggregation Unit) is a structure of an MLP and two hidden layers, which can align and activate the enhanced behavior interest distribution with the result refined by the attention mechanism:
[0086] Among them, xt re The enhanced feature vector of the target product or candidate product, which may have been enhanced by the cross-attention mechanism or other methods to include more contextual information, The user’s enhanced interest vector is generated by combining the user’s historical interaction data and current contextual information (e.g., the product details viewed by the user).
[0087] Aligned interest information After enhancement behavioral interests, and then a self-attention layer to capture the mutual influence:
[0088] The query vector is the representation of the user’s interest in the query space. is the representation of product features in the key space, is the representation of product features in value space, D a is the dimension of the key (K) vector, The dimension is D a , divided by It is a normalization method that aims to prevent the dot product result from being too large and helps stabilize the training process.
[0089] The attention score is converted into a probability distribution through the softmax function, so that the weight of each element is 1. This means that the model assigns a weight to each representation of the user's interest, and the size of the weight indicates the importance of the interest point relative to other interest points.
[0090] Finally, the user's preference results can be expressed as:
[0091] x S =MLP(AvgPool(xc m )), xc m The behavior interest vector output by the self-attention layer. AvgPool is an average pooling layer used to aggregate the interest representation processed by the self-attention layer.
[0092] The MLP (Multilayer Perceptron) mentioned above in this embodiment is a feed-forward artificial neural network. In MLP, data flows from the input layer to the hidden layer, and then to the output layer.
[0093] An MLP can have one or more hidden layers, each consisting of multiple neurons. Each neuron usually uses some form of activation function, such as ReLU, sigmoid, or tanh.
[0094] When an MLP has two hidden layers, its structure can be as follows: Input Layer: This is the first layer of the network and receives the raw data input. First Hidden Layer: The data from the input layer is passed to each neuron in the first hidden layer. Each neuron takes a weighted sum of its inputs and then applies an activation function to introduce nonlinearity, allowing the network to learn complex patterns. Second Hidden Layer: The output of the first hidden layer becomes the input to the second hidden layer. The process is the same as the first hidden layer, with each neuron taking a weighted sum of its inputs and applying an activation function. Output Layer: The output of the second hidden layer is passed to the output layer. The number of neurons in the output layer depends on the task (for example, for classification tasks, the number of neurons in the output layer is usually equal to the number of categories). Loss Function: The output of the network is compared to the true labels using a loss function to calculate the error or loss. Backpropagation: The results of the loss function are used to backpropagate through the network, adjusting the weights to minimize the loss. Optimizer: Optimizers (such as SGD, Adam, etc.) are used to update the network weights to improve the performance of the network.
[0095] In this way, the AIAU module can provide more personalized and accurate recommendations, increase user satisfaction and the platform's sales conversion rate. This recommendation system based on user behavior and interests can significantly improve the user's shopping experience and the platform's commercial effectiveness.
[0096] The data recommendation method provided in this embodiment can obtain user product information click data and convert it into a product feature vector; input the product feature vector into a preset neural network model to calculate the relevance of product information, and extract the first candidate product feature vector within the preset relevance sorting interval; obtain the client's interaction data, and use a cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use a self-attention mechanism to calculate the interaction data of the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, which solves the current technical problem of low product recommendation accuracy due to the explosive growth of information acquisition and the limited ability of users to select information when mining user behavior to mine their intentions and preferences.
[0097] At the same time, the client's interaction data can be distinguished and extracted effective contextual information by adopting preset neural network models, cross-attention mechanisms, and self-attention mechanisms. It not only considers the direct interaction between users and products, but also considers the user's environment, historical behavior patterns, and other factors that may affect user decisions, which helps to improve the effectiveness of the recommendation system and make recommendations more in line with users' actual needs and preferences.
[0098] Furthermore, a preset neural network model is used to adaptively align the feature vector of the second candidate product, which can automatically learn the optimal mapping of input features to output space. This means that interest representations can be adaptively converted without manually designing feature conversion logic, effectively processing user interest representations, and generating more personalized and accurate recommendations by learning the complex interactions between users and products.
