Product recommendation method and system based on privacy computing, and electronic device and medium

Abstract

Disclosed in the embodiments of the present description are a product recommendation method and system based on privacy computing, and an electronic device and a medium. The product recommendation method based on privacy computing comprises: acquiring a product list candidate set; on the basis of a privacy computing network established by several participating nodes, acquiring linear combination data of the several participating nodes for products in the product list candidate set; and sorting the products on the basis of the linear combination data of the several participating nodes for the products in the product list candidate set, and recommending a sorting result to a user. By means of the embodiments of the present description, the accuracy of product recommendation is improved.

Description

Product recommendation method, system, electronic device and medium based on privacy computing Technical Field

[0001] One or more embodiments of this specification relate to the field of information technology, and specifically to product recommendation methods, systems, electronic devices, and media based on privacy computing. Background Art

[0002] In the field of information technology, product sales platforms recommend suitable products to users based on the characteristics of the users and the products. However, in traditional recommendation algorithms, product sales platforms can only use the personal information authorized by users and the public data transmitted to the product sales platform by product manufacturers to train models and then recommend products to users. This method has relatively poor recommendation results. In addition, product manufacturers have a lot of undisclosed internal information, which raises issues with the use of private data: product manufacturers cannot provide internal non-public information to product sales platforms, and product sales platforms cannot share users' personal information with product manufacturers. Therefore, how to better recommend products without sharing the private data of all parties has become a technical problem that needs to be solved in this field.

[0003] Summary of the Invention

[0004] The embodiments of this specification provide a product recommendation method, system, electronic device, and medium based on privacy computing, and the technical solutions are as follows.

[0005] In the first aspect, an embodiment of this specification provides a product recommendation method based on privacy computing, including: obtaining a candidate set of a product list; based on a privacy computing network established by several participating nodes, obtaining linear combination data of several participating nodes on each product in the candidate set of the product list; sorting each product according to the linear combination data of several participating nodes on each product in the candidate set of the product list, and recommending the sorting results to the user.

[0006] On the second aspect, an embodiment of this specification provides a product recommendation system based on privacy computing, including: a list acquisition module for obtaining a candidate set of product lists; a privacy computing module for obtaining linear combination data of each product in the candidate set of product lists by several participating nodes based on a privacy computing network established by several participating nodes; a sorting module for sorting each product according to the linear combination data of each product in the candidate set of product lists by several participating nodes, and recommending the sorting results to users.

[0007] In the third aspect, an embodiment of this specification provides an electronic device, including a processor and a memory; the processor is connected to the memory; the memory is used to store executable program code; the processor runs the program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the steps of the product recommendation method based on privacy computing in the first aspect of the above embodiment.

[0008] In a fourth aspect, an embodiment of this specification provides a computer storage medium, which stores multiple instructions, and the instructions are suitable for being loaded by a processor and executing the steps of the product recommendation method based on privacy computing in the first aspect of the above embodiment.

[0009] The technical solutions provided in some embodiments of this specification provide at least the following beneficial effects: Based on a privacy-preserving computing network established by several participating nodes, the linear combination data of each product in a candidate product list is obtained by the participating nodes; the products are then sorted based on the linear combination data of each product in the candidate product list by the participating nodes, and the sorted results are recommended to the user. The embodiments of this specification, based on a privacy-preserving computing network established by several participating nodes, can protect the confidentiality of the data of each participant while ensuring that the data is not leaked or misused during the computation process, protect the private data of users and each participant from being shared, and improve the accuracy of product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of this specification, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0011] Figure 1 is a schematic diagram of an application scenario of a product recommendation system based on privacy computing provided in this specification.

[0012] FIG2 is a flowchart of a product recommendation method based on privacy computing provided in this specification.

[0013] FIG3 is a flowchart of the training process of the product recommendation model provided in this specification.

[0014] FIG4 is a schematic diagram of the process of sorting products provided in this specification.

[0015] FIG5 is a flowchart of another product recommendation method based on privacy computing provided in this specification.

[0016] FIG6 is a flow chart of the training process of the linear regression model provided in this specification.

[0017] FIG7 is a schematic diagram of the structure of a product recommendation device based on privacy computing provided in this specification.

[0018] FIG8 is a schematic structural diagram of an electronic device provided in this specification. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of this specification will be described clearly and completely below in conjunction with the drawings in the embodiments of this specification.

[0020] Throughout this specification, the claims, and the accompanying drawings, the terms "first," "second," and so forth are used to distinguish between different items, not to describe a particular order. Furthermore, the term "comprises" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may include other steps or elements inherent to the process, method, product, or apparatus.

[0021] The product recommendation method based on privacy computing provided in multiple embodiments of this specification may be executed by the product recommendation device based on privacy computing provided in an embodiment of the present invention, or a server integrated with the product recommendation device based on privacy computing, wherein the product recommendation device based on privacy computing may be implemented in hardware or software.

[0022] Before describing the technical solution of the present invention, a brief explanation of related technical terms is given first.

[0023] Privacy-preserving computing: Privacy-preserving computing is a computing method that uses cryptography or other technical means to protect data privacy and security. Its goal is to safeguard the confidentiality, integrity, and availability of data while ensuring that data is not leaked or misused during computation. Privacy-preserving computing techniques can be used to provide a trusted environment for processing or analyzing sensitive data, perform processing and analysis in a distributed manner, and transform data and algorithms before processing or analysis.

[0024] Linear regression model: A statistical model used to predict a numerical target variable. It establishes a linear relationship between the independent and dependent variables. The goal of a linear regression model is to find the optimal regression coefficients that minimize the error between the model's predicted values ​​and the actual observed values.

[0025] Federated Learning: Federated learning is a distributed machine learning technology or framework. It aims to enable collaborative modeling and improve model effectiveness while ensuring data privacy, security, and legal compliance. Federated learning emphasizes protecting the privacy of data owners during model training and is an effective measure to address data privacy.

