CN117726403A - Product recommendation method and device, storage medium and electronic equipment - Google Patents

Abstract

The application discloses a product recommendation method and device, a storage medium and electronic equipment, and relates to the technical field of computers. The method comprises the following steps: obtaining N target products, wherein the N target products are products to be recommended to a target object; acquiring a data information set based on N target products, wherein the data information set at least comprises: t pieces of historical behavior data, attribute information of each target product and attribute information of a target object; determining S prediction models, wherein the S prediction models at least comprise: a first prediction model, a second prediction model, and a third prediction model; and determining a target recommendation sequence for recommending N target products according to the data information set and the S prediction models, and recommending the N target products to the target object according to the target recommendation sequence. By the method and the device, the problem of low accuracy of recommending the product to the user in the related technology is solved.

Description

Recommended methods and devices for products, storage media and electronic equipment

Technical field

This application relates to the field of computer technology, specifically, to a product recommendation method and device, storage media and electronic equipment.

Background technique

In related technologies, machine learning models are generally used to score products and recommend products based on the scores. However, with the improvement of product richness and user behavior richness, traditional recommendation methods based on a single model (such as GBDT, DNN) are difficult to accurately characterize and capture users' interest preferences and changes in their interests, which will lead to User recommended products are less accurate.

To address the problem of low accuracy in recommending products to users in related technologies, no effective solution has yet been proposed.

Contents of the invention

The main purpose of this application is to provide a product recommendation method and device, storage media and electronic equipment, so as to solve the problem of low accuracy in recommending products to users in related technologies.

In order to achieve the above objectives, according to one aspect of the present application, a product recommendation method is provided. The method includes: obtaining N target products, wherein the N target products are products to be recommended to target objects, and N is a positive integer greater than 1; obtaining a data information set based on the N target products, wherein the The data information set at least includes: T historical behavior data, attribute information of each target product and attribute information of the target object. The T historical behavior data includes at least M objects for the N target products. Historical behavior data, the M objects include at least the target object, T and M are both positive integers greater than 1; S prediction models are determined, wherein the prediction model is used to predict and recommend the N targets Recommendation order of products, the S prediction models at least include: a first prediction model, a second prediction model and a third prediction model, the first prediction model at least includes a decision tree model, the second prediction model at least includes Factorization machine model, the third prediction model at least includes a combination of a multi-layer perceptron model and a factorization machine model, S is a positive integer greater than 1; based on the data information set and the S prediction models, Determine a target recommendation order for recommending the N target products, and recommend the N target products to the target object according to the target recommendation order.

Further, the T pieces of historical behavior data include at least: first historical behavior data, second historical behavior data and third historical behavior data. The first historical behavior data is the target object's history of the target product. Behavior data, the second historical behavior data is the historical behavior data of each of the M objects for the target product, and the third historical behavior data is the historical behavior data of each of the M objects for the target product. Based on the historical behavior data of each target product and the data information set and the S prediction models, determining the target recommendation sequence for recommending the N target products includes: combining the second historical behavior data and the third The historical behavior data is input into the first prediction model to predict and recommend the recommendation order of the N target products, and the first recommendation order is output; the first recommendation order, the second historical behavior data, and the third historical behavior are The data, the attribute information of each target product and the attribute information of the target object are input into the second prediction model to predict and recommend the recommendation order of the N target products, and output the second recommendation order; the first recommendation order, The first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object are input into the third prediction model to predict and recommend. The recommendation order of the N target products is output, and a third recommendation order is output; based on the second recommendation order and the third recommendation order, the target recommendation order for recommending the N target products is determined.

Further, based on the second recommendation order and the third recommendation order, determining the target recommendation order for recommending the N target products includes: determining a score corresponding to each target product based on the second recommendation order. value, N first score values are obtained; the score value corresponding to each target product is determined based on the third recommendation order, and N second score values are obtained; the weight sum corresponding to the second prediction model is determined respectively. The weight corresponding to the third prediction model; based on the N first score values, the N second score values, the weight corresponding to the second prediction model and the weight corresponding to the third prediction model , determine the target recommendation order for recommending the N target products.

Further, the S prediction models are obtained in the following manner: obtaining a target training sample set, wherein the target training sample set at least includes Q sample behavior data obtained in the historical process, attribute information of W sample products, and Attribute information of P sample objects, the Q sample behavior data at least includes the behavior data of the P sample objects for the W sample products, Q, W and P are all positive integers greater than 1; using all The target training sample set performs learning and training on each of the S original prediction models to obtain the S prediction models, wherein the S original prediction models at least include: the first original prediction model, the first The second original prediction model and the third original prediction model.

Further, using the target training sample set to perform learning and training on each of the S original prediction models, obtaining the S prediction models includes: learning and training the Q sample behaviors in the target training sample set The data is classified and processed to obtain the first sample behavior data, the second sample behavior data and the third sample behavior data, wherein the first sample behavior data is the behavior data of the sample object towards the sample product, so The second sample behavior data is the behavior data of each sample object among the P sample objects to the sample product, and the third sample behavior data is the behavior data of each sample object among the P sample objects to each sample product. Behavior data of sample products; using the second sample behavior data and the third sample behavior data to learn and train the first original prediction model among the S original prediction models to obtain the first prediction Model; obtain the prediction result output by the first prediction model, and use the prediction result, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the P Learn and train the second original prediction model among the S original prediction models with the attribute information of sample objects to obtain the second prediction model; use the prediction results, the first sample behavior data, The second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the attribute information of the P sample objects have a positive impact on the third original prediction model among the S original prediction models. The prediction model is learned and trained to obtain the third prediction model; based on the first prediction model, the second prediction model and the third prediction model, the S prediction models are obtained.

Further, obtaining the prediction result output by the first prediction model includes: obtaining the second sample behavior data and the third sample behavior data; inputting the second sample behavior data and the third sample behavior data The first prediction model predicts and recommends the order of recommendation of the W sample products, and outputs the predicted order of recommendation of the W sample products; and uses the predicted order of recommendation of the W sample products as the forecast result.

Further, obtaining the target training sample set includes: obtaining the portrait information of each sample object among the P sample objects, and determining the attribute information of the P sample objects based on the portrait information of each sample object, wherein, The attribute information of the P sample objects at least includes: ID information of each sample object and gender information of each sample object; obtain attribute information of the W sample products, wherein the attribute information of the W sample products including at least: ID information of each sample product and category information of each sample product; obtaining behavior data of the P sample objects for the W sample products; analyzing the W sample products based on the P sample objects. Behavior data of sample products, determine the Q sample behavior data; obtain the target training sample set based on the attribute information of the P sample objects, the attribute information of the W sample products and the Q sample behavior data .

In order to achieve the above object, according to another aspect of the present application, a product recommendation device is provided. The device includes: a first acquisition unit, used to acquire N target products, wherein the N target products are products to be recommended to target objects, and N is a positive integer greater than 1; a second acquisition unit, based on The N target products obtain a data information set, wherein the data information set at least includes: T historical behavior data, attribute information of each target product, and attribute information of the target object. The T historical behaviors The data includes at least M objects' historical behavior data for the N target products, the M objects include at least the target object, and both T and M are positive integers greater than 1; the first determination unit is used to Determine S prediction models, wherein the prediction models are used to predict the recommendation order of the N target products, and the S prediction models at least include: a first prediction model, a second prediction model and a third prediction model. , the first prediction model at least includes a decision tree model, the second prediction model at least includes a factorization machine model, and the third prediction model at least includes a model that is a combination of a multi-layer perceptron model and a factorization machine model, S is a positive integer greater than 1; the first processing unit is used to determine the target recommendation sequence for recommending the N target products based on the data information set and the S prediction models, and to recommend the target products to the target products based on the target recommendation sequence. The target object recommends the N target products.

