Personalized recommendation method and device based on high-order feature interaction and medium

Abstract

The invention discloses a personalized recommendation method and device based on high-order feature interaction and a medium, and relates to the field of deep learning, and the method comprises the steps: inputting obtained user information and knowledge in a knowledge base into a pre-trained personalized recommendation model, and outputting a classification variable used for representing whether a user obtains the knowledge or not; and performing personalized knowledge recommendation on the user according to the classification variables. The personalized recommendation model is a deep learning-based model, and compared with a traditional shallow model (such as a factorization machine, a domain factorization machine and the like), the personalized recommendation model can effectively learn high-interaction low-order features by using the domain factorization machine, and can construct high-order features in combination with a cross network module and a deep learning module.

Description

Technical Field

[0001] The present application relates to the field of deep learning, and specifically to a personalized recommendation method, device, and medium based on high-order feature interaction. Background Art

[0002] With the rapid development of the Internet and big data technologies, recommendation systems have made progress in intelligence and personalization. However, in traditional solutions, recommendation systems mainly rely on simple statistics of rating data and user behavior, which makes it difficult to meet users' higher demands for personalization, real-time and diversity.

[0003] The application of deep learning and data mining technologies has further promoted the intelligent development of recommendation systems. However, existing technologies still face the following problems that need to be solved: data sparsity, dynamic changes in user interests, insufficient content attributes, and poor model interpretability. Summary of the invention

[0004] In order to solve the above problems, this application proposes a personalized recommendation method based on high-order feature interaction, including: Input the acquired user information and the knowledge in the knowledge base into a pre-trained personalized recommendation model, output a classification variable indicating whether the user has acquired the knowledge, and make personalized recommendations of knowledge to the user based on the classification variable; The personalized recommendation model includes an input layer, an embedding layer, a feature interaction layer, and a combined output layer; The input layer inputs the user information and the knowledge in the form of feature vectors; The embedding layer converts the high-dimensional sparse feature vector into a low-dimensional dense embedding vector; The feature interaction layer includes a deep neural network module, a cross network module, a domain factorization machine module, and an attention mechanism module; The deep neural network module and the cross network module construct high-order feature interactions through the cross network and the deep neural network; The domain factorization machine module defines a feature domain through a domain factorization machine, groups features, and constructs linear relationships of features and second-order feature interaction relationships; The attention mechanism module learns the weights of second-order feature interactions through an attention network; The combined output layer makes predictions based on the output vectors of the deep neural network module, the cross network module, and the attention mechanism module.

[0005] In one example, the personalized recommendation model includes the following steps in the model training process: The input layer obtains a data set and converts the information features in the data set into a feature vector through encoding; The information features include a feature column and a label column. The feature column includes user information and multi-dimensional features corresponding to knowledge. The label column includes whether the user has acquired the knowledge. The feature types of the multi-dimensional features include categorical features and continuous features, which are processed by encoding respectively.

[0006] In one example, the embedding layer converts the feature vector of each feature domain into an embedding vector corresponding to the feature domain through the embedding matrix corresponding to the feature domain; The embedding vectors corresponding to the feature fields are combined to obtain the output vector of the embedding layer.

[0007] In one example, the deep neural network module includes a Product layer, a hidden layer, and an output layer; The Product layer includes a linear part and a nonlinear part; The linear part outputs the corresponding linear part output result through the corresponding weight matrix and the linear signal vector; wherein the linear signal vector is obtained through the corresponding embedding vector; The nonlinear part outputs the corresponding nonlinear part output result through the corresponding weight matrix and the secondary signal vector; wherein the linear signal vector is obtained through multiple embedding vectors.

[0008] In one example, the cross network module includes a plurality of cross layers; Each cross layer performs cross calculations with the feature vector of the current layer input and the output vector of the previous layer, and adjusts the contribution of each layer in the cross calculation through the weight vector of the current layer, and obtains the output of the current layer through the adjusted result, the original input vector and the bias term.

[0009] In one example, the domain factorization machine module defines that each feature domain corresponds to a single-class feature, and each feature domain includes multiple features; For each feature, multiple latent vectors are assigned to the feature; For each feature in the feature interaction, the latent vector of the feature in the feature domain of the other feature is calculated, and based on the calculation results of all feature interactions, as well as the feature weights and global bias terms corresponding to all traits, the corresponding linear relationship and second-order feature interaction relationship are output.