[0026] Neural network model: A model composed of a large number of artificial neurons (nodes). It typically includes components such as an input layer, multiple hidden layers, and an output layer. In a neural network, information is transmitted through the network layers. Each neuron receives input from the neurons in the previous layer, adds a weighted sum to the input, processes it through an activation function, and then passes it to the neurons in the next layer. Neural networks use training data to adjust connection weights so that the network can effectively classify or predict input data. Common neural network models include multilayer perceptrons, convolutional neural networks, and recurrent neural networks. These models have a wide range of applications in fields such as computer vision, natural language processing, and speech recognition.

[0027] Homomorphic encryption: A method that encrypts data into a hard-to-decipher digital string, allowing mathematical manipulation of the encrypted string and the decrypted result. Homomorphic encryption is an encryption technique that allows certain computational operations to be performed while the data is encrypted, without decrypting it. Homomorphic encryption allows addition and multiplication of two or more encrypted data to be performed while the data is encrypted, and the correct result is obtained after decryption. There are two main types of homomorphic encryption: partially homomorphic encryption and fully homomorphic encryption. Partially homomorphic encryption allows either addition or multiplication to be performed on encrypted data, but not both simultaneously. Fully homomorphic encryption allows both addition and multiplication to be performed simultaneously. Homomorphic encryption protects data privacy while enabling certain computational operations. For example, in cloud computing, users can upload encrypted data to a cloud server and then perform computational operations such as searching and sorting while the data is encrypted, without having to decrypt it. This improves both data privacy and security.

[0028] Before this specification elaborates on the product recommendation method based on privacy computing in combination with one or more embodiments, it first introduces the application scenarios of the product recommendation method based on privacy computing.

[0029] Please refer to Figure 1, which is a schematic diagram of a privacy-preserving product recommendation system 100 based on privacy computing, provided in an embodiment of the present invention. The privacy-preserving product recommendation system 100 may include several participating nodes 110, a privacy-preserving product recommendation device 120, and the like. Participating nodes 110 may be service nodes corresponding to product sales platforms, product manufacturing organizations, and the like, and may be electronic devices such as servers. Participating nodes 110 may establish a privacy-preserving computing network, each of which includes a product recommendation model, such as a linear combination model. The privacy-preserving computing network also includes a cooperative node 1100, which is communicatively connected to the participating nodes 110. The cooperative node is configured to participate in the model training process of the product recommendation model in each participating node 110 and to optimize the product recommendation model in each participating node 110. Participating nodes 110 are each communicatively connected to the product recommendation device 120, and are also communicatively connected to each other.

[0030] Specifically, the product recommendation device 120 based on privacy computing can be integrated into an electronic device, which can be a terminal, a server, or other devices. The terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, or a personal computer (PC); the server can be a single server or a server cluster composed of multiple servers. In some embodiments, the product recommendation device based on privacy computing can also be integrated into multiple electronic devices. For example, the product recommendation device based on privacy computing can be integrated into multiple servers, and the product recommendation method based on privacy computing of the present application can be implemented by multiple servers.

[0031] In the embodiments of this specification, the product recommendation device 120 based on privacy computing can be used to obtain a candidate set of product lists; based on a privacy computing network established by several participating nodes, obtain linear combination data of several participating nodes on each product in the candidate set of product lists; sort each product according to the linear combination data of several participating nodes on each product in the candidate set of product lists, and recommend the sorting results to users, etc.

[0032] It should be noted that the scenario diagram of the product recommendation system based on privacy computing shown in Figure 1 is only an example. The product recommendation system based on privacy computing and the scenario described in the embodiment of the present invention are for the purpose of more clearly illustrating the technical solution of the embodiment of the present invention, and do not constitute a limitation on the technical solution provided by the embodiment of the present invention. Ordinary technicians in this field can know that with the evolution of the product recommendation system based on privacy computing and the emergence of new scenarios, the technical solution provided by the embodiment of the present invention is also applicable to similar technical problems.

[0033] Please refer to Figure 2, which is a flowchart of a privacy-preserving computing-based product recommendation method provided by an embodiment of the present invention. This privacy-preserving computing-based product recommendation method can be performed by the privacy-preserving computing-based product recommendation device 120 shown in Figure 1. This privacy-preserving computing-based product recommendation method can include at least the following steps 200 to 220.

[0034] 200. Get a candidate set of product lists.

[0035] The product list candidate set is a product list used to recommend products to users. The product list may include several pieces of product information, including product names and the like.

[0036] In this embodiment, the product recommendation device 120 may obtain the product list candidate set through participating nodes corresponding to the product sales platform or participating nodes corresponding to the product manufacturing organization.

[0037] 210. Based on a privacy computing network established by several participating nodes, obtain linear combination data of each product in the product list candidate set by several participating nodes.

[0038] In this embodiment, the privacy-preserving computing network is composed of several participating nodes and their collaborating nodes. Each participating node is equipped with a product recommendation model, and the participating nodes and collaborating nodes jointly participate in the privacy-preserving computation of each product recommendation model. The product recommendation model can be, for example, a linear regression model.

[0039] In this embodiment, the linear combination data of a certain participating node for a certain product may be a linear combination result obtained based on a product recommendation model in the certain participating node.

[0040] This embodiment can be based on a privacy computing network established by several participating nodes, and through federated learning among the participating nodes, determine the product recommendation models that have been trained in the participating nodes, and then calculate the linear combination data of each product in the product list candidate set based on the trained product recommendation model in each participating node, thereby obtaining the linear combination data of each product in the product list candidate set by the participating nodes.

[0041] In some embodiments, based on a privacy computing network established by several participating nodes, linear combination data of several participating nodes for each product in a product list candidate set is obtained, including: based on the privacy computing network established by several participating nodes, determining a trained product recommendation model in each participating node, the product recommendation model including a linear regression model; obtaining product data of each product in a product list candidate set from any participating node among the several participating nodes; based on a product recommendation model trained in any participating node, obtaining a model output of the product recommendation model for the product data of each product, the model output being the linear combination data of any participating node for each product; traversing all participating nodes to obtain linear combination data of all participating nodes for each product in a product list candidate set.

