Product recommendation method and device integrating multiple recommendation models, medium and product

Abstract

The invention discloses a product recommendation method and device integrating multiple recommendation models, a medium and a product, and relates to the technical field of product recommendation, and the method comprises the steps: inputting a target user and all to-be-recommended products into a plurality of correction models, and obtaining an interaction score set corresponding to each correction model; the correction model is a model obtained by performing probability correction on the recommendation model, and different correction models correspond to different recommendation models; adding all the first interaction scores corresponding to each to-be-recommended product to obtain a plurality of second interaction scores; and according to the second interaction score, arranging the corresponding to-be-recommended products according to a priority sequence, and recommending the arranged to-be-recommended products to the target user. According to the method, integration of the multiple recommendation models is realized through a probability correction method, so that results of the multiple models can be combined, the view of each recommendation model is comprehensively considered, and compared with the performance of the multiple integrated models, the integrated model has a better recommendation effect.

Description

Technical Field

[0001] The present application relates to the technical field of product recommendation, and in particular to a product recommendation method, device, medium and product integrating multiple recommendation models. Background Art

[0002] In the face of the explosive growth of information in modern times, the recommendation system, as an information filtering system, can predict and recommend products or services that users may be interested in. These systems play a vital role in e-commerce, social media, content sharing platforms, and many other online services. The core goal of the recommendation system is to connect users with items that they may like but have not yet discovered, thereby improving user satisfaction, increasing user engagement, and improving business performance. Common recommendation scenarios include: Taobao recommends items that users may like to users, thereby increasing users' desire to shop; Douyin recommends interesting videos to users to improve user retention. In general, the recommendation system is an information filtering system that aims to predict items that users may like from a huge pool of candidate items to be recommended. As of 2023, there are about 2 billion items on Taobao and Tmall, China's well-known e-commerce platforms. It is impossible for users to find their favorite items from such a large number of items. At this time, the recommendation system needs to play a role. The recommendation system comprehensively considers many aspects of information, such as items that users have purchased before, and the content characteristics of items (such as the color and image characteristics of items).

[0003] However, existing recommendation system methods have their own starting points, and each recommendation model generates recommendation results based on its own view. Summary of the invention

[0004] The purpose of this application is to provide a product recommendation method, device, medium and product that integrates multiple recommendation models, which can combine the results of multiple models, comprehensively consider multiple views, and achieve improved recommendation effects.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a product recommendation method integrating multiple recommendation models, the product recommendation method integrating multiple recommendation models comprising:

[0007] Input the target user and all the products to be recommended into several correction models to obtain the interaction score set corresponding to each correction model; the interaction score set includes multiple elements, and the element is the first interaction score corresponding to the target user and a product to be recommended; wherein the number of elements is the same as the number of products to be recommended, and different elements correspond to different products to be recommended; the correction model is a model obtained after probability correction of the recommendation model, and different correction models correspond to different recommendation models; the recommendation model is determined according to sample data and a recommendation neural network, and different recommendation models correspond to different recommendation neural networks; the sample data is the interaction record between the sample user and the product; the sample data is used to train the recommendation neural network to obtain the recommendation model;

[0008] According to the interaction score set corresponding to each of the correction models, all first interaction scores corresponding to each of the to-be-recommended products are added together to obtain a plurality of second interaction scores;

[0009] According to the second interaction score, the corresponding products to be recommended are arranged in order of priority, and the products to be recommended arranged in order of priority are recommended to the target user.

[0010] Optionally, the interaction record includes: historical data of the user's shopping on the e-commerce platform.

[0011] Optionally, probability correction is performed on the recommendation model, specifically including:

[0012] A Gaussian correction function or a gamma correction function is used to perform probability correction on each of the recommended models to obtain a plurality of corrected models.

[0013] Optionally, a Gaussian correction function is used to perform probability correction on each of the recommendation models to obtain several correction models, specifically including:

[0014] By formula Map the interaction score output by the recommendation model; where a, b, c are learnable parameters, s is the interaction score output by the recommendation model, and σ is the Sigmoid function. is the score after mapping, 2as+b>0.

[0015] Optionally, the gamma correction function is:

[0016] Among them, φ = {a, b, c} is a learnable parameter, σ is the Sigmoid function, and s is the interaction score output by the recommendation model. is the corrected score.