[0010] In one example, the attention mechanism module processes the weight matrix corresponding to the second-order feature interaction, the calculation result of the latent vector, and the corresponding bias term through an activation function, obtains the corresponding attention score based on the processing result and the attention network parameters, and normalizes the attention score.

[0011] In one example, the combined output layer concatenates the output vectors of the deep neural network module, the cross network module, the attention mechanism module, and the bias term, and predicts the concatenated vector through an activation function.

[0012] On the other hand, the present application also proposes a personalized recommendation device based on high-order feature interaction, including: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: provide a personalized recommendation method based on high-order feature interaction as described in any of the above examples.

[0013] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer executable instructions, characterized in that the computer executable instructions are set to: a personalized recommendation method based on high-order feature interaction as described in any of the above examples.

[0014] The personalized recommendation method based on high-order feature interaction proposed in this application can bring the following beneficial effects: The personalized recommendation model in this application is a deep learning-based model. Compared with traditional shallow models (such as factor decomposition machines, domain factor decomposition machines, etc.), it can effectively learn low-order features with strong interactions by using domain factor decomposition machines, and can combine cross-network modules and deep learning modules to construct high-order features.

[0015] Compared with traditional models such as DeepFM, which only implicitly process feature interactions through deep neural networks, the parallel architecture of the personalized recommendation model in this application enables it to explicitly construct high-order feature interactions of arbitrary orders through cross networks.

[0016] The personalized recommendation model introduces the Product layer before the hidden layer of the deep learning module to capture the nonlinear relationship between features, further enhancing the ability of high-order feature interaction.

[0017] Compared with traditional models such as xDeepFM and EDCN, although they also combine explicit and implicit high-order feature construction, they are not able to use shallow modules to construct second-order feature interactions like the personalized recommendation model in this application, and they do not pay attention to the weights of each second-order feature interaction, which makes the traditional model lose the ability to capture low-order key information.

[0018] The personalized recommendation model can effectively capture the complex relationship between users and knowledge, thereby effectively improving the pertinence and accuracy of knowledge recommendations for users. It can also solve problems such as data sparsity, dynamic changes in user interests, insufficient content attributes, and poor model interpretability in traditional solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a personalized recommendation method based on high-order feature interaction in an embodiment of the present application; Figure 2 This is a schematic diagram of the architecture of a personalized recommendation model in one scenario in an embodiment of the present application; Figure 3 This is a schematic diagram of the conversion process of the embedded layer in one scenario in an embodiment of the present application; Figure 4 This is a schematic diagram of the structure of a deep neural network module in one scenario in an embodiment of the present application; Figure 5 A visualization diagram of calculation of a cross-network module in one scenario in an embodiment of the present application; Figure 6 Schematic diagram of a personalized recommendation device based on high-order feature interaction in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0021] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, the embodiment of the present application provides a personalized recommendation method based on high-order feature interaction, including: S101: Input the acquired user information and the knowledge in the knowledge base into a pre-trained personalized recommendation model, and output a classification variable for indicating whether the user has acquired the knowledge.

[0023] Personalized recommendations can be initiated actively or passively. For example, after a user asks a question on a search platform, personalized recommendations are passively made based on the user's needs. Or, on an e-commerce platform, personalized recommendations are actively made based on the user's purchase history.

[0024] User information and knowledge in the knowledge base are pre-set and obtained. Based on different actual scenarios, knowledge and knowledge base can also be set to corresponding types. For example, knowledge can include: web page address, product knowledge, skill knowledge, business knowledge, etc.

[0025] When personalized recommendations are needed for users, the corresponding knowledge base is selected based on the current recommendation scenario (for example, including recommended products, recommended books, recommended businesses, etc.), and the classification variables output by the personalized recommendation model indicate whether the user has acquired the corresponding knowledge.

[0026] The categorical variable output by the personalized recommendation model (Higher-Order Feature Interaction Model, HFIM) is a binary variable 0 or 1, indicating whether the user has acquired the knowledge.

[0027] S102: Performing personalized knowledge recommendations for users based on the classification variables.

