A recommendation method, device and computer-readable storage medium

Abstract

The present invention provides a recommendation method, apparatus and computer-readable storage medium, belonging to the field of computer technology. In this method, the server splits the first recommendation model into an input layer model and a perception layer model, and sends the perception layer model to the client, so that the client determines the products to be recommended for the target user according to the perception layer model; after the server obtains the historical operation data of the target user from the client, it inputs the historical operation data and the product feature information in the database into the second recommendation model for prediction processing to obtain a set of candidate products corresponding to the target user, and sends the set of candidate products to the client, so that the client determines the recommendation result according to the set of candidate products and the products to be recommended. The present invention can improve the recommendation efficiency of the client.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and particularly relates to a recommendation method, device, and computer-readable storage medium. Background Art

[0002] With the continuous development of Internet technology, how to capture the content required by users in real time and accurately and provide accurate recommendation services for users, such as product recommendations and personalized material recommendations, has become a popular research project.

[0003] Existing content recommendation methods generally determine recommended content based on machine learning models. However, for model-based recommendation methods, the recommendation accuracy is greatly affected by the complexity of the model. The more complex the model structure, the higher the recommendation accuracy. However, complex large-scale machine learning models have high requirements for the operating environment, and general user devices cannot provide a suitable operating environment for large-scale machine learning models, resulting in limited application scenarios for large-scale machine learning models and affecting the recommendation efficiency of user devices. Summary of the Invention

[0004] The present invention provides a recommendation method, device, and computer-readable storage medium, which can break through the limitations of the operating environment on large-scale machine learning models by running a lightweight perception layer model on the client side and improve the recommendation efficiency of the client side.

[0005] According to the first aspect of the present invention, a recommendation method is provided, which is applied to a server, and the method includes:

[0006] Split the first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client so that the client determines the products to be recommended for the target user according to the perception layer model;

[0007] Obtain the historical operation data of the target user from the client;

[0008] Input the historical operation data and product feature information in the database into the second recommendation model for prediction processing to obtain a set of candidate products corresponding to the target user;

[0009] Send the set of candidate products to the client so that the client determines the recommendation result according to the set of candidate products and the products to be recommended.

[0010] According to the second aspect of the present invention, another recommendation method is provided, which is applied to a client, and the method includes:

[0011] Determine the user feature information of the target user according to the real-time operation data of the target user;

[0012] Receive the perception layer model and the candidate product set sent by the server. The perception layer model is obtained by the server splitting the first recommendation model; the candidate product set is obtained by the server performing prediction processing on the historical operation data of the target user and the product feature information in the database based on the second recommendation model;

[0013] Input the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user;

[0014] Update the candidate product set according to the products to be recommended, and obtain and display the recommendation result.

[0015] According to the third aspect of the present invention, a recommendation device is provided, which is applied to the server. The device includes:

[0016] A model splitting module, configured to split the first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client, so that the client determines the products to be recommended for the target user according to the perception layer model;

[0017] A data acquisition module, configured to acquire the historical operation data of the target user from the client;

[0018] A candidate product determination module, configured to input the historical operation data and the product feature information in the database into the second recommendation model for prediction processing to obtain the candidate product set corresponding to the target user;

[0019] A candidate product sending module, configured to send the candidate product set to the client, so that the client determines the recommendation result according to the candidate product set and the products to be recommended.

[0020] According to the fourth aspect of the present invention, another recommendation device is provided, which is applied to the client. The device includes:

[0021] A feature determination module, configured to determine the user feature information of the target user according to the real-time operation data of the target user;

[0022] A data receiving module, configured to receive the perception layer model and the candidate product set sent by the server. The perception layer model is obtained by the server splitting the first recommendation model; the candidate product set is obtained by the server performing prediction processing on the historical operation data of the target user and the product feature information in the database based on the second recommendation model;

[0023] A recommended product determination module, configured to input the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user;

[0024] An update module, configured to update the candidate product set according to the product to be recommended, obtain a recommendation result and display it.

[0025] According to a fifth aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the recommendation method described in any one of the first aspects is implemented.