[0017] Optionally, the step of inputting the target user and all the products to be recommended into a plurality of the correction models to obtain a set of interaction scores corresponding to each correction model specifically includes:

[0018] Inputting the target user and all the products to be recommended into several correction models;

[0019] Each correction model is used to calculate the interaction score of each of the to-be-recommended products for the target user, to obtain a plurality of first interaction scores.

[0020] Optionally, according to the second interaction score, the corresponding to-be-recommended products are arranged in order of priority, and the to-be-recommended products arranged in order of priority are recommended to the target user, specifically including:

[0021] Sort the corresponding recommended products according to the second interaction scores to obtain a sorting result;

[0022] The first k recommended products in the ranking results are recommended to the target user, where k is a natural number.

[0023] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described product recommendation methods integrating multiple recommendation models.

[0024] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described product recommendation methods integrating multiple recommendation models.

[0025] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned product recommendation methods integrating multiple recommendation models.

[0026] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0027] The present application provides a product recommendation method, device, medium and product integrating multiple recommendation models, the method comprising: inputting a target user and all products to be recommended into several correction models to obtain an interaction score set corresponding to each correction model; the interaction score set comprises multiple elements, wherein the element is a first interaction score corresponding to a target user and a product to be recommended; wherein the number of elements is the same as the number of products to be recommended, and different elements correspond to different products to be recommended; the correction model is a model obtained after probability correction of the recommendation model, and different correction models correspond to different recommendation models; the recommendation model is determined based on sample data and a recommendation neural network, and different recommendation models correspond to different recommendation neural networks; the sample data is an interaction record between a sample user and a product; the sample data is used to train the recommendation neural network to obtain a recommendation model; according to the interaction score set corresponding to each correction model, all first interaction scores corresponding to each product to be recommended are added to obtain multiple second interaction scores; according to the second interaction scores, the corresponding products to be recommended are arranged in priority order, and the products to be recommended arranged in priority order are recommended to the target user. This application realizes the integration of multiple recommendation models through the probability correction method, so that the results of multiple models can be combined and the view of each recommendation model can be comprehensively considered. Compared with the performance of the integrated multiple models, the integrated model has a better recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 This is an application environment diagram of a product recommendation method integrating multiple recommendation models in one embodiment of the present application;

[0030] Figure 2 A flowchart of a product recommendation method integrating multiple recommendation models provided in one embodiment of the present application;

[0031] Figure 3 A schematic diagram of a recommended data set for an e-commerce platform mall provided in an embodiment of the present application;

[0032] Figure 4 A schematic diagram of the predicted score density of positive samples and negative samples of the recommendation model provided in one embodiment of the present application;

[0033] Figure 5 A schematic diagram of a common recommendation model provided in an embodiment of the present application;

[0034] Figure 6 A schematic diagram of a calibration model provided in an embodiment of the present application;

[0035] Figure 7 A schematic diagram comparing a common recommendation model and a correction model provided in an embodiment of the present application;

[0036] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. 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 creative work are within the scope of protection of this application.

[0038] The recommendation system comprehensively considers many aspects of information, such as items that users have purchased (it is generally believed that user purchases mean that users like them), content features of items (such as the color of items, image features, etc.). Existing recommendation system methods have their own starting points. Each recommendation model can generate recommendation results in its own view. If the results of multiple models can be combined and multiple views can be comprehensively considered, the recommendation effect can be improved. However, there is currently no method to achieve the integration of recommendation models. Therefore, this application proposes a general method for integrating recommendation models.

[0039] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0040] The product recommendation method integrating multiple recommendation models provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers.

[0041] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0042] In an exemplary embodiment, Figure 2 As shown, a product recommendation method integrating multiple recommendation models is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps S1 to S6. Among them:

[0043] S1. Input the target user and all the products to be recommended into several correction models to obtain the interaction score set corresponding to each correction model; the interaction score set includes multiple elements, and the element is the first interaction score corresponding to the target user and a product to be recommended; wherein the number of elements is the same as the number of the products to be recommended, and different elements correspond to different products to be recommended; the correction model is a model obtained after probability correction of the recommendation model, and different correction models correspond to different recommendation models; the recommendation model is determined based on sample data and a recommendation neural network, and different recommendation models correspond to different recommendation neural networks; the sample data is the interaction record between the sample user and the product; the sample data is used to train the recommendation neural network to obtain a recommendation model.

[0044] Specifically:

[0045] A1. Obtain the interaction record between the user and the product. The interaction record includes: the historical data of the user's shopping on the e-commerce platform. In addition to the recommendation on the e-commerce platform, the recommendation system in this application also has many application scenarios, such as video recommendation, music recommendation, takeaway recommendation, etc.