Further, the T pieces of historical behavior data include at least: first historical behavior data, second historical behavior data and third historical behavior data. The first historical behavior data is the target object's history of the target product. Behavior data, the second historical behavior data is the historical behavior data of each of the M objects for the target product, and the third historical behavior data is the historical behavior data of each of the M objects for the target product. For historical behavior data of each target product, the first processing unit includes: a first processing module for inputting the second historical behavior data and the third historical behavior data into the first prediction model to predict and recommend the data. The recommended order of the N target products is output, and the first recommended order is output; the second processing module is used to combine the first recommended order, the second historical behavior data, the third historical behavior data, and each target product. The attribute information and the attribute information of the target object are input into the second prediction model to predict and recommend the recommendation order of the N target products, and output the second recommendation order; a third processing module is used to convert the first recommendation order , the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object are input into the third prediction model to predict and recommend. The recommendation order of the N target products is output, and a third recommendation order is output; a first determination module is configured to determine the target recommendation that recommends the N target products based on the second recommendation order and the third recommendation order. order.

Further, the first determination module includes: a first determination sub-module, used to determine the score value corresponding to each target product based on the second recommendation sequence, and obtain N first score values; a second determination sub-module. Module, used to determine the score value corresponding to each target product based on the third recommendation sequence, and obtain N second score values; a third determination sub-module, used to determine the weight corresponding to the second prediction model respectively The weight corresponding to the third prediction model; the fourth determination sub-module is used to determine the weight sum corresponding to the N first score values, the N second score values, and the second prediction model based on the The weight corresponding to the third prediction model determines the target recommendation order for recommending the N target products.

Further, the S prediction models are obtained through the following units: a third acquisition unit, used to obtain a target training sample set, wherein the target training sample set includes at least Q sample behavior data, W obtained in the historical process. Attribute information of sample products and attribute information of P sample objects. The Q sample behavior data at least includes the behavior data of the P sample objects for the W sample products. Q, W and P are all greater than a positive integer of 1; the first training unit is used to use the target training sample set to perform learning and training on each of the S original prediction models to obtain the S prediction models, wherein the S The original prediction model at least includes: a first original prediction model, a second original prediction model and a third original prediction model.

Further, the first training unit includes: a fourth processing module for classifying the Q sample behavior data in the target training sample set to obtain the first sample behavior data and the second sample behavior data. and third sample behavior data, wherein the first sample behavior data is the behavior data of the sample object toward the sample product, and the second sample behavior data is each sample in the P sample objects. The behavior data of the object to the sample product, the third sample behavior data is the behavior data of each sample object in the P sample objects to each sample product; the first training module is used to utilize the third sample object. The second sample behavioral data and the third sample behavioral data perform learning and training on the first original prediction model among the S original prediction models to obtain the first prediction model; a second training module is used to obtain the The prediction results output by the first prediction model are used, and the prediction results, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the P sample objects are used. The attribute information is used to learn and train the second original prediction model among the S original prediction models to obtain the second prediction model; a third training module is used to use the prediction results, the first sample Behavior data, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products, and the attribute information of the P sample objects are critical to the S original prediction models. The third original prediction model performs learning and training to obtain the third prediction model; a second determination module is used to obtain the S based on the first prediction model, the second prediction model and the third prediction model. a prediction model.

Further, the second training module includes: a first acquisition sub-module for acquiring the second sample behavior data and the third sample behavior data; a first output sub-module for converting the second sample into The behavior data and the third sample behavior data are input into the first prediction model to predict the recommendation order of the W sample products, and output the predicted recommendation order of the W sample products; the fifth determination sub-module is used The predicted recommendation order of the W sample products is used as the predicted result.

Further, the third acquisition unit includes: a fifth processing module, used to acquire the portrait information of each sample object among the P sample objects, and determine the P samples based on the portrait information of each sample object. Attribute information of the object, wherein the attribute information of the P sample objects at least includes: ID information of each sample object and gender information of each sample object; a first acquisition module, used to obtain the W sample products attribute information, wherein the attribute information of the W sample products at least includes: ID information of each sample product and category information of each sample product; the second acquisition module is used to obtain the P sample object pairs The behavior data of the W sample products; the third determination module is used to determine the Q sample behavior data based on the behavior data of the P sample objects for the W sample products; the third acquisition module is used The target training sample set is obtained based on the attribute information of the P sample objects, the attribute information of the W sample products, and the Q sample behavior data.

In order to achieve the above object, according to another aspect of the present application, a computer-readable storage medium is provided, the storage medium stores a program, wherein the program executes any one of the above product recommendation methods.

In order to achieve the above object, according to another aspect of the present application, an electronic device is provided. The electronic device includes one or more processors and a memory, and the memory is used to store one or more programs, wherein when the When the one or more programs are executed by the one or more processors, the one or more processors enable the one or more processors to implement the recommended method for any of the above-mentioned products.

Through this application, the following steps are adopted: obtain N target products, where N target products are products to be recommended to target objects, and N is a positive integer greater than 1; obtain a data information collection based on N target products, where, data The information collection at least includes: T pieces of historical behavior data, attribute information of each target product, and attribute information of the target object. The T pieces of historical behavior data include at least M objects' historical behavior data for N target products, and M objects. includes at least the target object, and both T and M are positive integers greater than 1; determine S prediction models, where the prediction models are used to predict the recommendation order of N target products, and the S prediction models at least include: the first prediction model, a second prediction model and a third prediction model, the first prediction model at least includes a decision tree model, the second prediction model at least includes a factor decomposition machine model, and the third prediction model at least includes a combination of a multi-layer perceptron model and a factor decomposition machine model. In the latter model, S is a positive integer greater than 1; based on the data information collection and S prediction models, the target recommendation sequence for recommending N target products is determined, and N target products are recommended to the target object according to the target recommendation sequence, solving the related problem The problem in technology of low accuracy in recommending products to users. By obtaining multiple target products to be recommended to users, and obtaining a data information collection based on the multiple target products, and then determining the target recommendation sequence for recommending multiple target products based on the data information collection and multiple prediction models, and based on the target recommendation sequence Recommend multiple target products to users, thereby avoiding the problem of low accuracy in recommending products to users in related technologies, thereby achieving the effect of improving the accuracy of recommending products to users.

Description of the drawings

The drawings that form a part of this application are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an improper limitation of this application. In the attached picture:

Figure 1 is a flowchart 1 of a product recommendation method provided according to an embodiment of the present application;

Figure 2 is a flow chart 2 of a product recommendation method provided according to an embodiment of the present application;

Figure 3 is a flow chart 3 of a product recommendation method provided according to an embodiment of the present application;

Figure 4 is a schematic diagram of a product recommendation device provided according to an embodiment of the present application;

Figure 5 is a schematic diagram of an electronic device provided according to an embodiment of the present application.

Detailed ways

It should be noted that, as long as there is no conflict, the embodiments and features in the embodiments of this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

In order to enable those in the technical field to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only These are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts should fall within the scope of protection of this application.

It should be noted that the terms "first", "second", etc. in the description and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that data so used may be interchanged where appropriate for the embodiments of the application described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus that encompasses a series of steps or units and need not be limited to those explicitly listed. Those steps or elements may instead include other steps or elements not expressly listed or inherent to the process, method, product or apparatus.

It should be noted that the relevant information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analysis data, etc.) involved in this disclosure are authorized by the user. Or information and data fully authorized by all parties. For example, there is an interface between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned users or institutions through the interface, and after receiving the consent information fed back by the aforementioned users or institutions, obtain the relevant information. information.

For the convenience of description, some nouns or terms involved in the embodiments of this application are described below:

DNN: DNN is the abbreviation of Deep Neural Network, which is an artificial neural network composed of multiple layers of neurons.