[0026] Regarding the prior art, the present invention has the following advantages:

[0027] In the recommendation method provided by the present invention, the server splits the first recommendation model into an input layer model and a perception layer model, and sends the perception layer model to the client, so that the client determines the product to be recommended for the target user according to the perception layer model; after the server obtains the historical operation data of the target user from the client, it inputs the historical operation data and the product feature information in the database into the second recommendation model for prediction processing, obtains the candidate product set corresponding to the target user, and sends the candidate product set to the client, so that the client determines the recommendation result according to the candidate product set and the product to be recommended. In the embodiment of the present invention, the perception layer model running in the client is a lightweight one. The perception layer model obtained by splitting the first recommendation model has low requirements for the running environment, and its model performance is the same as that of the first recommendation model, and it can run normally in the client, which is beneficial to improving the recommendation efficiency of the client.

[0028] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specific embodiments of the present invention are specifically given. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0030] Figure 1 is a flowchart of the steps of a recommendation method provided by an embodiment of the present invention;

[0031] Figure 2 is an application scenario architecture diagram of a recommendation method provided by an embodiment of the present invention;

[0032] Figure 3It is a flowchart of steps of another recommendation method provided by an embodiment of the present invention;

[0033] Figure 4 It is a flowchart of steps of yet another recommendation method provided by an embodiment of the present invention;

[0034] Figure 5 It is a block diagram of a recommendation device provided by an embodiment of the present invention;

[0035] Figure 6 It is a block diagram of another recommendation device provided by an embodiment of the present invention. Detailed implementation manners

[0036] Hereinafter, exemplary embodiments of the present invention will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be completely conveyed to those skilled in the art.

[0037] Figure 1 It is a flowchart of steps of a recommendation method provided by an embodiment of the present invention. As Figure 1 shown, the method may include:

[0038] Step 101, the server splits the first recommendation model into an input layer model and a perception layer model, and sends the perception layer model to the client.

[0039] Step 102, the server obtains historical operation data of the target user from the client.

[0040] Step 103, the server inputs the historical operation data and product feature information in the database into a second recommendation model for prediction processing to obtain a set of candidate products corresponding to the target user.

[0041] Step 104, the server sends the set of candidate products to the client.

[0042] Step 105, the client determines user feature information of the target user according to real-time operation data of the target user.

[0043] Step 106, the client receives the perception layer model and the set of candidate products sent by the server.

[0044] Step 107, the client inputs the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user.

[0045] Step 108: The client updates the candidate product set according to the product to be recommended, obtains the recommendation result and displays it.

[0046] The recommendation method provided by the embodiment of the present invention can be applied to the server and the client. Refer to Figure 2 , which shows the application scenario architecture diagram of the recommendation method provided by the embodiment of the present invention. As Figure 2 shown, the application scenario of the embodiment of the present invention may include a server 201 and a client 202. The server 201 and the client 202 are connected through a wireless or wired network. The server 201 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, and may also provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, cloud communications, network services, middleware services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. In an optional embodiment, the server 201 may be the background server of the application corresponding to the client 202.

[0047] The client 202 may be an application, a web page, etc. that provides recommendation / search services for users. The recommendation / search services may include at least one of the following: food, movies, hotels, tourism, online products, etc. It should be noted that the products (candidate products / products to be recommended) mentioned in the embodiments of the present invention include takeaway meals, products / product vouchers in offline stores, movie tickets, hotel room information, tourist attraction tickets, online products, etc. As long as the content that can be displayed in the form of products on the client is within the scope covered by the embodiments of the present invention. The client 202 may run on a terminal device, and the terminal device may include, but is not limited to, electronic devices such as mobile phones, intelligent robots, AI artificial customer service, mobile computers, and tablet computers.

[0048] In a possible application scenario, a user interacts with a client. The client determines and stores the user's characteristic information based on the user's operation data. The server splits the first recommendation model into an input layer model and a perception layer model, and obtains the historical operation data of the target user from the client through a network connection. The historical operation data of the target user and the product characteristic information in the database are input into the second recommendation model for prediction processing to determine the set of candidate products corresponding to the target user. Then, the server sends the set of candidate products and the split perception layer model to the client. After receiving the perception layer model and the set of candidate products sent by the server, the client determines the characteristic information of the target user based on the real-time operation data of the target user, and then inputs the user characteristic information and the product characteristic information of the target user into the perception layer model for prediction processing to obtain the products to be recommended for the target user. Finally, the client updates the set of candidate products according to the determined products to be recommended, obtains the recommendation result for the target user and displays it.