[0046] The data set in this embodiment is derived from a real application scenario and is simplified and abstracted to some extent. To introduce the method, the baby data set is taken as an example. Figure 3 As shown in the figure, it is an e-commerce platform mall recommendation dataset, which records the user's shopping history on the e-commerce platform. Each row records the interaction record between a user and an item. The first row in the figure records that the user with id 0 gave a score of 5 to the item with id 1587. timestamp is the timestamp, which represents the time when this score was made. x_label is the training set, validation set and test set. 0 is the training set, 1 is the validation set, and 2 is the test set. The recommendation dataset can only record which items the user has interacted with. Interactions that have not occurred cannot be recorded. The recommendation system needs to predict which items the user may interact with next based on the existing interaction records. The recommendation model should be trained on the training set to obtain reasonable model parameters, select the best parameters of the model on the validation set data, and report the test results of the model on the test set.

[0047] A2. Using the interaction records to train several different recommendation network models to obtain several recommendation models; the input of the recommendation model is the user and the product; the output of the recommendation model is the interaction score.

[0048] Dataset and pre-trained recommendation model preparation. The user set is recorded as U, the item set is recorded as I, and for implicit recommendation scenarios, Y is generally used. u,i ∈R |U|×|I| It represents the interaction between user u∈U and item i∈I. 1 means there is an interaction between the user and the item, that is, positive interaction, and 0 means there is no interaction record, that is, negative interaction. The dataset is usually divided into training set, validation set and test set.

[0049] Prepare multiple recommendation models f pre-trained on the dataset for integration. When given a user u and an item i, the recommendation model f can output an interaction score s ui =f(u,i), taking the integration of two models as an example, such as the two recommendation models f1 and f2.

[0050] Here is a brief introduction to the recommendation process of the recommendation model. When making a recommendation to a user, all the items to be recommended and the user's representation are input into the recommendation model. The recommendation model calculates the score of each item and the user one by one. For example, user 1's score for item 1 is 0.6, and user 1's score for item 2 is 0.9. Relatively speaking, the model is more inclined to recommend item 2 to user 1. After calculating the user's scores for all items, the items are sorted according to the scores, and the ones with the highest scores are recommended to the user. The specific number of recommendations is not the key to the method. For example, the top 10 or 20 items with the highest scores can be recommended to the user. It is just that a fixed number will be used when calculating the performance of the model, such as how many of the top 20 (or other number) recommended items are items that the user really likes, so as to quantify the recommendation effect of the model.

[0051] A3. Probability correction is performed on each of the recommended models to obtain several corrected models.

[0052] In this embodiment, a Gaussian correction function or a gamma correction function is used to perform probability correction on each of the recommended models to obtain a plurality of corrected models.

[0053] Wherein, each of the recommendation models is probability corrected using a Gaussian correction function to obtain several correction models, specifically including:

[0054] By formula Map the interaction score output by the recommendation model; where a, b, c are learnable parameters, s is the interaction score output by the recommendation model, and σ is the Sigmoid function. is the score after mapping.

[0055] Specifically, in this embodiment, the Gaussian correction method is as follows:

[0056] For a binary classification model, if the model predicts 100 times and each time it considers the probability of class 1 to be p=0.3, then it is hoped that 30 times the true class is 1. Such a model is probability-corrected. This is not the case with general recommendation models. The scores output by the recommendation model do not have the meaning of probability. Therefore, this embodiment needs to use the probability correction function g φ Map prediction scores to corrected probabilities.

[0057] The probability correction function needs to be monotonically increasing and cannot change the order of the predicted scores before and after correction. In addition, the output of the probability correction function represents the probability and should satisfy g φ :

[0058] like Figure 4 As shown, Figure 4 The predicted score density diagrams of the recommendation model for positive samples and negative samples are plotted respectively, and it can be seen that the two have different distributions.

[0059] Therefore, it is assumed that the prediction scores of the recommendation model for positive interactions and negative interactions conform to different Gaussian distributions:

[0060]

[0061] in are the mean and variance of the two Gaussian distributions respectively, and the posterior probability is obtained according to the Bayesian formula.

[0062]

[0063] Among them, π0 and π1 are the prior probabilities of the two categories,

[0064] Therefore, the Gaussian correction function can be written as:

[0065]

[0066] Among them, φ = {a, b, c} is a learnable parameter, and σ is a Sigmoid function. This function accepts the predicted score as input and outputs the corrected probability, which is the probability that user u will interact with item i. In order to constrain the Gaussian correction function to be monotonic, it is necessary to satisfy:

[0067] 2as+b>0;

[0068] As long as max and minJust satisfy the above constraints.