[0042] In this embodiment, the linear regression model is a statistical model used to predict numerical target variables. Product data refers to the relevant data about users and products in participating nodes. The product data in different participating nodes is different, and the private data in the product data of each participating node is not shared. For example, product data in a product sales platform may include authorized user data, total product sales volume, product manufacturer ratings, etc. Product data in product manufacturers may include product industry research report data, market analysis data, etc.

[0043] In this embodiment, a trained product recommendation model can be provided in each participating node. First, based on the trained product recommendation model in any participating node, the product data of each product is input into the trained linear regression model respectively, and then the linear combination data of each product is output through the trained linear regression model. In this embodiment, all participating nodes are traversed in turn, so as to obtain the linear combination data of each product in the product list candidate set from all participating nodes.

[0044] In some embodiments, the privacy computing network includes several participating nodes and cooperative nodes that are respectively communicatively connected to the participating nodes; the cooperative nodes are provided with private keys, and the participating nodes are respectively provided with public keys paired with the private keys.

[0045] In this embodiment, a private key is a key component used in encrypted communications. It is a confidential encryption key, held only by cooperating nodes in this embodiment. A private key is typically used to decrypt data encrypted with its corresponding public key or to generate digital signatures to verify the authenticity and integrity of data.

[0046] In this embodiment, a public key is a key paired with a private key and used to encrypt data or verify digital signatures. In this embodiment, a cooperating node can send the public key to each participating node. Each participating node receives the public key and uses it to encrypt data during the privacy-preserving computation process. The public key is publicly visible and can be used by any participating node to encrypt data or verify digital signatures, but the corresponding private key cannot be derived from the public key.

[0047] Please refer to Figure 3, which shows a flowchart of a training process of a product recommendation model by any participating node provided in another embodiment of this specification. The training process can be performed by any participating node shown in Figure 1.

[0048] As shown in FIG3 , the training process of the product recommendation model by any participating node may include at least the following steps:

[0049] 300. Obtain product training data;

[0050] 310. Based on the product training data, encryption alignment processing is performed with other nodes in each participating node to obtain encrypted product training data after encryption alignment processing;

[0051] 320. Calculate the linear combination training data corresponding to the encrypted product training data through the product recommendation model in any participating node;

[0052] 330. Encrypt the linear combination training data using the public key to obtain the encrypted linear combination training data of any participating node;

[0053] 340. Send the encrypted linear combination training data of any participating node to other nodes, and receive the encrypted linear combination training data of other nodes;

[0054] 350. Calculate the gradient and loss of the product recommendation model in any participating node using the encrypted linear combination training data from other nodes;

[0055] 360. Send the gradient value and loss value of the product recommendation model in any participating node to the cooperation node; the cooperation node is used to calculate the combined gradient value and combined loss value based on the gradient value and loss value of each participating node, and send the combined gradient value and combined loss value to each participating node respectively;

[0056] 370. Receive the combined gradient value and the combined loss value sent by the cooperation node;

[0057] 380. Update the parameters of the product recommendation model in any participating node according to the combined gradient value and the combined loss value to obtain an updated product recommendation model, until the updated product recommendation model meets the preset conditions, and obtain the trained product recommendation model in any participating node.

[0058] In this embodiment, product training data can be a dataset acquired by any participating node for training a product recommendation model. Each participating node has different product training data, and the product training data within each participating node is not shared. For example, product training data on a product sales platform may include authorized user data, total product sales volume, and manufacturer ratings for several historical products. Product training data on a manufacturer may include industry research reports and market analysis data for several historical products.

[0059] In this embodiment, the linear combination training data corresponding to the product training encrypted data may be a linear combination result obtained by calculating the product training encrypted data based on the product recommendation model.

[0060] In this example, the gradient is the partial derivative of the loss function with respect to the model parameters. The loss is the difference between the model's predicted value for a given data sample and the true value. The loss is calculated using the loss function and is used to measure the model's performance on the training data.

[0061] In this embodiment, the combined gradient value may be a value obtained by combining the gradient values ​​(and / or loss values) of each participating node through the cooperating node. The combined loss value may be a value obtained by combining the loss values ​​(and / or gradient values) of each participating node through the cooperating node. The combination may be implemented by a neural network model in the cooperating node.

[0062] In this embodiment, the updated product recommendation model meets the preset conditions, where the preset conditions may be that the loss value reaches a preset threshold, or the number of iterations reaches a preset number, etc.

[0063] In some embodiments, the cooperative node is provided with a trained neural network model, and the combined gradient value and the combined loss value are calculated according to the gradient value and the loss value of each participating node, including: based on the trained neural network model, the combined gradient value and the combined loss value are calculated according to the gradient value and the loss value of each participating node.

[0064] In this embodiment, the cooperating nodes can set up two neural network models, one for gradient value calculation and the other for loss value calculation. In this embodiment, the input and output of each neural network model can be set according to the number of participating nodes.

[0065] For example, if this embodiment has n participating nodes, the number of inputs to the neural network model (neural network gradient model or neural network loss model) can be set to n, and the number of outputs can also be set to n. For example, this embodiment can have participating nodes A and B, with the gradient value of participating node A and the gradient value of participating node B serving as the two inputs of the neural network gradient model. The output of participating node A (i.e., the combined gradient value corresponding to participating node A) and the output of participating node B (i.e., the combined gradient value corresponding to participating node B) are then obtained through the neural network gradient model.

[0066] In other embodiments, the cooperative nodes may also set up a neural network model, use the gradient values ​​and loss values ​​of each participating model as inputs to the neural network model, and then output the gradient values ​​and loss values ​​of each participating node through the neural network model. For example, in this embodiment, there may be a participating node C and a participating node D, and the gradient value and loss value of participating node C and the gradient value and loss value of participating node D are used as the four inputs of the neural network model, and then four outputs are obtained through the neural network model (i.e., the combined gradient value of participating node C, the combined loss value of participating node C, the combined gradient value of participating node D, and the combined loss value of participating node D).