[0049] It should be noted that in the embodiments of the present invention, the operation data of the target user may include, but is not limited to, the user's browsing operation data, search operation data, click operation data, sharing operation data, comment content, order data, etc. The historical operation data of the target user may be the operation data of the target user within a historical time period. For example, the operation data of the target user in the past year or month. The historical operation data is mainly used to reflect the preferences, habits, etc. of the target user in the past period of time. The real-time operation data of the target user refers to the operation data of the target user within a recent period of time. For example, the operation data of the target user in the recent two days or one week, etc. Compared with the historical operation data, the real-time operation data can reflect the preferences, demands, etc. of the target user at the current stage for products.

[0050] The client determines the user characteristic information of the target user based on the real-time operation data of the target user, so as to determine the habits and preferences of the target user at the current stage according to the user characteristic information of the target user.

[0051] The first recommendation model in the present invention takes the user characteristic information and the product characteristic information as inputs and the product name / product identifier as the output, and is used to predict the products preferred by the user according to the user information and the product characteristic information. Among them, the first recommendation model may be a Deep Neural Networks (DNN) model, a Convolutional Neural Networks (CNN) model, a Recurrent Neural Networks (RNN) model, etc., which are not limited herein.

[0052] In an embodiment of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model. Specifically, the computational graph of the first recommendation model can be split. The part of the computational graph of the first recommendation model that is used to process the input data, extract the features of the input data, and generate the feature vector / feature matrix corresponding to the input data is used as the input layer model; the part of the computational graph of the first recommendation model that is used to analyze and calculate based on the feature vector / feature matrix corresponding to the input data and generate the calculation result is used as the perception layer model. After the server splits the first recommendation model to obtain the perception layer model and before sending the perception layer model to the client, the perception layer model can also be processed such as quantization and pruning, so as to compress the perception layer model, reduce the model complexity, and ensure that the client can run the perception layer model normally.

[0053] The second recommendation model in the embodiment of the present invention is used to determine the candidate product set corresponding to the target user according to the historical operation data of the target user and the product feature information. It should be noted that in the embodiment of the present invention, the first recommendation model and the second recommendation model can be the same model, or the second recommendation model can be used as the teacher model, and the first recommendation model can be obtained by compressing the second recommendation model, optimizing the magnitude and model performance of the first recommendation model, so that the magnitude of the perception layer model received by the client is smaller and the performance is better.

[0054] After the client receives the perception layer model and the candidate product set sent by the server, it first inputs the user feature information and product feature information of the target user into the perception layer model for prediction processing to obtain the products to be recommended corresponding to the target user. Specifically, the user feature information of the target user and the product feature information in the database can be input into the perception layer model at the same time. The perception layer model determines the matching degree between the user feature information and each product feature information, and determines the products to be recommended according to the matching degree. As an example, a threshold can be set in advance, and the products with a matching degree greater than the threshold with the target user are determined as the products to be recommended; or, according to the matching degree between each product and the target user, each product in the database can be sorted in descending order, and the products to be recommended are determined according to the sorting result. Finally, the client updates the candidate product set according to the determined products to be recommended, so as to further optimize the prediction result of the second recommendation model and obtain the final recommendation result for display.

[0055] In summary, in the embodiment of the present invention, the lightweight perception layer model runs in the client. The perception layer model obtained by splitting the first recommendation model has low requirements for the operating environment, and the model performance is the same as that of the first recommendation model, and it can run normally in the client, which is beneficial to improving the recommendation efficiency of the client.

[0056] Moreover, in the present invention, the historical operation data of the target user is first processed based on the second recommendation model in the server to determine the candidate product set corresponding to the target user. Then, the client processes the user feature information corresponding to the real-time operation data of the target user according to the perception layer model to obtain the products to be recommended, and uses the obtained products to be recommended to update the candidate product set to obtain the final recommendation result. In other words, the present invention predicts the preferred products of the target user twice through the perception layer model and the second recommendation model respectively, and adjusts the prediction result obtained based on the historical operation data of the target user with the prediction result obtained based on the real-time operation data of the target user, so that the finally obtained recommendation result more conforms to the current preferences and needs of the target user, and improves the accuracy of the recommendation result.

[0057] Figure 3 FIG. 4 is a flowchart of steps of another recommendation method provided by an embodiment of the present invention, which is applied to a server, as Figure 3 shown. The method may include:

[0058] Step 301: Split the first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client so that the client determines the products to be recommended for the target user according to the perception layer model.

[0059] Step 302: Obtain the historical operation data of the target user from the client.

[0060] Step 303: Input the historical operation data and the product feature information in the database into the second recommendation model for prediction processing to obtain the candidate product set corresponding to the target user.