[0069] In order to solve the parameters while satisfying the constraints, the model training is performed using the constrained logistic regression method. Since the recommendation data set generally contains only positive samples, it is necessary to use negative sampling technology in the training set to construct negative feedback data. The correction model is trained using the training set data and negative feedback data. The groundtruth of the positive feedback data is 1, and the groundtruth of the negative feedback data is 0. For the Gaussian correction model, during model training, it is necessary to ensure that the constraint 2as+b>0. To satisfy this, as long as S max and S min As long as the extreme value satisfies the constraint, the other values ​​also satisfy the condition.

[0070] Model ensemble requires each recommendation model to be individually probability corrected, and the output score of the model ensemble is obtained by adding the corrected scores. The probability correction process (for a single model):

[0071] Data preparation: The learning of parameters in probability correction depends on data training. The data here includes not only which items the user likes (positive samples) but also which items the user dislikes (negative samples). Therefore, negative sampling is required. A simple negative sampling method is to regard items that the user has interacted with as items that the user dislikes. There are also other negative sampling methods. The specific choice of negative sampling has little to do with this method.

[0072] Model training: Use constrained logistic regression to train parameters. The input data is the score predicted by the recommendation model. If the input score is a positive sample, the groundtruth is 1. If the input score is a predicted score of a negative sample, the groundtruth is 0.

[0073] After the probability correction model training is completed, the scores are corrected according to different probability correction methods to obtain the corrected scores.

[0074] Probability correction is performed on the two recommendation models f1 and f2 respectively to obtain two correction models g1 and g2.

[0075] In addition to using the Gaussian correction function, a gamma correction function may also be used, which is: Among them, φ = {a, b, c} is a learnable parameter, σ is the Sigmoid function, and s is the interaction score output by the recommendation model. is the corrected score.

[0076] Similarly, assume that the prediction scores of the recommendation model for positive and negative interactions conform to different Gaussian distributions:

[0077]

[0078] Where Γ is the gamma function, Represent the shape parameter and rate parameter of the two gamma distributions respectively, and the posterior probability is:

[0079]

[0080]

[0081] Among them, a=α1-α0, b=β0-β1, The gamma correction function can be written as:

[0082]

[0083] Where φ = {a, b, c} is a learnable parameter. Since the gamma distribution is defined on positive real numbers, it is necessary to move the interaction score predicted by the recommendation model to the positive real number domain:

[0084] s←ss min

[0085] where s min To minimize the interaction score, the gamma correction function described above can be used instead of the Gaussian correction function.

[0086] A4. Input the target user and the product to be recommended into several correction models to obtain several first interaction scores.

[0087] In this embodiment, the target user and all the products to be recommended are input into a plurality of the correction models; the correction model is used to calculate the interaction score of each product to be recommended for the target user to obtain a plurality of first interaction scores.

[0088] Using the obtained correction model, the sum of the results of probability correction of multiple recommendation models is used to replace the original interaction score as the recommendation basis of the final integrated model:

[0089]

[0090] in, represents the corrected interaction score, f1(u,i) and f2(u,i) represent the uncorrected recommendation models, and g1 and g2 represent the correction process.

[0091] S2. According to the interaction score set corresponding to each of the corrected models, all first interaction scores corresponding to each of the products to be recommended are added together to obtain multiple second interaction scores. That is, the scores of the models to be integrated are corrected and added together as the result of multi-model integration.

[0092] S3. Arrange the corresponding products to be recommended in order of priority according to the second interaction score, and recommend the products to be recommended arranged in order of priority to the target user.

[0093] According to the obtained second interaction score Recommending the k items with the highest scores to each user as the recommendation results, or recommending to users in descending order of scores, is a unified practice of the recommendation system. The higher the predicted score, the more the recommendation model believes that the item will be liked by the user.

[0094] In this embodiment, the corresponding products to be recommended are sorted according to the second interaction scores to obtain a sorting result; the first k products to be recommended in the sorting result are recommended to the target user, where k is a natural number.

[0095] In the prior art, there is no method for implementing model integration for recommendation models. By integrating multiple recommendation models through the probability correction method, the recommendation performance can be improved. Compared with the performance of the multiple integrated models, the integrated model has a better recommendation effect.