[0067] In this embodiment, during the training process of the neural network model of the cooperative node, the cooperative node can generate a universal neural network model. Each participating node downloads this universal model locally and uses local data to train the model, and uploads the updated parameter content of the trained model to the cooperative node. The cooperative node optimizes the initial universal model by fusing and evenly dividing the parameter update content of multiple participating nodes. Then, each participating party downloads the updated universal model to perform the above processing. This process is repeated until the loss function in the neural network model converges, and a trained neural network model is obtained.

[0068] In some embodiments, in step 310, based on the product training data, encryption alignment processing is performed between each participating node and other nodes to obtain encrypted product training data after encryption alignment processing, including: based on the homomorphic encryption method, sample alignment processing is performed on the product training data of any participating node and the product training data of other participating nodes to obtain encrypted product training encrypted data after encryption alignment processing in any participating node.

[0069] In this embodiment, an encrypted sample alignment method is used to preprocess the product training data of each participating node. Because the users of each participating node do not completely overlap, the purpose of encrypted sample alignment is to combine homogeneous user features from each participating node for model algorithm training. Homogeneous user features can be features related to a newly issued task. For example, a product sales platform can encrypt the data of authorized users, while a product manufacturer can encrypt its own user features and internal industry analysis data. Both parties use a homomorphic encryption algorithm to align samples, retaining homogeneous user features and industry research encrypted samples while removing irrelevant samples.

[0070] In some embodiments, a combined gradient value and a combined loss value are calculated based on the gradient value and loss value of each participating node, including: using a private key to decrypt the encrypted gradient value and loss value sent by each participating node to obtain the decrypted gradient value of each participating node and the decrypted loss value of each participating node; and combining the decrypted gradient value of each participating node and the decrypted loss value of each participating node to obtain a combined gradient value and a combined loss value.

[0071] In this embodiment, based on its own product recommendation model, each participating node calculates a linear combination result based on its own data, and then each participating node can encrypt the linear combination result with a public key and transmit it to other participating nodes, and then calculate the gradient value and loss value of the product recommendation model based on the encrypted linear combination result of other participating nodes; then, each participating node uploads the gradient value and loss value to the cooperative node, and the cooperative node uses its own private key to decrypt the gradient value and loss to obtain the decrypted result, and combines the decrypted gradient value and loss value of each participating node to obtain the combined gradient value and combined loss value; then, the combined gradient value and combined loss value are transmitted back to each participating node respectively, and each participating node receives the combined gradient value and combined loss value and updates and iterates until the loss function of each model converges, completing the model training in the privacy calculation process, and obtaining the trained product recommendation model in any participating node, and then each participating node can deploy its own product recommendation model to the production environment. In this embodiment, the data of each participating node cannot be exchanged directly. In this embodiment, when the data of each participating node are not shared with each other, a more effective product recommendation model can be trained through privacy computing, while protecting the private data of users and each participating node from being shared.

[0072] 220. Sort the products in the product list candidate set based on the linear combination data of the participating nodes, and recommend the sorted results to the user.

[0073] Please refer to Figure 4, which shows a schematic diagram of a process for sorting products according to another embodiment of this specification. The sorting process can be performed by the product recommendation device 120 based on privacy computing shown in Figure 1.

[0074] As shown in Figure 4, the linear combination data of each product in the product list candidate set is sorted based on several participating nodes, and the sorting results are recommended to the user, including:

[0075] 400. Obtain linear combination data of several participating nodes for any product in the product list candidate set;

[0076] 410. Add the linear combination data of any product in the product list candidate set by several participating nodes to obtain the total linear combination data of any product;

[0077] 420. Traverse all products in the product list candidate set to obtain the total linear combination data corresponding to all products;

[0078] 430. Sort the products according to the total linear combination data corresponding to all products to obtain a sorting result;

[0079] 440. Recommend the sorting results to the user.

[0080] In this embodiment, when it is detected that a user accesses the organization's recommended product list through the application interface of the product recommendation device 120, the product recommendation device 120 can first recall the product list candidate set through each participating node, and then obtain the linear combination data of each product in the product list candidate set by several participating nodes based on the privacy computing network established by several participating nodes, and then sort each product according to the linear combination data of each product in the product list candidate set by several participating nodes.

[0081] For example, there are 3 products: product a, product b and product c, and 2 participating nodes: participating node E and participating node F; among them, the linear combination results of participating node E for product a, product b and product c are 0.3, 0.4 and 0.6 respectively, and the linear combination results of participating node F for product a, product b and product c are 0.2, 0.5 and 0.7 respectively. Then, the linear combination results of participating node E and participating node F for each product are added together, and the total linear combination data of product a is 0.5, the total linear combination data of product b is 0.9, and the total linear combination data of product c is 1.3; then, the products are sorted from high to low according to the scores, product c is greater than product b, and product b is greater than product a.

[0082] The embodiments of this specification are based on a privacy computing network established by several participating nodes, which can protect the confidentiality of the data of each participant while ensuring that the data is not leaked or abused during the calculation process, protecting the private data of users and each participant from being shared, and improving the accuracy of product recommendations.

[0083] Please refer to Figure 5, which shows a flow chart of a product recommendation method based on privacy computing provided in another embodiment of this specification, which is specifically applied in the new fund sales scenario. This method can be executed by the product recommendation device 120 based on privacy computing shown in Figure 1.

[0084] As shown in FIG5 , the product recommendation method based on privacy computing may at least include the following steps 500 to 520 .

[0085] 500. Obtain a candidate list of newly issued fund products.

[0086] In this embodiment, the candidate set of the newly issued fund product list may include information of several newly issued funds, and the newly issued fund information may include the name of the newly issued fund and the like.

[0087] 510. Based on a privacy computing network established by several participating nodes, obtain linear combination data of each newly issued fund product in the candidate set of the newly issued fund product list by several participating nodes, and the participating nodes include the fund sales platform and the service nodes corresponding to the fund company.

[0088] In this embodiment, based on the privacy computing network established by the fund sales platform and the fund company, linear combination data of several participating nodes for each product in the candidate set of the newly issued fund product list is obtained, which can include determining the trained linear regression model in each participating node based on the privacy computing network established by the several fund sales platforms and the fund company; obtaining product data of each product in the candidate set of the product list from any participating node among the several participating nodes; obtaining the model output of the linear regression model for the product data of each product based on the linear regression model trained in any participating node, and the model output is the linear combination data of any participating node for each product; traversing all participating nodes to obtain the linear combination data of all participating nodes for each product in the candidate set of the product list.