[0061] Step 304: Send the candidate product set to the client so that the client determines the recommendation result according to the candidate product set and the products to be recommended.

[0062] It should be noted that the recommendation method provided by the embodiment of the present invention can be applied to a server, and the server may be Figure 2 the server 201 in the application scenario architecture diagram shown in FIG. 5. The server 201 may be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, cloud communications, network services, middleware services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. In an alternative embodiment, the server 201 may be the background server of the application corresponding to the client 202.

[0063] In an embodiment of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model, obtains historical operation data of a target user from the client through a network connection, inputs the historical operation data of the target user and product feature information in the database into the second recommendation model for prediction processing, and determines a set of candidate products corresponding to the target user. Then, the server sends the set of candidate products and the split perception layer model to the client. After receiving the perception layer model and the set of candidate products sent by the server, the client inputs the user feature information of the target user into the perception layer model for prediction processing, obtains the products to be recommended for the target user, updates the set of candidate products according to the determined products to be recommended, and obtains and displays the recommendation result for the target user.

[0064] Among them, the historical operation data of the target user obtained by the server from the client is determined by the client according to the operation data of the target user in the historical time period. For example, the target user interacts with the client, browses products, evaluates products, purchases products, etc. through the client. On the premise of following the corresponding data protection regulations and policies of the country where the client is located and obtaining the authorization of the target user, based on various operations performed by the target user through the client, the client collects the browsing operation data, search operation data, click operation data, sharing operation data, comment content, order data, etc. of the target user in the past year and in the past month to obtain the historical operation data of the target user.

[0065] The first recommendation model in the present invention takes user feature information and product feature information as inputs and product names / product identifiers as outputs, and is used to predict products preferred by users according to user feature information, product feature information, and user operation data. Among them, the first recommendation model can be a Deep Neural Networks (DNN) model, a Convolutional Neural Networks (CNN) model, a Recurrent Neural Networks (RNN) model, etc., which is not limited herein.

[0066] As an example, the server can split the computational graph of the first recommendation model. The part of the computational graph of the first recommendation model that is used to process input data, extract features of the input data, and generate a feature vector / feature matrix corresponding to the input data is used as the input layer model; the part of the computational graph of the first recommendation model that is used to analyze and calculate based on the feature vector / feature matrix corresponding to the input data and generate a calculation result is used as the perception layer model.

[0067] In the embodiment of the present invention, the second recommendation model is used to determine a candidate product set corresponding to a target user according to the historical operation data of the target user and the product feature information. It should be noted that in the embodiment of the present invention, the first recommendation model and the second recommendation model can be the same model, or the second recommendation model can be used as a teacher model, and the first recommendation model can be obtained by performing knowledge distillation on the second recommendation model to optimize the scale and model performance of the first recommendation model, so that the scale of the perception layer model received by the client is smaller and the performance is better.

[0068] In an alternative embodiment of the present invention, the step 301 of splitting the first recommendation model into an input layer model and a perception layer model and sending the perception layer model to the client includes:

[0069] Step S11: Split the first recommendation model into an input layer model and a perception layer model;

[0070] Step S12: Perform compression processing on the perception layer model to obtain a compressed perception layer model;

[0071] Step S13: Send the compressed perception layer model to the client.

[0072] In the embodiment of the present invention, after the server splits the first recommendation model to obtain the perception layer model and before sending the perception layer model to the client, compression processing can also be performed on the perception layer model, such as quantization, pruning, etc. on the perception layer model, to achieve sparsity and compression of the perception layer model, reduce the model complexity, shorten the inference latency of the perception layer model, so as to ensure that the perception layer model received by the client is a lightweight model and improve the running fluency and inference efficiency of the perception layer model on the client.

[0073] In an alternative embodiment of the present invention, the method further includes:

[0074] Step S21: Generate and store a feature embedding vector data table according to the input layer model;

[0075] Step S22: Configure a query interface for the feature embedding vector data table for the client to obtain perception layer input parameters based on query information and determine the products to be recommended for the target user according to the perception layer input parameters.

[0076] Wherein, the feature embedding vector data table includes vector information of user feature information and product feature information, and the input layer model is used to preprocess the user feature information and product feature information input by the client, obtain corresponding vector information according to the preprocessed user feature information and product feature information, and construct perception layer input parameters according to the obtained vector information.