[0096] The scores output by different recommendation models are not comparable, so different models need to be aligned. The specific method is probability correction, which converts the original interaction score into the probability of user-item interaction, so that different models are aligned to this probability and comparable. In addition, this model integration method is applicable to different types of recommendation models and is universal. It has no specific requirements for the structure and method of the model. As long as the recommendation model can output the interaction score for any user-item pair, it can be integrated. Most recommendation models can do this.

[0097] The present application provides a product recommendation method, device, medium and product that integrates multiple recommendation models, the method comprising: obtaining interaction records between users and products; using the interaction records to train several different recommendation network models to obtain several recommendation models; the input of the recommendation model is the user and the product; the output of the recommendation model is the interaction score; probability correction is performed on each of the recommendation models to obtain several corrected models; the target user and the product to be recommended are input into several corrected models to obtain several first interaction scores; and the corresponding product to be recommended is recommended to the target user according to the first interaction score. The present application integrates multiple recommendation models through the probability correction method, so that the results of multiple models can be combined, and the view of each recommendation model can be comprehensively considered. Compared with the performance of the multiple integrated models, the integrated model has a better recommendation effect. Among them, the existing recommendation model and the model provided in this embodiment are compared. Figure 5 , Figure 6 and Figure 7 shown.

[0098] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a product recommendation method integrating multiple recommendation models is implemented.

[0099] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0100] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0101] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0102] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0104] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0105] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0106] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A product recommendation method integrating multiple recommendation models, characterized in that: The product recommendation method integrating multiple recommendation models includes: Input the target user and all the products to be recommended into several correction models to obtain the interaction score set corresponding to each correction model; the interaction score set includes multiple elements, and the element is the first interaction score corresponding to the target user and a product to be recommended; wherein the number of elements is the same as the number of products to be recommended, and different elements correspond to different products to be recommended; the correction model is a model obtained after probability correction of the recommendation model, and different correction models correspond to different recommendation models; the recommendation model is determined according to sample data and a recommendation neural network, and different recommendation models correspond to different recommendation neural networks; the sample data is the interaction record between the sample user and the product; the sample data is used to train the recommendation neural network to obtain the recommendation model; According to the interaction score set corresponding to each of the correction models, all first interaction scores corresponding to each of the to-be-recommended products are added together to obtain a plurality of second interaction scores; According to the second interaction score, the corresponding products to be recommended are arranged in order of priority, and the products to be recommended arranged in order of priority are recommended to the target user.

2. The product recommendation method integrating multiple recommendation models according to claim 1, characterized in that: The interaction record includes: historical data of the user's shopping on the e-commerce platform.

3. The product recommendation method integrating multiple recommendation models according to claim 1, characterized in that: Probability correction is performed on the recommendation model, specifically including: A Gaussian correction function or a gamma correction function is used to perform probability correction on each of the recommended models to obtain a plurality of corrected models.

4. The product recommendation method integrating multiple recommendation models according to claim 3, characterized in that: Each of the recommended models is probability corrected using a Gaussian correction function to obtain several correction models, specifically including: By formula Map the interaction score output by the recommendation model; where a, b, c are learnable parameters, s is the interaction score output by the recommendation model, and σ is the Sigmoid function. is the score after mapping, 2as+b>

0.

5. The product recommendation method integrating multiple recommendation models according to claim 3, characterized in that: The gamma correction function is: Among them, φ = {a, b, c} is a learnable parameter, σ is the Sigmoid function, and s is the interaction score output by the recommendation model. is the corrected score.

6. The product recommendation method integrating multiple recommendation models according to claim 1, characterized in that: The step of inputting the target user and all the products to be recommended into a plurality of the correction models to obtain a set of interaction scores corresponding to each correction model specifically includes: Inputting the target user and all the products to be recommended into several correction models; Each correction model is used to calculate the interaction score of each of the to-be-recommended products for the target user, to obtain a plurality of first interaction scores.

7. The product recommendation method integrating multiple recommendation models according to claim 1, characterized in that: According to the second interaction score, the corresponding to-be-recommended products are arranged in order of priority, and the to-be-recommended products arranged in order of priority are recommended to the target user, specifically including: Sort the corresponding recommended products according to the second interaction scores to obtain a sorting result; The first k recommended products in the ranking results are recommended to the target user, where k is a natural number.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the product recommendation method integrating multiple recommendation models as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the product recommendation method integrating multiple recommendation models described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the product recommendation method integrating multiple recommendation models described in any one of claims 1 to 7 is implemented.

Images