[0099] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.
[0100] In one embodiment, a data recommendation device is provided, which corresponds one-to-one to the data recommendation method in the above embodiment. Figure 3 As shown, the data recommendation device includes an acquisition module 101, a first candidate product feature vector module 102, a cross-attention mechanism module 103, a second candidate product feature vector module 104, and a product recommendation module 105. The functional modules are described in detail as follows:
[0101] The acquisition module 101 is used to acquire user product information click data and convert it into a product feature vector;
[0102] In one embodiment, the acquisition module 101 is specifically used for the following steps:
[0103] Obtain user product information click data, and parse out product click sequence and product information data;
[0104] Filter the product information data according to the product click order to obtain related product information click data;
[0105] The related product information click data is converted into the product feature vector.
[0106] The first candidate product feature vector module 102 is used to input the product feature vector into a preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector within a preset relevance sorting interval according to the product information relevance;
[0107] In one embodiment, the first candidate product feature vector module 102 is specifically used for the following steps:
[0108] Input the product feature vector into a preset neural network model, and calculate the relevance of at least two pieces of product information corresponding to the click data of the associated product information;
[0109] At least two of the product feature vectors corresponding to the preset relevance ranking interval are determined as the first candidate product feature vectors for listing.
[0110] A cross-attention mechanism module 103 is used to obtain the interaction data of the client, and associate the interaction data with the first candidate product feature vector using a cross-attention mechanism;
[0111] In one embodiment, the cross attention mechanism module 103 is specifically used for the following steps:
[0112] Acquire the client's interaction data, build a behavior interest model, and convert it into an interaction data feature vector;
[0113] The cross-attention mechanism is adopted to take the interaction data feature vector as a query sequence and the first candidate product feature vector as a key-value sequence, and associate them.
[0114] A second candidate product feature vector module 104, configured to calculate the interest weight of the interaction data and the first candidate product feature vector, and output a corresponding second candidate product feature vector;
[0115] In one embodiment, the second candidate product feature vector module 104 is specifically used for the following steps:
[0116] Calculate the interest weight of the interaction data and the first candidate product feature vector using a softmax function;
[0117] The interest weight is weighted and summed with the first candidate product feature vector to obtain the second candidate product feature vector.
[0118] The product recommendation module 105 is used to calculate the interaction data between the second candidate product feature vector and the client by using a self-attention mechanism, output a behavior interest weight vector, and generate an associated product recommendation list.
[0119] In one embodiment, the product recommendation module 105 specifically includes:
[0120] An output unit, configured to calculate the interaction data between the second candidate product feature vector and the client by using a self-attention mechanism, and output a behavior interest weight vector;
[0121] A product recommendation vector unit, used to input the behavior interest weight vector to the average pooling layer for aggregation to obtain a product recommendation vector;
[0122] A generating unit is used to generate a product recommendation list associated with the product recommendation vector.
[0123] In one embodiment, the output unit is used for the following steps:
[0124] Adopting a preset neural network model to perform adaptive alignment processing on the second candidate product feature vector;
[0125] The second candidate product feature vector after adaptive alignment is combined with the behavioral interest weight vector corresponding to the client's interaction data using a self-attention mechanism.
[0126] The data recommendation device provided in this embodiment can obtain user product information click data and convert it into a product feature vector; input the product feature vector into a preset neural network model to calculate the relevance of product information, and extract the first candidate product feature vector within the preset relevance sorting interval; obtain the client's interaction data, and use a cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use a self-attention mechanism to calculate the interaction data of the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, which solves the current technical problem of low product recommendation accuracy due to the explosive growth of information acquisition and the limited ability of users to select information when mining user behavior to mine their intentions and preferences.
[0127] At the same time, the client's interaction data can be distinguished and extracted effective contextual information by adopting preset neural network models, cross-attention mechanisms, and self-attention mechanisms. It not only considers the direct interaction between users and products, but also considers the user's environment, historical behavior patterns, and other factors that may affect user decisions, which helps to improve the effectiveness of the recommendation system and make recommendations more in line with users' actual needs and preferences.