[0028] For different recommendation scenarios, the final personalized recommendation results may also be different. For example, for user search questions, you can give priority to recommending knowledge that the user has not yet acquired. For product recommendations, you can recommend the same or similar products based on the products that the user has already purchased (corresponding to the knowledge that has been acquired). For book recommendations, you can recommend similar or unread books based on the books that the user has already read (corresponding to the knowledge that has been acquired). For business recommendations, you can recommend the next business to be executed based on the business that the user has already executed (corresponding to the knowledge that has been acquired) and referring to the business execution process.

[0029] The HIFM model uses a domain factorization machine to realize the linear relationship between features and the second-order feature interaction, and introduces an attention network to learn the weight information of the second-order interaction features; it explicitly constructs high-order feature interactions through a cross network, and implicitly uses a deep neural network to construct high-order feature interactions; finally, the outputs of the deep neural network, the cross network, and the domain factorization machine after the attention network are spliced ​​as the final prediction result of the model.

[0030] like Figure 2 As shown in the figure, the personalized recommendation model includes an input layer, an embedding layer, a feature interaction layer, and a combined output layer. The input layer inputs the processed user information and knowledge in the data set in the form of a feature vector. The embedding layer converts the high-dimensional sparse feature vector into a low-dimensional dense embedding vector. The feature interaction layer includes a deep neural network module, a cross network module, a domain factor decomposition machine module, and an attention mechanism module; the deep neural network module and the cross network module construct high-order feature interactions through cross networks and deep neural networks; the domain factor decomposition machine module defines the feature domain through a domain factor decomposition machine, groups features, and constructs linear relationships and second-order feature interaction relationships of features; the attention mechanism module learns the weights of second-order feature interactions through an attention network. The combined output layer makes predictions based on the output vectors of the deep neural network module, the cross network module, and the attention mechanism module.

[0031] The input layer is the first step of the HIFM model. Its purpose is to input the information features of users and knowledge into the model in the form of vectors. During the model training process of the personalized recommendation model, the original data set is set as the training sample (after the training is completed, the data set corresponds to the actual data content that needs to be predicted).

[0032] During the training process, a data set is obtained and the information features in the data set are converted into feature vectors through encoding.

[0033] Among them, information features include feature columns and label columns. The feature columns include user information and multi-dimensional features corresponding to knowledge, such as user gender, age, hobbies, knowledge type, etc. The label column includes whether the user has acquired the knowledge (for example, whether the user has acquired the knowledge is determined by whether the user clicks on the knowledge on the web page or platform).

[0034] The feature types of multi-dimensional features include categorical features and continuous features, which are processed by encoding. For categorical features, the feature processing method is to convert the feature data into a One-Hot encoding form. For example, the gender feature is encoded as 10 or 01. If the first digit is 1, it represents a male, and if the second digit is 1, it represents a female. For continuous features, the discretization method is used to process continuous feature data.

[0035] The embedding layer converts the feature vector of each feature domain into the embedding vector corresponding to the feature domain through the embedding matrix corresponding to the feature domain. The embedding vectors corresponding to each feature domain are combined to obtain the output vector of the embedding layer.

[0036] Although One-Hot encoding standardizes the model input to a certain extent, it will cause a sharp increase in data dimensions when processing features with many categories such as knowledge types, generating high-dimensional sparse data. In this case, the effective expression of features may be affected. If calculated directly, the model calculation amount will be very large, causing loss of model performance. Therefore, the HFIM model introduces an embedding layer after the input layer to convert high-dimensional sparse features into low-dimensional dense vectors. The conversion of the embedding process is shown in Formula 1: Formula 1; in, Indicates The feature vector of the feature domain, Indicates The embedding matrix of the feature domain, Indicates The embedding vector of the feature domain.

[0037] like Figure 3 As shown, feature vectors of different dimensions , after being processed by the embedding layer, a set of embedding vectors of the same dimension can be obtained During the conversion process, each embedding matrix Each parameter in will first be initialized to a floating point value, and finally the optimal parameter result will be learned through the back propagation of the model. The output vector E of the embedding layer is defined as follows: Formula 2: in, represents the output vector of the embedding layer, Indicates The embedding vector of the feature domain, represents the total number of feature domains, Indicates the dimension size of the embedding vector.