The present invention will be described below with reference to preferred implementation steps. Figure 1 is a flow chart 1 of a product recommendation method provided according to an embodiment of the present application. As shown in Figure 1, the method includes the following steps:

Step S101: Obtain N target products, where the N target products are products to be recommended to the target object, and N is a positive integer greater than 1.

For example, the above-mentioned target object may be a user, and the above-mentioned N target products may be multiple products to be recommended to the user.

Step S102: Obtain a data information set based on N target products, where the data information set at least includes: T pieces of historical behavior data, attribute information of each target product, and attribute information of the target object. The T pieces of historical behavior data include at least Historical behavioral data of M objects for N target products. The M objects include at least the target object. Both T and M are positive integers greater than 1.

For example, the above-mentioned T pieces of historical behavior data may include the historical behavior data of the user (the above-mentioned target object), including product exposure, clicks, additional purchases, purchase data, etc., and may also include the historical behavior data of different users on the same product in the historical process. Clicks, additional purchases, transaction frequency data, etc. and different users' clicks, additional purchases, transaction frequency data, etc. for different products in the historical process. The above-mentioned attribute information of each target product may include product ID (Identification, identification) identification and category information, etc.; the above-mentioned attribute information of the target object may include user ID (Identification, identification) identification, gender information, etc.

Step S103: Determine S prediction models, where the prediction models are used to predict the recommendation order of N target products. The S prediction models at least include: a first prediction model, a second prediction model and a third prediction model. The first prediction model The prediction model at least includes a decision tree model, the second prediction model at least includes a factorization machine model, the third prediction model at least includes a model that is a combination of a multi-layer perceptron model and a factorization machine model, and S is a positive integer greater than 1.

For example, the above-mentioned first prediction model can be a gradient boosting decision tree (GBDT) prediction model; the above-mentioned second prediction model can be a factorization machine (FM) prediction model; the above-mentioned third prediction model can be It is a model combining multi-layer perceptron and factorization machine.

Step S104: Determine a target recommendation order for recommending N target products based on the data information set and S prediction models, and recommend N target products to the target object according to the target recommendation order.

For example, a model that combines a gradient boosting decision tree (GBDT) prediction model (the above-mentioned first prediction model), a factorization machine (FM) prediction model (the above-mentioned second prediction model), a multi-layer perceptron and a factorization machine (the above-mentioned third prediction model) are fused to obtain a fusion model. Then, the acquired multiple historical behavioral data, user ID identification and gender information, as well as product ID identification and category information can be input into the fusion model, and recommendations can be made to the user (the above-mentioned target object) through the prediction of the fusion model The recommendation sequence of multiple products can then recommend multiple products to be recommended to the user according to the recommendation sequence.

Through the above-mentioned steps S101 to S104, multiple target products to be recommended to the user are obtained, a data information set is obtained based on the multiple target products, and then a goal of recommending multiple target products is determined based on the data information set and multiple prediction models. recommendation sequence, and recommends multiple target products to users based on the target recommendation sequence, thereby avoiding the problem of low accuracy in recommending products to users in related technologies, thereby achieving the effect of improving the accuracy of recommending products to users.

Optionally, in the product recommendation method provided by the embodiment of the present application, obtaining the target training sample set includes: obtaining the portrait information of each sample object among the P sample objects, and determining P based on the portrait information of each sample object. Attribute information of sample objects, wherein the attribute information of P sample objects at least includes: ID information of each sample object and gender information of each sample object; obtain attribute information of W sample products, where, W samples The attribute information of the product at least includes: ID information of each sample product and category information of each sample product; obtaining behavior data of P sample objects on W sample products; based on the behavior of P sample objects on W sample products data, determine Q sample behavioral data; obtain the target training sample set based on the attribute information of P sample objects, the attribute information of W sample products, and Q sample behavioral data.

For example, you can first obtain the user's user portrait information on the platform, and then obtain the user's ID identification information and gender information based on the user's user portrait information. Then the ID identification and category information of the product can be obtained; and then the behavioral data of multiple sample objects for multiple sample products can be obtained, and these behavioral data can be the sample objects' clicks, additional purchases, transaction frequency and other data on the sample products. The obtained ID information and gender information of the user, the ID identification and category information of the product, as well as the sample subject's clicks, additional purchases, transaction frequency and other data on the sample product can be summarized together as used to train the prediction model training sample set.

Through the above solution, the training sample set for training the prediction model can be easily obtained.

Optionally, in the product recommendation method provided by the embodiment of the present application, S prediction models are obtained in the following manner: obtaining a target training sample set, where the target training sample set includes at least Q sample behaviors obtained in the historical process. data, attribute information of W sample products and attribute information of P sample objects. Q sample behavior data includes at least the behavior data of P sample objects for W sample products. Q, W and P are all positive values greater than 1. Integer; use the target training sample set to learn and train each of the S original prediction models to obtain S prediction models. Among them, the S original prediction models include at least: the first original prediction model, the second original prediction model, and the second original prediction model. prediction model and the third original prediction model.

For example, you can use the user's ID identification information and gender information in the training sample set, the ID identification and category information of the product, as well as the sample object's clicks, additional purchases, transaction frequency and other data on the sample product to learn multiple original prediction models. training, and can obtain the trained gradient boosting decision tree (GBDT) prediction model, factorization machine (FM) prediction model and multi-layer perceptron and factorization machine combination model, and then the trained gradient boosting decision tree ( GBDT) prediction model, factorization machine (FM) prediction model and a model that combines multi-layer perceptron and factorization machine as the above S prediction models.

Through the above solution, the original prediction model can be easily learned and trained using the training sample set, and the trained model can be obtained quickly and accurately.

Optionally, in the product recommendation method provided by the embodiment of the present application, obtaining the prediction result output by the first prediction model includes: obtaining the second sample behavior data and the third sample behavior data; combining the second sample behavior data and the third sample behavior data. The sample behavior data is input into the first prediction model to predict the recommendation order of W sample products, and output the predicted recommendation order of W sample products; the predicted recommendation order of W sample products is used as the prediction result.

For example, the above-mentioned second sample behavior data can be different sample objects' clicks, additional purchases, transaction frequency data, etc. on the same sample product; the above-mentioned third sample behavior data can be different sample objects' clicks on different sample products. , additional purchases, transaction frequency data, etc.; the above-mentioned first prediction model can be a gradient boosting decision tree (GBDT) prediction model. For example, the data of clicks, additional purchases, and transaction frequency of different sample objects on the same sample product, and the data of clicks, additional purchases, and transaction frequency of the same sample object on different sample products can be input into the gradient boosting decision tree ( GBDT) prediction model, then the predicted recommendation order for recommending multiple sample products can be output, and the output predicted recommendation order for recommending multiple sample products can be used as the above prediction result.

In summary, by inputting user clicks and purchase data on products into the gradient boosting decision tree prediction model, the order of recommended products can be quickly and accurately predicted.

Optionally, in the product recommendation method provided by the embodiment of the present application, using the target training sample set to learn and train each of the S original prediction models, obtaining the S prediction models includes: The concentrated Q sample behavior data are classified and processed to obtain the first sample behavior data, the second sample behavior data and the third sample behavior data. Among them, the first sample behavior data is the behavior data of the sample object towards the sample product, and the The second sample behavior data is the behavior data of each sample object among the P sample objects for the sample product, and the third sample behavior data is the behavior data of each sample object among the P sample objects for each sample product; using the second The sample behavior data and the third sample behavior data learn and train the first original prediction model among the S original prediction models to obtain the first prediction model; obtain the prediction result output by the first prediction model, and use the prediction result and the second sample The behavioral data, the third sample behavioral data, the attribute information of W sample products and the attribute information of P sample objects are learned and trained on the second original prediction model among the S original prediction models to obtain the second prediction model; the prediction results are used , the first sample behavior data, the second sample behavior data, the third sample behavior data, the attribute information of W sample products and the attribute information of P sample objects are used to learn the third original prediction model among the S original prediction models. After training, a third prediction model is obtained; based on the first prediction model, the second prediction model and the third prediction model, S prediction models are obtained.