[0128] Furthermore, a preset neural network model is used to adaptively align the feature vector of the second candidate product, which can automatically learn the optimal mapping of input features to output space. This means that interest representations can be adaptively converted without manually designing feature conversion logic, effectively processing user interest representations, and generating more personalized and accurate recommendations by learning the complex interactions between users and products.
[0129] For the specific definition of the data recommendation device, please refer to the definition of the data recommendation method above, which will not be repeated here. Each module in the above data recommendation device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0130] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a data recommendation method server side.
[0131] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a client side of a data recommendation method.
[0132] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:
[0133] Obtain user product information click data and convert it into product feature vector;
[0134] Input the product feature vector into a preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector within the preset relevance ranking interval;
[0135] Acquire client interaction data, and associate the interaction data with the first candidate product feature vector using a cross-attention mechanism;
[0136] Calculating interest weights of the interaction data and the first candidate product feature vector, and outputting a corresponding second candidate product feature vector;
[0137] A self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list.
[0138] The computer device provided in this embodiment can obtain user product information click data and convert it into a product feature vector; input the product feature vector into a preset neural network model to calculate the relevance of the product information, and extract the first candidate product feature vector within the preset relevance sorting interval; obtain the client's interaction data, and use a cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use a self-attention mechanism to calculate the interaction data of the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, which solves the current technical problem of low product recommendation accuracy due to the explosive growth of information acquisition and the limited ability of users to select information when mining user behavior to mine their intentions and preferences.
[0139] At the same time, the client's interaction data can be distinguished and extracted effective contextual information by adopting preset neural network models, cross-attention mechanisms, and self-attention mechanisms. It not only considers the direct interaction between users and products, but also considers the user's environment, historical behavior patterns, and other factors that may affect user decisions, which helps to improve the effectiveness of the recommendation system and make recommendations more in line with users' actual needs and preferences.
[0140] Furthermore, a preset neural network model is used to adaptively align the feature vector of the second candidate product, which can automatically learn the optimal mapping of input features to output space. This means that interest representations can be adaptively converted without manually designing feature conversion logic, effectively processing user interest representations, and generating more personalized and accurate recommendations by learning the complex interactions between users and products.
[0141] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
[0142] Obtain user product information click data and convert it into product feature vector;
[0143] Input the product feature vector into a preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector within the preset relevance ranking interval;
[0144] Acquire client interaction data, and associate the interaction data with the first candidate product feature vector using a cross-attention mechanism;
[0145] Calculating interest weights of the interaction data and the first candidate product feature vector, and outputting a corresponding second candidate product feature vector;
[0146] A self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list.
[0147] The computer-readable storage medium provided in this embodiment can obtain user product information click data and convert it into a product feature vector; input the product feature vector into a preset neural network model to calculate the relevance of product information, and extract the first candidate product feature vector within the preset relevance sorting interval; obtain the client's interaction data, and use a cross-attention mechanism to associate the interaction data with the first candidate product feature vector; calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; use a self-attention mechanism to calculate the interaction data of the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, which solves the current technical problem of low product recommendation accuracy due to the explosive growth of information acquisition and the limited ability of users to select information when mining user behavior to mine their intentions and preferences.
[0148] At the same time, the client's interaction data can be distinguished and extracted effective contextual information by adopting preset neural network models, cross-attention mechanisms, and self-attention mechanisms. It not only considers the direct interaction between users and products, but also considers the user's environment, historical behavior patterns, and other factors that may affect user decisions, which helps to improve the effectiveness of the recommendation system and make recommendations more in line with users' actual needs and preferences.
[0149] Furthermore, a preset neural network model is used to adaptively align the feature vector of the second candidate product, which can automatically learn the optimal mapping of input features to output space. This means that interest representations can be adaptively converted without manually designing feature conversion logic, effectively processing user interest representations, and generating more personalized and accurate recommendations by learning the complex interactions between users and products.