[0038] like Figure 4 As shown in the figure, the deep neural network module includes a product layer, a hidden layer, and an output layer. HIFM implicitly constructs a high-order feature combination through a deep neural network. It is essentially a feedforward neural network, but it is difficult to fully capture the relationship between features by only learning high-order features in a fully connected way. Therefore, the Product layer is added before the first hidden layer to further capture the nonlinear relationship between features and realize the intersection between features.

[0039] The Product layer includes a linear part and a nonlinear part. The linear part outputs the corresponding linear part output result through the corresponding weight matrix and the linear signal vector; wherein the linear signal vector is obtained through the corresponding embedding vector; the nonlinear part outputs the corresponding nonlinear part output result through the corresponding weight matrix and the quadratic signal vector; wherein the linear signal vector is obtained through multiple embedding vectors.

[0040] The Product layer focuses on the original information of the embedded vector and the interactive information of each embedded vector, making the information richer and thus improving the model effect. and the nonlinear part The specific calculation formulas are shown in Formula 3 and Formula 4: Formula 3; Formula 4; in, represents a linear signal vector, represents the quadratic signal vector, represents the Hadamard product operation, represents the embedding vector, express and The inner product of vectors, , represents the weight matrix.

[0041] like Figure 4 As shown in the figure, the left side of the Product layer directly copies the embedding vector and introduces a constant signal "1" in the embedding layer to multiply it. The right side performs inner product cross calculations between the embedding vectors. After that, the fully connected layer converts the dimensions of the two parts to the same dimension and inputs them into the hidden layer behind. By introducing the Product layer, the interaction between features is further strengthened.

[0042] The cross network module includes multiple cross layers; each cross layer performs cross calculations with the output vector of the previous layer through the feature vector input by the current layer, and adjusts the contribution of each layer in the cross calculation through the weight vector of the current layer, and obtains the output of the current layer through the adjusted result, the original input vector and the bias term.

[0043] The intersection between features can provide richer information than a single feature. Although deep neural networks can implicitly construct high-order features through multiple layers of nonlinear transformations, they are not interpretable enough and are inefficient. Therefore, feature intersection is applied at each layer through the cross network to explicitly construct high-order feature combinations. The cross network is stacked with multiple cross layers, and each layer captures higher-order intersections of input features. Specifically, each layer calculates the outer product of the input feature vector and its own output based on the previous layer, and then adjusts the contribution of these intersections by dot multiplication with the weight vector of the layer. Finally, the result is added to the original input vector and a bias term to form the output of the layer. The calculation formula is shown in Formula 5: Formula 5; in, and Respectively represent the cross network Layer and The cross-layer output of the layer, Indicates The learning parameters of the layer, Indicates The offset of the layer. Figure 5 As shown, based on the residual mechanism, the output of each cross layer needs to be added to the output of one layer, for example, The output of the layer includes the constructed cross features and Output of the layer .

[0044] The above is the macro calculation method of the cross network. The following is a detailed description of the calculation process of each layer through the specific calculation of the first layer. is the embedding vector output by the embedding layer, and the embedding dimension is , the bias term is 0, and the output of the first cross layer is shown in Formula 6: Formula 6; From formula 6, we can see that in the first layer, first Transpose The outer product operation is performed to form a matrix in which each element is the product of two elements in the embedding vector. The matrix is ​​then combined with the parameter vector Finally, based on the residual mechanism, the second-order cross term is combined with the output of the previous layer By repeating this process, multiple cross layers can be stacked to effectively construct high-order features. In addition, the time and space complexity of the cross network is linear in the input dimension, so the complexity brought by the introduction of the cross network is negligible compared to the deep neural network module.

[0045] The domain factor decomposition machine module defines the features of a single class corresponding to each feature domain, and each feature domain includes multiple features; for each feature, multiple latent vectors are assigned to the feature; for each feature in the feature interaction, the latent vector of the feature in the feature domain where the other feature is located is calculated, and according to the calculation results of all feature interactions, as well as the feature weights and global bias items corresponding to all traits, the corresponding linear relationship and second-order feature interaction relationship are output.