For example, the above-mentioned first sample behavior data can be the sample object's exposure, clicks, additional purchases, purchase data, etc. for the same sample product in the historical process; the above-mentioned second sample behavior data can be the historical process of different sample objects. Clicks, additional purchases, transaction frequency data, etc. on the same sample product; the above-mentioned third sample behavior data can be different sample objects' clicks, additional purchases, transaction frequency data, etc. on different sample products in the historical process; the above-mentioned The first original prediction model can be an untrained gradient boosting decision tree (GBDT) prediction model; the above-mentioned second original prediction model can be an untrained factorization machine (FM) prediction model; the above-mentioned third original prediction The model can be a combination of an untrained multi-layer perceptron and a factorization machine; the above-mentioned first prediction model can be a trained gradient boosting decision tree (GBDT) prediction model; the above-mentioned second prediction model can be a trained Factorization machine (FM) prediction model; the above-mentioned third prediction model can be a model that is a combination of a trained multi-layer perceptron and a factorization machine.

For example, you can first use data on clicks, additional purchases, and transaction frequencies of the same sample product by different sample objects in the historical process, and data on clicks, additional purchases, and transaction frequencies of different sample objects on different sample products in the historical process. Use the data to learn and train the untrained Gradient Boosting Decision Tree (GBDT) prediction model to obtain the trained Gradient Boosting Decision Tree (GBDT) prediction model; then obtain the output of the trained Gradient Boosting Decision Tree (GBDT) prediction model. Predicted recommendation order of multiple sample products, and then you can use the well-trained gradient boosting decision tree (GBDT) prediction model output to predict the recommended order of multiple sample products, different sample objects in the historical process for the same sample Product clicks, additional purchases, transaction frequency data, etc., as well as different sample objects' clicks, additional purchases, transaction frequency data, etc. for different sample products in the historical process, sample product ID identification and category information, etc. and user ID Identify and gender information, etc., learn and train the untrained factorization machine (FM) prediction model to obtain the trained factorization machine (FM) prediction model; then use the trained gradient boosting decision tree (GBDT) prediction model output Predictive recommendation of the order of recommendation of multiple sample products, exposure, clicks, additional purchases, and purchase data of the same sample product by sample objects in the historical process, and clicks, additions, and purchases of the same sample product by different sample objects in the historical process. Purchase and transaction frequency data, etc., as well as the clicks, additional purchases, transaction frequency data of different sample objects on different sample products in the historical process, the ID identification and category information of the sample products, and the user ID identification and gender information, etc. Learn and train the model that is a combination of an untrained multi-layer perceptron and a factorization machine, and obtain a model that is a combination of a trained multi-layer perceptron and a factorization machine. Then the model combining the trained gradient boosting decision tree (GBDT) prediction model, the trained factorization machine (FM) prediction model, and the trained multi-layer perceptron and factorization machine can be used as the above S prediction models.

Through the above solution, the trained gradient boosting decision tree (GBDT) prediction model, the trained factorization machine (FM) prediction model, and the trained multi-layer perceptron and factorization machine combination model can be quickly and accurately obtained.

Figure 2 is a flowchart 2 of the product recommendation method provided according to the embodiment of the present application. As shown in Figure 2, in the product recommendation method provided by the embodiment of the present application, T pieces of historical behavior data at least include: the first historical behavior data, the second historical behavior data and the third historical behavior data. The first historical behavior data is the historical behavior data of the target object towards the target product, and the second historical behavior data is the historical behavior of each of the M objects towards the target product. Data, the third historical behavior data is the historical behavior data of each of the M objects for each target product. Based on the data information collection and S prediction models, the target recommendation sequence for recommending the N target products is determined, including:

Step S201, input the second historical behavior data and the third historical behavior data into the first prediction model to predict and recommend the recommendation order of N target products, and output the first recommendation order;

Step S202, input the first recommendation order, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object into the second prediction model to predict and recommend the recommendation order of N target products, and output Second recommendation order;

Step S203, input the first recommendation sequence, the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object into the third prediction model to predict and recommend N targets. Product recommendation order, output the third recommendation order;

Step S204: Based on the second recommendation order and the third recommendation order, determine the target recommendation order for recommending N target products.

For example, the above-mentioned first historical behavior data can be the historical behavior data of the user (the above-mentioned target object), including product exposure, clicks, additional purchases, purchase data, etc.; the above-mentioned second historical behavior data can be the historical behavior data of different users in the historical process. Clicks, additional purchases, transaction frequency data, etc. on the same product; the above third historical behavior data can be different users' clicks, additional purchases, transaction frequency data, etc. on different products in the historical process; each of the above target products The attribute information may include the ID identification and category information of each product; the attribute information of the target object may include the user ID identification, gender information, etc. The above-mentioned first prediction model can be a trained gradient boosting decision tree (GBDT) prediction model; the above-mentioned second prediction model can be a trained factorization machine (FM) prediction model; the above-mentioned third prediction model can be a trained A good model combining multilayer perceptron and factorization machine.

For example, the data of clicks, additional purchases, and transaction frequencies of different users on the same product in the historical process, as well as the data of clicks, additional purchases, and transaction frequencies of different users on different products in the historical process, can be input into the trained gradient. Boost the decision tree (GBDT) prediction model, and output the prediction output results of the gradient boosting decision tree (GBDT) prediction model (the first recommended order above); then you can use the prediction output results of the gradient boosting decision tree (GBDT) prediction model, The data of clicks, additional purchases, and transaction frequency of different users on the same product in the historical process, as well as the data of clicks, additional purchases, and transaction frequency of different users on different products in the historical process, as well as the ID identification and category of each product. The item information, user ID identification and gender information, etc. are input into the trained factorization machine (FM) prediction model, and the prediction output result of the factorization machine (FM) prediction model is output (the second recommendation sequence mentioned above); Combine the prediction output results of the gradient boosting decision tree (GBDT) prediction model, the historical behavior data of users (the above-mentioned target objects), including product exposure, clicks, additional purchases, purchase data, etc., as well as the views of different users on the same product in the historical process. Clicks, additional purchases, transaction frequency data, etc., as well as different users' clicks, additional purchases, transaction frequency data on different products in the historical process, etc., as well as the ID identification and category information of each product, and the user's ID identification and Gender information, etc. are input into the trained model combining multi-layer perceptron and factor decomposition machine, and the predicted output result of the model combining multi-layer perceptron and factor decomposition machine is output (the third recommended order above); then the model can be based on the factors The prediction output result of the factorization machine (FM) prediction model (the above-mentioned second recommendation sequence) and the prediction output result of the model combined with the multi-layer perceptron and factor decomposition machine (the above-mentioned third recommendation sequence) determine the final recommendation to the user. Recommended order of products.

Through the above solution, the fusion model can be used to quickly and accurately determine the order of recommendation of multiple products to the user.

Optionally, in the product recommendation method provided by the embodiment of the present application, based on the second recommendation order and the third recommendation order, determining the target recommendation order for recommending N target products includes: determining each target product based on the second recommendation order. According to the corresponding score value, N first score values are obtained; the score value corresponding to each target product is determined based on the third recommendation order, and N second score values are obtained; the weight sum corresponding to the second prediction model is determined respectively. The weight corresponding to the third prediction model; based on the N first score values, N second score values, the weight corresponding to the second prediction model and the weight corresponding to the third prediction model, determine the target recommendation for recommending N target products order.