[0150] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.
[0151] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0152] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0153] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.
[0154] It should be noted that if software tools or components other than those of our company appear in the embodiments of the present application, they are only used for illustrative purposes and do not represent actual use.
Claims
1. A data recommendation method, characterized in that: include: Obtain user product information click data and convert it into product feature vector; Inputting the product feature vector into a preset neural network model to calculate product information relevance, and extracting a first candidate product feature vector within a preset relevance sorting interval according to the product information relevance; Acquire client interaction data, and associate the interaction data with the first candidate product feature vector using a cross-attention mechanism; Calculating interest weights of the interaction data and the first candidate product feature vector, and outputting a corresponding second candidate product feature vector; A self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list.
2. The data recommendation method according to claim 1, characterized in that: The step of obtaining user product information click data and converting it into a product feature vector includes: Obtain user product information click data, and parse out product click sequence and product information data; Filter the product information data according to the product click order to obtain related product information click data; The related product information click data is converted into the product feature vector.
3. The data recommendation method according to claim 1, characterized in that: The product feature vector is input into a preset neural network model to calculate the product information relevance, and the first candidate product feature vector within the preset relevance ranking interval is extracted, including: Input the product feature vector into a preset neural network model, and calculate the relevance of at least two pieces of product information corresponding to the click data of the associated product information; At least two of the product feature vectors corresponding to the preset relevance ranking interval are determined as the first candidate product feature vectors for listing.
4. The data recommendation method according to claim 1, characterized in that: Acquiring client interaction data, and using a cross-attention mechanism to associate the interaction data with the first candidate product feature vector, including: Acquire the client's interaction data, build a behavior interest model, and convert it into an interaction data feature vector; The cross-attention mechanism is adopted to take the interaction data feature vector as a query sequence and the first candidate product feature vector as a key-value sequence, and associate them.
5. The data recommendation method according to claim 1, characterized in that: Calculating the interest weights of the interaction data and the first candidate product feature vector, and outputting a second candidate product feature vector within a preset interest weight sorting interval, including: Calculating the interest weights of the interaction data and the first candidate product feature vector using a softmax function; The interest weight is weighted and summed with the first candidate product feature vector to obtain the second candidate product feature vector.
6. The data recommendation method according to claim 1, characterized in that: The self-attention mechanism is used to calculate the interaction data between the second candidate product feature vector and the client, output a behavior interest weight vector, and generate an associated product recommendation list, including: Using a self-attention mechanism to calculate the interaction data between the second candidate product feature vector and the client, and outputting a behavior interest weight vector; Input the behavior interest weight vector to the average pooling layer for aggregation to obtain a product recommendation vector; A product recommendation list associated with the product recommendation vector is generated.
7. The data recommendation method according to claim 6, characterized in that: The interaction data between the second candidate product feature vector and the client is calculated using a self-attention mechanism, and a behavior interest weight vector is output, including: Adopting a preset neural network model to perform adaptive alignment processing on the second candidate product feature vector; The second candidate product feature vector after adaptive alignment is combined with the behavioral interest weight vector corresponding to the client's interaction data using a self-attention mechanism.
8. A data recommendation device, characterized in that: include: The acquisition module is used to obtain the user's product information click data and convert it into a product feature vector; A first candidate product feature vector module, used to input the product feature vector into a preset neural network model to calculate the product information relevance, and extract the first candidate product feature vector within a preset relevance ranking interval; A cross-attention mechanism module, used for obtaining client interaction data, and associating the interaction data with the first candidate product feature vector using a cross-attention mechanism; A second candidate product feature vector module, used to calculate the interest weight of the interaction data and the first candidate product feature vector, and output the corresponding second candidate product feature vector; The product recommendation module is used to calculate the interaction data between the second candidate product feature vector and the client by using a self-attention mechanism, output a behavior interest weight vector, and generate an associated product recommendation list.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the data recommendation method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the data recommendation method according to any one of claims 1 to 7 are implemented.