[0089] Among them, the privacy computing network includes several participating nodes and cooperative nodes that are respectively communicated with the participating nodes; the cooperative nodes are provided with private keys, and the participating nodes are respectively provided with public keys paired with the private keys.

[0090] In some embodiments, the training process of the linear regression model of any participating node may include: obtaining product training data; performing sample alignment processing on the product training data of any participating node and the product training data of other participating nodes based on the homomorphic encryption method to obtain the encrypted product training encrypted data of any participating node; calculating the linear combination training data corresponding to the product training encrypted data through the linear regression model in any participating node; encrypting the linear combination training data using the public key to obtain the encrypted linear combination training data of any participating node; sending the encrypted linear combination training data of any participating node to other nodes, and receiving the encrypted linear combination training data of other nodes. The method comprises the following steps: first, calculating the gradient value and loss value of the linear regression model in any participating node using the encrypted linear combination training data of other nodes; sending the gradient value and loss value of the linear regression model in any participating node to the cooperative node; the cooperative node is used to calculate the combined gradient value and the combined loss value according to the gradient value and loss value of each participating node, and send the combined gradient value and the combined loss value to each participating node respectively; receiving the combined gradient value and the combined loss value sent by the cooperative node; updating the parameters of the linear regression model in any participating node according to the combined gradient value and the combined loss value to obtain an updated linear regression model, until the linear regression model meets the preset conditions, and obtaining a trained linear regression model in any participating node.

[0091] In this embodiment, product training data is obtained. For example, the sales platform has information such as industry track data of fund investments, historical performance data of fund managers, and rating data of institutions; the fund company has internal industry research report data, market analysis data, user research data, and other information. The data of the sales platform and the data of the fund company cannot be shared.

[0092] In this embodiment, encrypted alignment is performed between each participating node. Since the users of the sales platform and the fund company do not completely overlap, the purpose of encrypted sample alignment is to combine the homogeneous user features of both parties for algorithm training. In this embodiment, homogeneous user features refer to features related to the task of newly issuing funds. The sales platform first encrypts the data of authorized users, and the fund company encrypts its own user features and internal industry analysis data. Both parties use a homomorphic encryption algorithm to align samples, retaining homogeneous user features and industry research encrypted samples, and removing irrelevant samples. For example, for white-collar users aged 30-45, regarding their preference for the new energy industry, this embodiment can use encrypted sample data from both white-collar users and new energy industry funds for subsequent model training.

[0093] In some embodiments, a combined gradient value and a combined loss value are calculated based on the gradient value and loss value of each participating node, including: using a private key to decrypt the encrypted gradient value and loss value sent by each participating node to obtain the decrypted gradient value of each participating node and the decrypted loss value of each participating node; and combining the decrypted gradient value of each participating node and the decrypted loss value of each participating node to obtain a combined gradient value and a combined loss value.

[0094] Please refer to Figure 6, which shows the training process of the linear regression model by any participating node provided in another embodiment of the present specification. As shown in Figure 6, the cooperative node can send the public key to the fund sales platform and the fund company respectively. The fund sales platform can calculate the linear combination result A based on its own data according to the sales platform algorithm model (the linear regression model corresponding to the fund sales platform). The fund company can calculate the linear combination result B based on its own data according to the fund company algorithm model (the linear regression model corresponding to the fund company). Then the fund sales platform and the fund company will encrypt their respective linear combination results with the public key and transmit them to each other. For example, the fund sales platform will encrypt the linear combination result A with the public key and transmit it to the fund company. The fund company will encrypt the linear combination result B with the public key and transmit it to the fund sales platform. Then the fund sales platform will calculate the gradient value and loss value based on the linear combination result B encrypted with the fund company's public key. The fund company The gradient value and loss value are calculated based on the linear combination result A encrypted with the public key of the fund sales platform; then the fund company and the fund sales platform upload the gradient value and loss value they calculated to the cooperation node respectively, and the cooperation node uses its own private key to decrypt the gradient value and loss value of both parties to obtain the decrypted result, and then perform combination processing to obtain the combination value (combined gradient value or combined loss value) corresponding to the fund sales platform and the fund company, and then transmit the combination value corresponding to the fund sales platform to the fund sales platform, and transmit the combination value corresponding to the fund company to the fund company. After receiving the combination value, the fund sales platform and the fund company update and iterate the model until the loss function converges, completing the training of the linear regression model. The fund sales platform and the fund company deploy the trained linear regression model to the production environment.

[0095] In some embodiments, the cooperative node is provided with a trained neural network model, and the combined gradient value and the combined loss value are calculated according to the gradient value and the loss value of each participating node, including: based on the trained neural network model, the combined gradient value and the combined loss value are calculated according to the gradient value and the loss value of each participating node.

[0096] In this embodiment, a general neural network model is generated by the cooperative nodes jointly built by the fund sales platform and the fund company in the federated learning architecture. Each participating node downloads this general model to the local and uses local data to train the model, and uploads the updated content of the trained model to the cooperative node. The initial general model is optimized by fusing and evenly dividing the updated content of multiple participating nodes. Each participating node then downloads the updated general model and performs the above processing until the loss function of the neural network model is absorbed, thereby obtaining a trained neural network model.

[0097] 520. Sort each newly issued fund product in the candidate set of the newly issued fund product list according to the linear combination data of each newly issued fund product in the candidate set of the newly issued fund product list by several participating nodes, and recommend the sorting result to the user.

[0098] In some embodiments, the products in the product list candidate set are sorted according to the linear combination data of several participating nodes on each product, and the sorting results are recommended to the user, including: obtaining the linear combination data of several participating nodes on any product in the product list candidate set; adding the linear combination data of several participating nodes on any product in the product list candidate set to obtain the total linear combination data of any product; traversing all products in the product list candidate set to obtain the total linear combination data corresponding to all products; sorting each product according to the total linear combination data corresponding to all products to obtain the sorting result; and recommending the sorting result to the user.