[0046] Field-aware Factorization Machines (FFM) is an improvement on Factorization Machines (FM) that aims to improve the accuracy of prediction models when dealing with complex interactions between features. FFM introduces the concept of domains to group features, which fully improves the interaction between features. , each feature of knowledge can be called a feature domain, and each feature domain contains multiple features. For example, the knowledge type can be used as a feature domain, and the technology type as a knowledge type can be considered as a feature. In the FM model, each feature uses the same latent vector when interacting with other features, while the FFM model assigns multiple latent vectors to each feature, so that features from different domains can respond using different latent vectors when interacting. The expression of FFM is shown in Formula 7: Formula 7; in, represents the global bias term, represents the total number of features, Indicates The weight of the feature, Indicates The vector representation of the features, and Respectively represent characteristics and Features The hidden vector in the opponent domain, represents the dot product of the latent vector. The difference from the FM model is that the latent vector is Became , that is, each feature changes from the original unique latent vector to a group of latent vectors, which enables the FFM model to capture the interaction between features more carefully.

[0047] The attention mechanism module processes the weight matrix corresponding to the second-order feature interaction, the calculation results of the latent vector, and the corresponding bias terms through the activation function, obtains the corresponding attention score based on the processing results and the attention network parameters, and normalizes the attention score.

[0048] Generally speaking, not all feature interactions play a positive role in the performance of the model, and some unimportant feature interactions may even affect the performance of the model. Therefore, the HIF-CSR model introduces an attention network to assign different weights to different second-order feature interactions, so that the model can automatically learn the importance of feature interactions. The weights are calculated through an attention network, and the calculation process is shown in Formula 8 and Formula 9: Formula 8; Formula 9; in, are the attention network parameters, is the activation function, The weight matrix representing the feature interactions, represents the latent vector dot product, represents the bias term, Score for attention, Represents the normalized attention score.

[0049] The combined output layer concatenates the output vectors of the deep neural network module, the cross network module, the attention mechanism module, and the bias term, and predicts the concatenated vector through the activation function.

[0050] The combined output layer concatenates the output of the deep neural network module, the output of the cross network module, and the output of the domain factor decomposition machine after the attention network module, and predicts whether the user has accessed and browsed the knowledge through the activation function. The deep neural network captures complex patterns by learning nonlinear combinations of input features, while the cross network focuses on learning cross combinations between features, which is very valuable for understanding the interactions between features. The domain factor decomposition machine further enhances the model's ability to handle interactions between different feature domains, allowing the model to learn interactions between features at a finer granularity and learn the weights of different interactive features. Information is merged by connecting the output vectors of each layer end to end to form a more comprehensive feature representation. This fusion strategy not only enhances the model's ability to capture complex patterns in the data, but also provides a rich information basis for the final calculation output, thereby improving the prediction accuracy and generalization ability of the entire personalized recommendation model.

[0051] The combined output layer uses The activation function calculates the concatenated output vectors to obtain the final output of the model. The output of the final model is shown in Formula 10: Formula 10; in, and Represent the outputs of the deep neural network and the cross network respectively, represents the output of the domain factorization machine after the attention network, Represents the bias term.

[0052] The personalized recommendation model in this application is a deep learning-based model. Compared with traditional shallow models (such as factor decomposition machines, domain factor decomposition machines, etc.), it can effectively learn low-order features with strong interactions by using domain factor decomposition machines, and can combine cross-network modules and deep learning modules to construct high-order features.

[0053] Compared with traditional models such as DeepFM, which only implicitly process feature interactions through deep neural networks, the parallel architecture of the personalized recommendation model in this application enables it to explicitly construct high-order feature interactions of arbitrary orders through cross networks.

[0054] The personalized recommendation model introduces the Product layer before the hidden layer of the deep learning module to capture the nonlinear relationship between features, further enhancing the ability of high-order feature interaction.

[0055] Compared with traditional models such as xDeepFM and EDCN, although they also combine explicit and implicit high-order feature construction, they are not able to use shallow modules to construct second-order feature interactions like the personalized recommendation model in this application, and they do not pay attention to the weights of each second-order feature interaction, which makes the traditional model lose the ability to capture low-order key information.