[0077] It should be noted that, in the embodiments of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model. Among them, the input layer model is used to perform data preprocessing on the input user feature information and commodity feature information, such as data screening, data standardization, normalization, data augmentation, and so on. Exemplarily, performing data standardization (normalization) processing on the input user feature information and / or commodity feature information can scale the input user feature information and / or commodity feature information proportionally so that it falls into a small specific interval to ensure the convergence speed of the neural network; performing normalization processing on the input user feature information and / or commodity feature information can, to a certain extent, eliminate the influence of various reasons such as poor data quality or noise on the update of the model weights; performing data augmentation on the input user feature information and / or commodity feature information to increase the training data of the first recommendation model, avoid overfitting, and thus improve the accuracy of the first recommendation model. The output data of the input layer model is the input data of the perception layer model, that is, the perception layer input parameters in the present invention. The perception layer model analyzes and calculates the perception layer input parameters and outputs the prediction result for the target user.

[0078] Before the split, for the first recommendation model, its input data is the commodity feature information and the user feature information of the target user, and the output data is the commodity name / commodity identifier. After the server splits the first recommendation model, only the perception layer model is sent to the client, and the client cannot directly process the commodity feature information and user feature information through the perception layer model to obtain the recommended commodities for the target user. Therefore, in order to improve the prediction efficiency of the client for the recommended commodities for the target user, the embodiments of the present invention can generate a feature embedding vector data table based on the input layer model. The feature embedding vector data table records the vector information of the user feature information and the vector information of the commodity feature information. And a query interface for the feature embedding vector data table is configured in the server. When the client determines the recommended commodities for the target user, it can query the perception layer input parameters corresponding to the target user according to the user feature information and commodity feature information by accessing the query interface configured for the feature embedding vector data table in the server, so that there is no need to further process the user feature information and commodity feature information, and input the queried perception layer input parameters into the perception layer model to determine the recommended commodities for the target user.

[0079] In an alternative embodiment of the present invention, before step 301 of splitting the first recommendation model into an input layer model and a perception layer model, the method further includes:

[0080] Step S31: Train the second recommendation model based on a pre-acquired training sample set to obtain a trained second recommendation model;

[0081] Step S32: Compress the trained second recommendation model to obtain the first recommendation model.

[0082] In the embodiments of the present invention, the first recommendation model and the second recommendation model can be the same model. Alternatively, the second recommendation model can add more features or use a more complex network to optimize the prediction effect. To simplify the model processing process, the second recommendation model can be directly compressed to obtain the first recommendation model.

[0083] As an example, the second recommendation model can be used as a teacher model, and the trained second recommendation model can be processed by knowledge distillation to obtain the first recommendation model, optimizing the scale and performance of the first recommendation model, so that the scale of the perception layer model received by the client is smaller and the performance is better.

[0084] As another example, methods such as model pruning, quantization, and feature screening can be used to compress the second recommendation model. While trying to maintain the prediction effect, the model volume and complexity of the second recommendation model are compressed to obtain a lightweight first recommendation model.

[0085] In an alternative embodiment of the present invention, the splitting of the first recommendation model into an input layer model and a perception layer model in step 301 includes:

[0086] Step S41: Obtain the historical recommendation data of the target user;

[0087] Step S42: Adjust the first recommendation model according to the historical recommendation data to obtain an adjusted first recommendation model;

[0088] Step S43: Split the adjusted first recommendation model into an input layer model and a perception layer model.

[0089] In the embodiments of the present invention, a neural network model that is good at user recommendation in the art can be directly selected as the first recommendation model in the present invention, such as a deep neural network model, a convolutional neural network model, a recurrent neural network model, etc.

[0090] In order to further improve the accuracy of the first recommendation model, in the embodiments of the present invention, the first recommendation model can be adjusted according to the historical recommendation data of the target user to obtain the first recommendation model for the target user. Specifically, the historical recommendation data of the target user can be first obtained. The historical recommendation data can be the historical recommendation results determined by the client for the target user, or the products to be recommended for the target user determined by the client based on the perception layer model during the historical recommendation process before generating the current recommendation result, etc.; then, based on the historical recommendation data of the target user, the first recommendation model is optimized and trained, and the model parameters of the first recommendation model are adjusted according to the loss value during each round of training process until the loss value meets the model convergence condition to obtain the adjusted first recommendation model. Among them, the convergence condition can be that the loss value is less than a preset threshold, or the number of training times is greater than a preset number of training times, etc. Finally, the server splits the adjusted first recommendation model into an input layer model and a perception layer model. The perception layer model is a personalized model for the target user, and its prediction result is more in line with the personalized needs of the target user.