For example, products can be recommended based on the score of each product. For example, you can first determine the score value corresponding to each product based on the prediction output result of the factorization machine (FM) prediction model (the second recommendation sequence mentioned above); and you can also determine the score value corresponding to each product based on the model of the combination of multi-layer perceptron and factorization machine. The prediction output result (the third recommendation sequence mentioned above) determines the score value corresponding to each product; then determines the corresponding weights of the factorization machine (FM) prediction model and the model combined with the multi-layer perceptron and factorization machine, and then you can The score corresponding to each product is determined based on the prediction output result of the factorization machine (FM) prediction model, and the score corresponding to each product is determined based on the prediction output result of the model combining the multi-layer perceptron and factorization machine. The corresponding weights of the value, factorization machine (FM) prediction model and the model combined with the multi-layer perceptron and factorization machine are used to obtain the recommendation sequence for recommending multiple products to the user.

Through the above solution, the order of recommendation of multiple products to the user can be determined quickly and accurately based on the score value of each product.

For example, this embodiment provides a product recommendation method and system based on a fusion model, and this embodiment belongs to the field of computer technology, and particularly relates to the field of product recommendation systems. Moreover, in this embodiment, a product recommendation method based on a fusion model is provided, which can improve the accuracy of recommendation.

For example, the main technical problem solved by this embodiment is to overcome the problem that traditional single models are difficult to accurately recommend in product recommendation systems, and provide a recommendation method based on multi-model fusion to improve the accuracy of recommendations, user shopping experience and platform products. sales volume.

For example, Figure 3 is flow chart 3 of a product recommendation method provided according to an embodiment of the present application. As shown in Figure 3, a product recommendation method based on multi-model fusion provided by this embodiment provides the following technical solution:

S1. Obtain the user's user portrait and historical behavior data on the platform, including product exposure, clicks, additional purchases, and purchase data;

S2. Based on the data obtained in S1, determine the training sample set and the test sample set. The data occurrence time node of the training sample set should be earlier than the test sample set; the sample set needs to include the user's unique ID identification on the platform to distinguish different users;

S3. Construct model training features based on the sample set determined in S2. Features include user behavior sequence features and product features. Feature types can be divided into more fine-grained categories based on the type of feature values. Specifically, such as product clicks, additional purchases, Transaction frequency, etc. can be called posterior statistical features; user ID identification, gender, product ID identification, category information, etc. can be called category features; different users' clicks, additional purchases, transaction frequencies, etc. on different products can be called display cross characteristics;

S4. Based on the training data set determined in S2, train multiple models with different types of training features in S3. Specifically, a gradient boosting decision tree (GBDT) prediction model is trained using posterior statistical features, explicit cross-features, and prediction output results of the improved decision tree prediction model using posterior statistical features, categorical features, explicit cross-features, and improved decision tree prediction models. A factorization machine (FM) prediction model; using posterior statistical features, categorical features, display cross features, user behavior sequence features, and the prediction output results of the gradient boosting decision tree prediction model to train a multi-layer perceptron and factorization machine combined models;

S5. Use the prediction model trained in S4 to predict the test sample set in S2 and recommend products to users in S1.

In addition, compared with the existing technology, this embodiment has the following advantages:

1. Divide the features of different attributes and train different models based on the features, making full use of the characteristics of various models and maximizing the role of the models;

2. Fusion of the results of various prediction models, fully absorbing the advantages of various models, greatly improving the accuracy of product recommendation compared to the recommendation method of a single model.

In summary, the product recommendation method provided by the embodiments of this application obtains N target products, where the N target products are products to be recommended to the target objects, and N is a positive integer greater than 1; based on the N target products, A collection of data information, wherein the collection of data information includes at least: T pieces of historical behavior data, attribute information of each target product, and attribute information of the target object. The T pieces of historical behavior data include at least M objects for N target products. Historical behavioral data, M objects include at least the target object, T and M are both positive integers greater than 1; determine S prediction models, where the prediction model is used to predict the recommendation order of N target products, and S prediction models including at least: a first prediction model, a second prediction model and a third prediction model. The first prediction model at least includes a decision tree model, the second prediction model at least includes a factor decomposition machine model, and the third prediction model at least includes a multi-layer perceptron. The model after combining the model and the factor decomposition machine model, S is a positive integer greater than 1; based on the data information collection and S prediction models, determine the target recommendation sequence for recommending N target products, and recommend N to the target object based on the target recommendation sequence A target product solves the problem of low accuracy in recommending products to users in related technologies. By obtaining multiple target products to be recommended to users, and obtaining a data information collection based on the multiple target products, and then determining the target recommendation sequence for recommending multiple target products based on the data information collection and multiple prediction models, and based on the target recommendation sequence Recommend multiple target products to users, thereby avoiding the problem of low accuracy in recommending products to users in related technologies, thereby achieving the effect of improving the accuracy of recommending products to users.

It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and, although a logical sequence is shown in the flowchart, in some cases, The steps shown or described may be performed in a different order than here.

The embodiment of the present application also provides a product recommendation device. It should be noted that the product recommendation device of the embodiment of the present application can be used to execute the product recommendation method provided by the embodiment of the present application. The recommended device of the product provided by the embodiment of this application is introduced below.

Figure 4 is a schematic diagram of a product recommendation device provided according to an embodiment of the present application. As shown in Figure 4, the device includes: a first obtaining unit 401, a second obtaining unit 402, a first determining unit 403 and a first processing unit 404.

Specifically, the first acquisition unit 401 is used to acquire N target products, where the N target products are products to be recommended to the target object, and N is a positive integer greater than 1;

The second acquisition unit 402 is used to acquire a data information set based on N target products, where the data information set at least includes: T pieces of historical behavior data, attribute information of each target product and attribute information of the target object, T pieces of history The behavioral data includes at least M objects' historical behavioral data for N target products, the M objects include at least the target object, and both T and M are positive integers greater than 1;

The first determination unit 403 is used to determine S prediction models, where the prediction models are used to predict the recommendation order of N target products. The S prediction models at least include: a first prediction model, a second prediction model and a third prediction model. Prediction model, the first prediction model at least includes a decision tree model, the second prediction model at least includes a factor decomposition machine model, the third prediction model at least includes a combination of a multi-layer perceptron model and a factor decomposition machine model, S is greater than 1 positive integer;

The first processing unit 404 is configured to determine a target recommendation sequence for recommending N target products based on the data information set and S prediction models, and recommend N target products to the target object according to the target recommendation sequence.

In summary, the product recommendation device provided by the embodiment of the present application acquires N target products through the first acquisition unit 401, where the N target products are products to be recommended to the target object, and N is a positive integer greater than 1; The second acquisition unit 402 acquires a data information set based on N target products, wherein the data information set includes at least: T pieces of historical behavior data, attribute information of each target product, and attribute information of the target object, and at least T pieces of historical behavior data Including the historical behavior data of M objects for N target products, the M objects include at least the target object, and T and M are both positive integers greater than 1; the first determination unit 403 determines S prediction models, where the prediction model is For predicting and recommending the order of recommending N target products, the S prediction models at least include: a first prediction model, a second prediction model and a third prediction model. The first prediction model at least includes a decision tree model, and the second prediction model at least includes The factorization machine model, the third prediction model at least includes a combination of the multi-layer perceptron model and the factorization machine model, S is a positive integer greater than 1; the first processing unit 404 determines based on the data information set and S prediction models Recommend the target recommendation sequence of N target products, and recommend N target products to the target object according to the target recommendation sequence, which solves the problem of low accuracy in recommending products to users in related technologies. By obtaining multiple target products to be recommended to users, and obtaining a data information collection based on the multiple target products, and then determining the target recommendation sequence for recommending multiple target products based on the data information collection and multiple prediction models, and based on the target recommendation sequence Recommend multiple target products to users, thereby avoiding the problem of low accuracy in recommending products to users in related technologies, thereby achieving the effect of improving the accuracy of recommending products to users.