[0099] For example, when the product recommendation device 120 detects that a user accesses the institution's recommended product list through the application interface, it first recalls a candidate set of the product list, such as 4 newly issued funds. The fund sales platform and the fund company respectively give a linear combination score based on the 5 newly issued funds in combination with their respective models, such as product 1 (0.3 points for the fund sales platform and 0.4 points for the fund company), product 2 (0.6 points for the fund sales platform and 0.3 points for the fund company), product 3 (0.4 points for the fund sales platform and 0.2 points for the fund company), and product 4 (0.7 points for the fund sales platform and 0.4 points for the fund company). The scores of both parties are then added together to obtain product 1 (0.7 points), product 2 (0.9 points), product 3 (0.6 points), and product 4 (1.1 points). Finally, according to the scores from high to low, product 4 is greater than product 2, product 2 is greater than product 1, and product 1 is greater than product 3. The sorted products are then displayed to the user.

[0100] In the sales area of ​​newly issued funds, the fund sales platform will recommend suitable funds to users based on the characteristics of users and institutional products. In the traditional recommendation algorithm model, the sales platform can only use the personal information authorized by the user and the public data transmitted to the platform by the fund company for model training. To further optimize the algorithm recommendation effect, more data is needed to train the model. However, there is a lot of undisclosed internal information within the newly issued fund institutions, which involves the use of privacy data: the fund company cannot provide internal non-public information to the sales platform, and the sales platform cannot share the user's personal information with the fund company. This embodiment can use privacy computing to train the model to ensure that the private data of both parties are not shared, but a better product recommendation model can be trained based on the private data of both parties.

[0101] The embodiments of this specification are based on a privacy computing network established by several participating nodes, which can protect the confidentiality of the data of each participant while ensuring that the data is not leaked or abused during the calculation process, protecting the private data of users and each participant from being shared, and improving the accuracy of product recommendations.

[0102] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0103] Please refer to Figure 7, which is a structural diagram of a product recommendation device based on privacy computing provided in an embodiment of this specification.

[0104] As shown in FIG. 7 , the product recommendation device based on privacy computing may include at least a list acquisition module 700 , a privacy computing module 710 and a ranking module 720 .

[0105] The list acquisition module 700 is used to obtain a candidate set of product lists; the privacy computing module 710 is used to obtain the linear combination data of each product in the candidate set of product lists by several participating nodes based on a privacy computing network established by several participating nodes; the sorting module 720 is used to sort each product according to the linear combination data of each product in the candidate set of product lists by several participating nodes, and recommend the sorting results to the user.

[0106] In some embodiments, the privacy computing module includes: a model determination module, which is used to determine the product recommendation model trained in each participating node based on a privacy computing network established by several participating nodes, and the product recommendation model includes a linear regression model; a first data acquisition module, which is used to obtain product data of each product in the product list candidate set from any participating node among several participating nodes; a model output module, which is used to obtain the model output of the product recommendation model for the product data of each product based on the product recommendation model trained in any participating node, and the model output is the linear combination data of any participating node for each product; a node traversal module, which is used to traverse all participating nodes and obtain the linear combination data of all participating nodes for each product in the product list candidate set.

[0107] In some embodiments, in the privacy computing module, the privacy computing network includes several participating nodes and cooperative nodes that are respectively communicated with the several participating nodes; the cooperative nodes are provided with private keys, and the several participating nodes are respectively provided with public keys paired with the private keys.

[0108] In some embodiments, the privacy computing module includes model training modules corresponding to several participating nodes, and the model training module corresponding to any participating node includes: a training data module for obtaining product training data; an encryption alignment module for performing encryption alignment processing with other nodes in each participating node based on the product training data to obtain encrypted product training encrypted data after encryption alignment processing; a linear combination module for calculating the linear combination training data corresponding to the product training encrypted data through the product recommendation model in any participating node; a public key encryption module for encrypting the linear combination training data using the public key to obtain the encrypted linear combination training data of any participating node; a data sending module for sending the encrypted linear combination training data of any participating node to other nodes, and receiving the encrypted linear combination training data of other nodes. data; a gradient calculation module, used to calculate the gradient value and loss value of the product recommendation model in any participating node using the encrypted linear combination training data of other nodes; a combination calculation module, used to send the gradient value and loss value of the product recommendation model in any participating node to the cooperation node; the cooperation node is used to calculate the combined gradient value and combined loss value based on the gradient value and loss value of each participating node, and send the combined gradient value and combined loss value to each participating node respectively; a gradient receiving module, used to receive the combined gradient value and combined loss value sent by the cooperation node; a parameter updating module, used to update the parameters of the product recommendation model in any participating node according to the combined gradient value and combined loss value, and obtain an updated product recommendation model, until the updated product recommendation model meets the preset conditions, and a trained product recommendation model in any participating node is obtained.

[0109] In some embodiments, the combined calculation module includes a combined calculation sub-module applied to the cooperative node, and the combined calculation sub-module includes: a decryption module, which is used to use a private key to decrypt the encrypted gradient value and loss value sent by each participating node to obtain the decrypted gradient value of each participating node and the decrypted loss value of each participating node; a combination sub-module, which is used to combine the decrypted gradient value of each participating node and the decrypted loss value of each participating node to obtain a combined gradient value and a combined loss value.

[0110] In some embodiments, the cooperative node is provided with a trained neural network model, and the combined calculation submodule applied to the cooperative node includes: a model calculation module, which is used to calculate the combined gradient value and the combined loss value based on the trained neural network model and the gradient value and loss value of each participating node.

[0111] In some embodiments, the encryption alignment module includes: an encryption alignment sub-module, which is used to perform sample alignment processing on the product training data of any participating node and the product training data of other participating nodes based on the homomorphic encryption method, and obtain the encrypted product training data after encryption alignment processing in any participating node.