[0056] The personalized recommendation model can effectively capture the complex relationship between users and knowledge, thereby effectively improving the pertinence and accuracy of knowledge recommendations for users. It can also solve problems such as data sparsity, dynamic changes in user interests, insufficient content attributes, and poor model interpretability in traditional solutions.

[0057] like Figure 6 As shown, the embodiment of the present application also provides a personalized recommendation device based on high-order feature interaction, including: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: provide the personalized recommendation method based on high-order feature interaction as described in any of the above embodiments.

[0058] An embodiment of the present application further provides a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to be: a personalized recommendation method based on high-order feature interaction as described in any of the above embodiments.

[0059] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0060] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0061] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A personalized recommendation method based on high-order feature interaction, characterized in that: include: Input the acquired user information and the knowledge in the knowledge base into the pre-trained personalized recommendation model, and output a classification variable indicating whether the user has acquired the knowledge; Providing personalized knowledge recommendations to users based on the classification variables; The personalized recommendation model includes an input layer, an embedding layer, a feature interaction layer, and a combined output layer; The input layer inputs the user information and the knowledge in the form of feature vectors; The embedding layer converts the high-dimensional sparse feature vector into a low-dimensional dense embedding vector; The feature interaction layer includes a deep neural network module, a cross network module, a domain factorization machine module, and an attention mechanism module; The deep neural network module and the cross network module construct high-order feature interactions through the cross network and the deep neural network; The domain factorization machine module defines a feature domain through a domain factorization machine, groups features, and constructs linear relationships of features and second-order feature interaction relationships; The attention mechanism module learns the weights of second-order feature interactions through an attention network; The combined output layer makes predictions based on the output vectors of the deep neural network module, the cross network module, and the attention mechanism module.

2. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The personalized recommendation model includes the following steps in the model training process: The input layer obtains a data set and converts the information features in the data set into a feature vector through encoding; The information features include a feature column and a label column. The feature column includes user information and multi-dimensional features corresponding to knowledge. The label column includes whether the user has acquired the knowledge. The feature types of the multi-dimensional features include categorical features and continuous features, which are processed by encoding respectively.

3. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The embedding layer converts the feature vector of each feature domain into an embedding vector corresponding to the feature domain through the embedding matrix corresponding to the feature domain; The embedding vectors corresponding to the feature fields are combined to obtain the output vector of the embedding layer.

4. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The deep neural network module includes a Product layer, a hidden layer, and an output layer; The Product layer includes a linear part and a nonlinear part; The linear part outputs the corresponding linear part output result through the corresponding weight matrix and the linear signal vector; wherein the linear signal vector is obtained through the corresponding embedding vector; The nonlinear part outputs the corresponding nonlinear part output result through the corresponding weight matrix and the secondary signal vector; wherein the linear signal vector is obtained through multiple embedding vectors.

5. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The cross network module includes a plurality of cross layers; Each cross layer performs cross calculations with the feature vector of the current layer input and the output vector of the previous layer, and adjusts the contribution of each layer in the cross calculation through the weight vector of the current layer, and obtains the output of the current layer through the adjusted result, the original input vector and the bias term.

6. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The domain factorization machine module defines the features of each feature domain corresponding to a single class, and each feature domain includes multiple features; For each feature, multiple latent vectors are assigned to the feature; For each feature in the feature interaction, the latent vector of the feature in the feature domain of the other feature is calculated, and based on the calculation results of all feature interactions, as well as the feature weights and global bias terms corresponding to all traits, the corresponding linear relationship and second-order feature interaction relationship are output.

7. The personalized recommendation method based on high-order feature interaction according to claim 6, characterized in that: The attention mechanism module processes the weight matrix corresponding to the second-order feature interaction, the calculation result of the latent vector, and the corresponding bias term through an activation function, obtains the corresponding attention score according to the processing result and the attention network parameter, and normalizes the attention score.

8. The personalized recommendation method based on high-order feature interaction according to claim 1, characterized in that: The combined output layer concatenates the output vectors of the deep neural network module, the cross network module, the attention mechanism module, and the bias term, and predicts the concatenated vector through an activation function.

9. A personalized recommendation device based on high-order feature interaction, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: provide a personalized recommendation method based on high-order feature interaction as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured as: a personalized recommendation method based on high-order feature interaction as described in any one of claims 1 to 8 above.

Images