Optionally, in the product recommendation device provided by the embodiment of the present application, the T pieces of historical behavior data include at least: first historical behavior data, second historical behavior data and third historical behavior data, and the first historical behavior data is the target. The object's historical behavior data for the target product, the second historical behavior data is the historical behavior data of each of the M objects for the target product, and the third historical behavior data is the historical behavior data of each of the M objects for each target product. Historical behavior data, the first processing unit includes: a first processing module, used to input the second historical behavior data and the third historical behavior data into the first prediction model to predict and recommend the recommendation order of N target products, and output the first recommendation order ; The second processing module is used to input the first recommendation sequence, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object into the second prediction model to predict and recommend N target products. The recommendation sequence, outputs the second recommendation sequence; the third processing module is used to combine the first recommendation sequence, the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information and target of each target product The attribute information of the object is input into the third prediction model to predict the recommendation order of N target products and output the third recommendation order; the first determination module is used to determine the recommendation order of N target products based on the second recommendation order and the third recommendation order. Target recommendation sequence.

Optionally, in the product recommendation device provided by the embodiment of the present application, the first determination module includes: a first determination sub-module, used to determine the score value corresponding to each target product based on the second recommendation order, and obtain N The first score value; the second determination sub-module is used to determine the score value corresponding to each target product based on the third recommendation order, and obtain N second score values; the third determination sub-module is used to determine the first score value respectively. The weight corresponding to the second prediction model and the weight corresponding to the third prediction model; the fourth determination submodule is used to determine the weight based on N first score values, N second score values, the weight corresponding to the second prediction model and the third The weight corresponding to the prediction model determines the target recommendation sequence for recommending N target products.

Optionally, in the product recommendation device provided by the embodiment of the present application, S prediction models are obtained through the following units: a third acquisition unit, used to obtain a target training sample set, wherein the target training sample set at least includes The obtained Q sample behavior data, W sample product attribute information and P sample object attribute information, the Q sample behavior data at least includes the behavior data of P sample objects for W sample products, Q, W and P is a positive integer greater than 1; the first training unit is used to use the target training sample set to learn and train each of the S original prediction models to obtain S prediction models, among which, S original predictions The model at least includes: a first original prediction model, a second original prediction model and a third original prediction model.

Optionally, in the product recommendation device provided by the embodiment of the present application, the first training unit includes: a fourth processing module, used to classify Q sample behavior data in the target training sample set to obtain the first sample Behavior data, second sample behavior data and third sample behavior data, where the first sample behavior data is the behavior data of sample objects for sample products, and the second sample behavior data is each sample object pair among the P sample objects. The behavioral data of the sample product, the third sample behavioral data is the behavioral data of each sample object among the P sample objects for each sample product; the first training module is used to utilize the second sample behavioral data and the third sample behavioral data Learn and train the first original prediction model among the S original prediction models to obtain the first prediction model; the second training module is used to obtain the prediction results output by the first prediction model, and use the prediction results and the second sample behavior data , the third sample behavioral data, the attribute information of W sample products and the attribute information of P sample objects are learned and trained on the second original prediction model among the S original prediction models to obtain the second prediction model; the third training module, Used to use the prediction results, the first sample behavior data, the second sample behavior data, the third sample behavior data, the attribute information of W sample products and the attribute information of P sample objects to predict the third of the S original prediction models. The original prediction model undergoes learning and training to obtain a third prediction model; the second determination module is used to obtain S prediction models based on the first prediction model, the second prediction model and the third prediction model.

Optionally, in the product recommendation device provided by the embodiment of the present application, the second training module includes: a first acquisition sub-module, used to acquire the second sample behavior data and the third sample behavior data; a first output sub-module, Used to input the second sample behavior data and the third sample behavior data into the first prediction model to predict the recommendation order of W sample products, and output the predicted recommendation order of W sample products; the fifth determination sub-module is used to The predicted recommendation order of W sample products is used as the prediction result.

Optionally, in the product recommendation device provided by the embodiment of the present application, the third acquisition unit includes: a fifth processing module, used to acquire the portrait information of each sample object among the P sample objects, and based on each sample object The portrait information of P sample objects is determined to determine the attribute information of P sample objects, wherein the attribute information of P sample objects at least includes: the ID information of each sample object and the gender information of each sample object; the first acquisition module is used to obtain Attribute information of W sample products, where the attribute information of W sample products at least includes: ID information of each sample product and category information of each sample product; the second acquisition module is used to obtain P sample object pairs Behavioral data of W sample products; the third determination module is used to determine Q sample behavior data based on the behavior data of W sample products based on P sample objects; the third acquisition module is used to determine Q sample behavior data based on the attributes of P sample objects Information, attribute information of W sample products and Q sample behavioral data to obtain the target training sample set.

The product recommendation device includes a processor and a memory. The above-mentioned first acquisition unit 401, second acquisition unit 402, first determination unit 403, first processing unit 404, etc. are all stored in the memory as program units, and the storage is executed by the processor. The above program units in the memory implement corresponding functions.

The processor contains a core, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the accuracy of product recommendations to users can be improved by adjusting kernel parameters.

Memory may include non-permanent memory in computer-readable media, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory includes at least one memory chip.

Embodiments of the present invention provide a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the recommended method for the product is implemented.

An embodiment of the present invention provides a processor, which is configured to run a program, wherein when the program is run, the recommended method for the product is executed.

As shown in Figure 5, an embodiment of the present invention provides an electronic device. The device includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the following steps: Obtain N Target products, wherein the N target products are products to be recommended to target objects, and N is a positive integer greater than 1; a data information set is obtained based on the N target products, wherein the data information set at least includes : T pieces of historical behavior data, attribute information of each target product and attribute information of the target object. The T pieces of historical behavior data include at least the historical behavior data of M objects for the N target products, and the The M objects at least include the target object, and T and M are both positive integers greater than 1; determine S prediction models, wherein the prediction models are used to predict and recommend the recommendation order of the N target products, and the The S prediction models at least include: a first prediction model, a second prediction model and a third prediction model, the first prediction model at least includes a decision tree model, the second prediction model at least includes a factor decomposition machine model, the The third prediction model at least includes a combination of a multi-layer perceptron model and a factorization machine model, and S is a positive integer greater than 1; based on the data information set and the S prediction models, determine and recommend the N targets The target recommendation order of products, and recommend the N target products to the target object according to the target recommendation order.

When the processor executes the program, the following steps are also implemented: the T pieces of historical behavior data include at least: first historical behavior data, second historical behavior data and third historical behavior data, and the first historical behavior data is the target object. For the historical behavior data of the target product, the second historical behavior data is the historical behavior data of each of the M objects towards the target product, and the third historical behavior data is the M Based on the historical behavior data of each object in the object for each target product, based on the data information set and the S prediction models, determining the target recommendation sequence for recommending the N target products includes: converting the second history Behavior data and the third historical behavior data are input into the first prediction model to predict and recommend the recommendation order of the N target products, and the first recommendation order is output; the first recommendation order, the second historical behavior data are , the third historical behavior data, the attribute information of each target product and the attribute information of the target object are input into the second prediction model to predict and recommend the recommendation order of the N target products, and output the second recommendation order; The first recommendation sequence, the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object are input into the third The three prediction models predict and recommend the recommendation order of the N target products, and output a third recommendation order; based on the second recommendation order and the third recommendation order, determine the target recommendation order for recommending the N target products. .

When the processor executes the program, the processor also implements the following steps: based on the second recommendation order and the third recommendation order, determining the target recommendation order to recommend the N target products includes: determining each of the N target products based on the second recommendation order. Score values corresponding to the target products are obtained, and N first score values are obtained; the score value corresponding to each target product is determined based on the third recommendation sequence, and N second score values are obtained; and the first score values are determined respectively. The weight corresponding to the second prediction model and the weight corresponding to the third prediction model; based on the N first score values, the N second score values, the weight corresponding to the second prediction model and the The weight corresponding to the third prediction model determines the target recommendation order for recommending the N target products.