[0112] In some embodiments, the sorting module includes: a second data acquisition module, used to obtain the linear combination data of several participating nodes for any product in the product list candidate set; an addition module, used to add the linear combination data of several participating nodes for any product in the product list candidate set to obtain the total linear combination data of any product; a product traversal module, used to traverse all products in the product list candidate set to obtain the total linear combination data corresponding to all products; a product sorting module, used to sort each product according to the total linear combination data corresponding to all products to obtain a sorting result; a recommendation module, used to recommend the sorting result to the user.

[0113] Based on the content of the product recommendation system based on privacy computing in multiple embodiments of this specification, it can be seen that the embodiments of this specification are based on a privacy computing network established by several participating nodes, which can protect the confidentiality of the data of each participant while ensuring that the data is not leaked or abused during the calculation process, protecting the private data of users and each participant from being shared, and improving the accuracy of product recommendations.

[0114] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referenced across them. Each embodiment focuses on the differences from the other embodiments. In particular, the product recommendation system embodiment based on privacy computing is fundamentally similar to the product recommendation method embodiment based on privacy computing, so its description is relatively simple. For relevant parts, refer to the description of the method embodiment.

[0115] Please refer to FIG8 , which shows a schematic structural diagram of an electronic device provided in an embodiment of this specification.

[0116] As shown in FIG. 8 , the electronic device 800 may include: at least one processor 801 , at least one network interface 804 , a user interface 803 , a memory 805 , and at least one communication bus 802 .

[0117] The communication bus 802 may be used to implement connection and communication among the above components.

[0118] The user interface 803 may include buttons, and the optional user interface may also include a standard wired interface or a wireless interface.

[0119] The network interface 804 may include, but is not limited to, a Bluetooth module, an NFC module, a Wi-Fi module, and the like.

[0120] Among them, the processor 801 may include one or more processing cores. The processor 801 uses various interfaces and lines to connect the various parts of the entire electronic device 800, and executes various functions of the electronic device 800 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 805, and calling data stored in the memory 805. Optionally, the processor 801 can be implemented in at least one hardware form of DSP, FPGA, and PLA. The processor 801 can integrate one or a combination of CPU, GPU, and modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; and the modem is used to handle wireless communications. It is understandable that the above-mentioned modem may not be integrated into the processor 801, but may be implemented separately through a chip.

[0121] The memory 805 may include RAM or ROM. Optionally, the memory 805 includes a non-transitory computer-readable medium. The memory 805 can be used to store instructions, programs, codes, code sets, or instruction sets. The memory 805 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 805 may also optionally be at least one storage device located away from the aforementioned processor 801. The memory 805, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a product recommendation application based on privacy computing. The processor 801 may be used to call the product recommendation application based on privacy computing stored in the memory 805 and execute the steps of product recommendation and formulation based on privacy computing mentioned in the above-mentioned embodiments.

[0122] The embodiments of this specification also provide a computer-readable storage medium having instructions stored therein that, when executed on a computer or processor, cause the computer or processor to perform one or more steps of the embodiments shown in Figures 2 to 6 . If the components of the electronic device described above are implemented as software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium.

[0123] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of this specification is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. Available media may be magnetic media (eg, floppy disks, hard disks, tapes), optical media (eg, digital versatile discs (DVDs)), or semiconductor media (eg, solid state disks (SSDs)).

[0124] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. The technical features of this embodiment and the implementation scheme can be combined in any manner unless they conflict.

[0125] The above embodiments are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Without departing from the design spirit of this specification, various modifications and improvements made to the technical solutions of this specification by ordinary technicians in this field should fall within the scope of protection determined by the claims of this specification.

Claims

1. A product recommendation method based on privacy computing, comprising: Get the candidate set of product list; Based on a privacy computing network established by a number of participating nodes, obtain linear combination data of the participating nodes for each product in the product list candidate set; According to the linear combination data of each product in the product list candidate set by the several participating nodes, the products are sorted, and the sorting results are recommended to the user.

2. The method according to claim 1, wherein the step of obtaining linear combination data of each product in the candidate set of the product list by the plurality of participating nodes based on a privacy computing network established by the plurality of participating nodes comprises: Based on a privacy computing network established by a number of participating nodes, determine a product recommendation model trained in each participating node, wherein the product recommendation model includes a linear regression model; Obtain product data of each product in the product list candidate set from any participating node among the plurality of participating nodes; Based on the product recommendation model trained in any one of the participating nodes, obtain the model output of the product recommendation model for the product data of each product, wherein the model output is the linear combination data of each product by any one of the participating nodes; Traverse all participating nodes to obtain linear combination data of all participating nodes for each product in the candidate set of the product list.

3. According to the method described in claim 2, the privacy computing network includes the several participating nodes and cooperative nodes that are respectively communicated with the several participating nodes; the cooperative nodes are provided with a private key, and the several participating nodes are respectively provided with a public key paired with the private key.

4. The method according to claim 3, wherein the determining, based on a privacy computing network established by a plurality of participating nodes, of the product recommendation models trained in the participating nodes, the training process of the product recommendation model by any participating node is applied to the any participating node, and comprises: Get product training data; Based on the product training data, encryption alignment processing is performed with other nodes in each participating node to obtain encrypted product training data after encryption alignment processing; Calculate the linear combination training data corresponding to the product training encrypted data through the product recommendation model in any one of the participating nodes; Encrypting the linear combination training data using a public key to obtain the encrypted linear combination training data of any participating node; Sending the encrypted linear combination training data of any participating node to other nodes, and receiving the encrypted linear combination training data of other nodes; Calculate the gradient value and loss value of the product recommendation model in any participating node by using the encrypted linear combination training data of the other nodes; Send the gradient value and loss value of the product recommendation model in any participating node to the cooperation node point; the cooperation node is used to calculate the combined gradient value and the combined loss value according to the gradient value and the loss value of each participating node, and send the combined gradient value and the combined loss value to each participating node respectively; Receiving the combined gradient value and the combined loss value sent by the cooperation node; The parameters of the product recommendation model in any one of the participating nodes are updated according to the combined gradient value and the combined loss value to obtain an updated product recommendation model, until the updated product recommendation model meets the preset conditions, thereby obtaining a trained product recommendation model in any one of the participating nodes.