When the processor executes the program, the following steps are also implemented: the S prediction models are obtained in the following manner: obtaining a target training sample set, wherein the target training sample set includes at least Q sample behavioral data, W obtained in the historical process Attribute information of sample products and attribute information of P sample objects. The Q sample behavior data at least includes the behavior data of the P sample objects for the W sample products. Q, W and P are all greater than A positive integer of 1; use the target training sample set to perform learning and training on each of the S original prediction models to obtain the S prediction models, wherein the S original prediction models at least include: The first original prediction model, the second original prediction model and the third original prediction model.

When the processor executes the program, the following steps are also implemented: using the target training sample set to perform learning and training on each of the S original prediction models. Obtaining the S prediction models includes: using the target training sample set to learn and train The Q sample behavior data are classified and processed to obtain the first sample behavior data, the second sample behavior data and the third sample behavior data, wherein the first sample behavior data is the sample object's response to the Behavior data of sample products, the second sample behavior data is the behavior data of each sample object in the P sample objects to the sample product, and the third sample behavior data is the behavior data of each of the P sample objects. Behavior data of each sample object for each sample product; use the second sample behavior data and the third sample behavior data to learn and train the first original prediction model among the S original prediction models , obtain the first prediction model; obtain the prediction result output by the first prediction model, and use the prediction result, the second sample behavior data, the third sample behavior data, and the W sample products The attribute information of the P sample objects and the attribute information of the P sample objects are learned and trained on the second original prediction model among the S original prediction models to obtain the second prediction model; using the prediction results, the The first sample behavior data, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the attribute information of the P sample objects compare to the S original prediction models. The third original prediction model in is learned and trained to obtain the third prediction model; based on the first prediction model, the second prediction model and the third prediction model, the S prediction models are obtained .

When the processor executes the program, the following steps are also implemented: Obtaining the prediction result output by the first prediction model includes: obtaining the second sample behavior data and the third sample behavior data; combining the second sample behavior data and the The third sample behavior data is input into the first prediction model to predict and recommend the recommendation order of the W sample products, and output the predicted recommendation order of the W sample products; and the predicted recommendation order of the W samples is output The recommended order of products is used as the predicted result.

The processor also implements the following steps when executing the program: Obtaining the target training sample set includes: obtaining the portrait information of each sample object among the P sample objects, and determining the P sample objects based on the portrait information of each sample object. attribute information, wherein the attribute information of the P sample objects at least includes: ID information of each sample object and gender information of each sample object; obtain attribute information of the W sample products, wherein, the The attribute information of the W sample products at least includes: the ID information of each sample product and the category information of each sample product; obtaining the behavior data of the P sample objects for the W sample products; based on the P The sample object determines the Q sample behavior data based on the behavior data of the W sample products; based on the attribute information of the P sample objects, the attribute information of the W sample products and the Q sample behavior data Obtain the target training sample set.

The devices in this article can be servers, PCs, PADs, mobile phones, etc.

This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program initialized with the following method steps: acquiring N target products, wherein the N target products are to be recommended to the target The product of the object, N is a positive integer greater than 1; a data information set is obtained based on the N target products, wherein the data information set at least includes: T pieces of historical behavior data, attribute information of each target product and all The attribute information of the target object, the T historical behavior data includes at least the historical behavior data of M objects for the N target products, the M objects include at least the target object, and both T and M are A positive integer greater than 1; determine S prediction models, wherein the prediction models are used to predict the recommendation order of the N target products, and the S prediction models at least include: a first prediction model, a second prediction model and a third prediction model, the first prediction model at least includes a decision tree model, the second prediction model at least includes a factorization machine model, and the third prediction model at least includes a multi-layer perceptron model and a factorization machine model. In the combined model, S is a positive integer greater than 1; based on the data information set and the S prediction models, the target recommendation order for recommending the N target products is determined, and the target recommendation order is determined according to the target recommendation order. The target object recommends the N target products.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: the T pieces of historical behavior data at least include: first historical behavior data, second historical behavior data and third historical behavior data, so The first historical behavior data is the historical behavior data of the target object towards the target product, the second historical behavior data is the historical behavior data of each of the M objects towards the target product, so The third historical behavior data is the historical behavior data of each of the M objects for each target product. Based on the data information set and the S prediction models, it is determined to recommend the N target products. The target recommendation sequence includes: inputting the second historical behavior data and the third historical behavior data into the first prediction model to predict and recommend the recommendation sequence of the N target products, and outputting the first recommendation sequence; A recommendation sequence, the second historical behavior data, the third historical behavior data, attribute information of each target product and attribute information of the target object are input into the second prediction model to predict and recommend the N target products. the recommendation sequence, output the second recommendation sequence; combine the first recommendation sequence, the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and The attribute information of the target object is input into the third prediction model to predict and recommend the recommendation order of the N target products, and a third recommendation order is output; based on the second recommendation order and the third recommendation order, the recommended order is determined. The target recommendation order of the N target products.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: based on the second recommendation sequence and the third recommendation sequence, determine the target recommendation sequence for recommending the N target products The method includes: determining the score value corresponding to each target product based on the second recommendation sequence, and obtaining N first score values; determining the score value corresponding to each target product based on the third recommendation sequence, and obtaining N first score values. The second score value; respectively determine the weight corresponding to the second prediction model and the weight corresponding to the third prediction model; based on the N first score values, the N second score values, and the The weight corresponding to the second prediction model and the weight corresponding to the third prediction model determine the target recommendation order for recommending the N target products.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: the S prediction models are obtained in the following manner: obtaining a target training sample set, wherein the target training sample set at least includes historical processes Q sample behavior data, attribute information of W sample products, and attribute information of P sample objects obtained from Behavioral data, Q, W and P are all positive integers greater than 1; use the target training sample set to learn and train each of the S original prediction models to obtain the S prediction models, where, The S original prediction models include at least: a first original prediction model, a second original prediction model and a third original prediction model.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: using the target training sample set to perform learning and training on each of the S original prediction models to obtain the S The prediction model includes: classifying the Q sample behavior data in the target training sample set to obtain the first sample behavior data, the second sample behavior data and the third sample behavior data, wherein the first sample behavior data This behavior data is the behavior data of the sample object towards the sample product, the second sample behavior data is the behavior data of each sample object in the P sample objects towards the sample product, and the third The sample behavior data is the behavior data of each of the P sample objects for each sample product; the second sample behavior data and the third sample behavior data are used to analyze the S original prediction models. Perform learning and training on the first original prediction model to obtain the first prediction model; obtain the prediction results output by the first prediction model, and use the prediction results, the second sample behavior data, and the third Three samples of behavioral data, attribute information of the W sample products and attribute information of the P sample objects are used to learn and train the second original prediction model among the S original prediction models, and obtain the second original prediction model. Prediction model; using the prediction results, the first sample behavior data, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the P sample objects The third original prediction model among the S original prediction models is learned and trained with the attribute information to obtain the third prediction model; based on the first prediction model, the second prediction model and the third prediction model Three prediction models are used to obtain the S prediction models.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: obtaining the prediction result output by the first prediction model includes: obtaining the second sample behavior data and the third sample behavior data ; Input the second sample behavior data and the third sample behavior data into the first prediction model to predict the recommendation order of the W sample products, and output the predicted recommendation order of the W sample products; The predicted recommendation order of the W sample products is used as the prediction result.

When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: obtaining the target training sample set includes: obtaining the portrait information of each sample object among the P sample objects, and based on each sample object portrait information, determine the attribute information of the P sample objects, wherein the attribute information of the P sample objects at least includes: ID information of each sample object and gender information of each sample object; obtain the W Attribute information of the W sample products, wherein the attribute information of the W sample products at least includes: ID information of each sample product and category information of each sample product; Obtain the P sample objects for the W sample products Behavior data of sample products; Determine the Q sample behavior data based on the behavior data of the P sample objects for the W sample products; Based on the attribute information of the P sample objects, the W sample products The target training sample set is obtained from the attribute information and the Q sample behavior data.

Those skilled in the art will understand that embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a use A device for implementing the functions specified in one process or processes of the flowchart and/or one block or blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions The device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device. Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

Memory may include non-volatile memory in computer-readable media, random access memory (RAM), and/or non-volatile memory in the form of read-only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media includes both persistent and non-volatile, removable and non-removable media that can be implemented by any method or technology for storage of information. Information may be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), and read-only memory. (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, Magnetic tape cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium can be used to store information that can be accessed by a computing device. As defined in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprises," or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements not only includes those elements, but also includes Other elements are not expressly listed or are inherent to the process, method, article or equipment. Without further limitation, an element qualified by the statement "comprises a..." does not exclude the presence of additional identical elements in the process, method, good, or device that includes the element.

Those skilled in the art will understand that embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The above are only examples of the present application and are not used to limit the present application. To those skilled in the art, various modifications and variations may be made to this application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application shall be included in the scope of the claims of this application.

Claims (10)

1. A method of recommending a product, comprising:

obtaining N target products, wherein the N target products are products to be recommended to a target object, and N is a positive integer greater than 1;

acquiring a data information set based on the N target products, wherein the data information set at least comprises: t pieces of historical behavior data, attribute information of each target product and attribute information of the target object, wherein the T pieces of historical behavior data at least comprise historical behavior data of M objects on the N target products, the M pieces of historical behavior data at least comprise the target objects, and T and M are positive integers larger than 1;

Determining S prediction models, wherein the prediction models are used for predicting and recommending the recommendation sequence of the N target products, and the S prediction models at least comprise: the system comprises a first prediction model, a second prediction model and a third prediction model, wherein the first prediction model at least comprises a decision tree model, the second prediction model at least comprises a factorizer model, the third prediction model at least comprises a model formed by combining a multi-layer perceptron model and the factorizer model, and S is a positive integer greater than 1;

and determining a target recommendation sequence for recommending the N target products according to the data information set and the S prediction models, and recommending the N target products to the target object according to the target recommendation sequence.

2. The method of claim 1, wherein the T historical behavior data comprises at least: the method comprises the steps of determining a target recommendation sequence of recommending N target products according to a data information set and S prediction models, wherein the data information set comprises first historical behavior data, second historical behavior data and third historical behavior data, the first historical behavior data is the historical behavior data of the target objects on the target products, the second historical behavior data is the historical behavior data of each of the M objects on the target products, the third historical behavior data is the historical behavior data of each of the M objects on the target products, and the target recommendation sequence of recommending the N target products comprises:

Inputting the second historical behavior data and the third historical behavior data into the first prediction model to predict and recommend the recommendation sequence of the N target products, and outputting a first recommendation sequence;

inputting the first recommendation sequence, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object into the second prediction model to predict and recommend the recommendation sequence of the N target products, and outputting a second recommendation sequence;

inputting the first recommendation sequence, the first historical behavior data, the second historical behavior data, the third historical behavior data, the attribute information of each target product and the attribute information of the target object into the third prediction model to predict and recommend the recommendation sequence of the N target products, and outputting a third recommendation sequence;

and determining the target recommendation order of recommending the N target products based on the second recommendation order and the third recommendation order.

3. The method of claim 2, wherein determining the target recommendation order to recommend the N target products based on the second recommendation order and the third recommendation order comprises:

Determining a score value corresponding to each target product based on the second recommendation sequence to obtain N first score values;

determining a score value corresponding to each target product based on the third recommendation sequence to obtain N second score values;

respectively determining the weight corresponding to the second prediction model and the weight corresponding to the third prediction model;

and determining the target recommendation sequence for recommending the N target products based on the N first score values, the N second score values, the weights corresponding to the second prediction model and the weights corresponding to the third prediction model.

4. The method of claim 1, wherein the S predictive models are obtained by:

obtaining a target training sample set, wherein the target training sample set at least comprises Q sample behavior data, attribute information of W sample products and attribute information of P sample objects obtained in a history process, the Q sample behavior data at least comprises behavior data of the P sample objects on the W sample products, and Q, W and P are positive integers larger than 1;

learning and training each original prediction model in S original prediction models by using the target training sample set to obtain the S prediction models, wherein the S original prediction models at least comprise: the first, second and third original prediction models.

5. The method of claim 4, wherein learning each of the S original predictive models using the set of target training samples comprises:

classifying the Q sample behavior data in the target training sample set to obtain first sample behavior data, second sample behavior data and third sample behavior data, wherein the first sample behavior data is behavior data of the sample objects on the sample products, the second sample behavior data is behavior data of each sample object of the P sample objects on the sample products, and the third sample behavior data is behavior data of each sample object of the P sample objects on each sample product;

learning and training the first original prediction model in the S original prediction models by using the second sample behavior data and the third sample behavior data to obtain the first prediction model;

obtaining a prediction result output by the first prediction model, and learning and training the second original prediction model in the S original prediction models by using the prediction result, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the attribute information of the P sample objects to obtain the second prediction model;

Learning and training the third initial prediction model in the S initial prediction models by using the prediction result, the first sample behavior data, the second sample behavior data, the third sample behavior data, the attribute information of the W sample products and the attribute information of the P sample objects to obtain a third prediction model;

and obtaining the S prediction models based on the first prediction model, the second prediction model and the third prediction model.

6. The method of claim 5, wherein obtaining the prediction result output by the first prediction model comprises:

acquiring the second sample behavior data and the third sample behavior data;

inputting the second sample behavior data and the third sample behavior data into the first prediction model to predict and recommend the recommendation sequence of the W sample products, and outputting the predicted recommendation sequence of the W sample products;

and taking the predicted recommendation sequence of recommending the W sample products as the prediction result.

7. The method of claim 4, wherein obtaining a set of target training samples comprises:

Acquiring the portrait information of each sample object in the P sample objects, and determining the attribute information of the P sample objects based on the portrait information of each sample object, wherein the attribute information of the P sample objects at least comprises: ID information of each sample object and sex information of each sample object;

acquiring attribute information of the W sample products, wherein the attribute information of the W sample products at least comprises: ID information of each sample product and category information of each sample product;

acquiring behavior data of the P sample objects on the W sample products;

determining the Q sample behavior data according to the behavior data of the P sample objects on the W sample products;

and acquiring the target training sample set based on the attribute information of the P sample objects, the attribute information of the W sample products and the Q sample behavior data.

8. A recommendation device for a product, comprising:

the first acquisition unit is used for acquiring N target products, wherein the N target products are products to be recommended to a target object, and N is a positive integer greater than 1;

the second obtaining unit is configured to obtain a data information set based on the N target products, where the data information set at least includes: t pieces of historical behavior data, attribute information of each target product and attribute information of the target object, wherein the T pieces of historical behavior data at least comprise historical behavior data of M objects on the N target products, the M pieces of historical behavior data at least comprise the target objects, and T and M are positive integers larger than 1;

The first determining unit is configured to determine S prediction models, where the prediction models are used to predict and recommend a recommendation order of the N target products, and the S prediction models at least include: the system comprises a first prediction model, a second prediction model and a third prediction model, wherein the first prediction model at least comprises a decision tree model, the second prediction model at least comprises a factorizer model, the third prediction model at least comprises a model formed by combining a multi-layer perceptron model and the factorizer model, and S is a positive integer greater than 1;

the first processing unit is used for determining a target recommendation sequence for recommending the N target products according to the data information set and the S prediction models, and recommending the N target products to the target object according to the target recommendation sequence.

9. A computer-readable storage medium, characterized in that the storage medium stores a program, wherein the program performs the recommendation method of the product according to any one of claims 1 to 7.

10. An electronic device comprising one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of recommending an article of manufacture of any of claims 1-7.