5. The method according to claim 4, wherein the step of calculating the combined gradient value and the combined loss value according to the gradient value and the loss value of each participating node comprises: Decrypt the encrypted gradient value and loss value sent by each participating node using the private key to obtain the decrypted gradient value of each participating node and the decrypted loss value of each participating node; The gradient value of each participating node after decryption and the loss value of each participating node after decryption are respectively combined to obtain a combined gradient value and a combined loss value.

6. According to the method of claim 5, the cooperation node is provided with a trained neural network model, and the combined gradient value and the combined loss value are calculated according to the gradient value and the loss value of each participating node, comprising: Based on the trained neural network model, the combined gradient value and the combined loss value are calculated according to the gradient value and loss value of each participating node.

7. The method according to claim 4, wherein the step of performing encryption alignment processing with other nodes in each participating node based on the product training data to obtain encrypted product training data after encryption alignment processing comprises: Based on the homomorphic encryption method, sample alignment is performed on the product training data of any participating node and the product training data of other participating nodes to obtain encrypted product training encrypted data after encryption alignment in any participating node.

8. The method according to claim 1, wherein the sorting of the products in the candidate set of the product list according to the linear combination data of the several participating nodes and recommending the sorting results to the user comprises: Obtain linear combination data of a plurality of participating nodes for any product in the candidate set of the product list; Adding the linear combination data of any one product in the candidate set of the product list by the plurality of participating nodes to obtain the total linear combination data of the any one product; Traversing all products in the candidate set of the product list to obtain total linear combination data corresponding to all products; Sort the products according to the total linear combination data corresponding to all the products to obtain a sorting result; The ranking result is recommended to the user.

9. A product recommendation system based on privacy computing, comprising: List acquisition module, used to obtain product list candidate sets; A privacy computing module, configured to obtain linear combination data of each product in the product list candidate set by the plurality of participating nodes based on a privacy computing network established by the plurality of participating nodes; The sorting module is used to sort the products in the product list candidate set according to the linear combination data of the several participating nodes, and recommend the sorting results to the user.

10. The system according to claim 9, wherein the privacy computing module comprises: A model determination module, used to determine a product recommendation model trained in each participating node based on a privacy computing network established by several participating nodes, wherein the product recommendation model includes a linear regression model; A first data acquisition module is used to acquire product data of each product in the product list candidate set from any participating node among the plurality of participating nodes; A model output module, used for obtaining a model output of the product recommendation model for the product data of each product based on the product recommendation model trained in any one of the participating nodes, wherein the model output is a linear combination data of each product by any one of the participating nodes; The node traversal module is used to traverse all participating nodes and obtain the linear combination data of all participating nodes for each product in the candidate set of the product list.

11. According to the system of claim 10, in the privacy computing module, the privacy computing network includes the several participating nodes and cooperation nodes that are respectively communicated with the several participating nodes; the cooperation nodes are provided with a private key, and the several participating nodes are respectively provided with a public key paired with the private key.

12. According to the system of claim 11, the privacy computing module includes model training modules corresponding to a number of participating nodes, and the model training module corresponding to any participating node includes: Training data module, used to obtain product training data; An encryption alignment module, used to perform encryption alignment processing with other nodes in each participating node based on the product training data to obtain encrypted product training data after encryption alignment processing; A linear combination module, used for calculating the linear combination training data corresponding to the product training encrypted data through the product recommendation model in any one of the participating nodes; A public key encryption module, used to encrypt the linear combination training data using a public key to obtain the encrypted linear combination training data of any participating node; A data sending module, used to send the encrypted linear combination training data of any participating node to other nodes, and receive the encrypted linear combination training data of other nodes; A gradient calculation module, used to calculate the gradient value and loss value of the product recommendation model in any participating node by using the encrypted linear combination training data of the other nodes; A combined calculation module, used for sending the gradient value and loss value of the product recommendation model in any participating node to the cooperation node; The cooperation node is used to calculate a combined gradient value and a combined loss value according to the gradient value and the loss value of each participating node, and send the combined gradient value and the combined loss value to each participating node respectively; A gradient receiving module, used for receiving a combined gradient value and a combined loss value sent by the cooperation node; A parameter updating module is used to update the parameters of the product recommendation model in any participating node according to the combined gradient value and the combined loss value to obtain an updated product recommendation model, until the updated product recommendation model meets the preset conditions, thereby obtaining a trained product recommendation model in any participating node.

13. The system according to claim 12, wherein the combined computing module comprises a combined computing submodule applied to the cooperation node, the combined computing submodule comprising: A decryption module is used to decrypt the encrypted gradient value and loss value sent by each participating node using a private key to obtain the decrypted gradient value of each participating node and the decrypted loss value of each participating node; The combination submodule is used to combine the gradient value of each participating node after decryption and the loss value of each participating node after decryption to obtain a combined gradient value and a combined loss value.

14. The system according to claim 13, wherein the cooperation node is provided with a trained neural network model, and the combined calculation submodule applied to the cooperation node comprises: The model calculation module is used to calculate the combined gradient value and the combined loss value according to the gradient value and loss value of each participating node based on the trained neural network model.

15. The system according to claim 12, wherein the encryption alignment module comprises: The encryption alignment submodule is used to perform sample alignment processing on the product training data of any participating node and the product training data of other participating nodes based on the homomorphic encryption method to obtain the encrypted product training data after encryption alignment processing in any participating node.

16. The system according to claim 9, wherein the sorting module comprises: A second data acquisition module is used to acquire linear combination data of a plurality of participating nodes for any product in the candidate set of the product list; An adding module, used for adding the linear combination data of any one product in the candidate set of the product list by the plurality of participating nodes to obtain the total linear combination data of the any one product; A product traversal module, used to traverse all products in the candidate set of the product list to obtain the total linear combination data corresponding to all products; A product sorting module, used to sort each product according to the total linear combination data corresponding to all the products, to obtain a sorting result; A recommendation module is used to recommend the ranking results to users.

17. An electronic device comprising a processor and a memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the method according to any one of claims 1 to 8.

18. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images