[0091] In summary, for the recommendation method provided by the embodiments of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model, and sends the perception layer model to the client. The perception layer model running in the client is lightweight. The perception layer model obtained by splitting the first recommendation model has low requirements for the running environment, and the model performance is the same as that of the first recommendation model, and it can run normally in the client, which is beneficial to improving the recommendation efficiency of the client.

[0092] Figure 4 It is a step flowchart of another recommendation method provided by the embodiments of the present invention, which is applied to the client, as Figure 4 shown, and the method may include:

[0093] Step 401, determine the user feature information of the target user according to the real-time operation data of the target user.

[0094] Step 402, receive the perception layer model and the candidate product set sent by the server. The perception layer model is obtained by the server splitting the first recommendation model; the candidate product set is obtained by the server performing prediction processing on the historical operation data of the target user and the product feature information in the database based on the second recommendation model.

[0095] Step 403, input the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user.

[0096] Step 404, update the candidate product set according to the products to be recommended to obtain the recommendation result and display it.

[0097] It should be noted that the recommendation method provided in the embodiments of the present invention can be applied to a client, and the client can be the client 202 in the application scenario architecture diagram shown below. The client 202 can be an application program, a web page, etc. that provides recommendation / search services for users. The recommendation / search services can include at least one of the following: food, movies, hotels, tourism, online goods, etc. It should be noted that the goods (candidate goods / to-be-recommended goods) mentioned in the embodiments of the present invention include takeout food, goods / goods vouchers in offline stores, movie tickets, room information in hotels, tickets for tourist attractions, online goods, etc. As long as the content can be displayed in the form of goods in the client, it belongs to the content covered by the embodiments of the present invention. The client 202 can run in a terminal device, and the terminal device can include, but is not limited to, electronic devices such as mobile phones, intelligent robots, AI artificial customer service, mobile computers, and tablet computers. Figure 2 The client determines the user characteristic information of the target user according to the real-time operation data of the target user. Specifically, the target user interacts with the client, browses goods, evaluates goods, purchases goods, etc. through the client. On the premise of following the corresponding data protection regulations and policies of the country where the client is located and obtaining the authorization given by the target user, the client collects the real-time operation data of the target user based on various operations performed by the target user through the client, and analyzes and processes the collected operation data to obtain the user characteristic information of the target user. Exemplarily, in the embodiments of the present invention, the real-time operation data of the target user may include, but is not limited to, the browsing data, user search data, user click data, user sharing data, user comment data, user order data, etc. of the user in a recent period of time, such as the recent two days or within a week. The real-time operation data of the target user can reflect the preferences, demands, etc. of the target user at the current stage for goods.

[0098] In the embodiments of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model, and obtains the historical operation data of the target user from the client through a network connection. The historical operation data of the target user and the product feature information in the database are input into the second recommendation model for prediction processing to determine the set of candidate goods corresponding to the target user. Then, the server sends the set of candidate goods and the split perception layer model to the client. It should be noted that in the embodiments of the present invention, the historical operation data of the target user is mainly used to reflect the preferences, habits, etc. of the target user in the past period of time.

[0099]

[0100] ​After the client receives the perception layer model and the candidate product set sent by the server, it inputs the user feature information and product feature information of the target user into the perception layer model for prediction processing, obtains the products to be recommended for the target user, and updates the candidate product set according to the determined products to be recommended, so as to obtain the recommendation result for the target user and display it.

[0101] As an example, the client determines whether there are products in the candidate product set that are the same as or similar to the products to be recommended. If so, the products in the candidate product set are retained; if not, the products to be recommended are added to the candidate product set to obtain the recommendation result for the target user.

[0102] As another example, the client can calculate the matching degree between each product in the candidate product set and the user feature information of the target user; then, replace the product in the candidate product set with a lower matching degree with the user feature information with the product to be recommended to obtain the recommendation result for the target user.

[0103] In the present invention, the historical operation data of the target user is first processed in the server based on the second recommendation model to determine the candidate product set corresponding to the target user, and then the client processes the user feature information corresponding to the real-time operation data of the target user according to the perception layer model to obtain the products to be recommended, and uses the obtained products to be recommended to update the candidate product set to obtain the final recommendation result. In other words, the present invention predicts the preferred products of the target user twice through the perception layer model and the second recommendation model respectively, and uses the prediction result obtained based on the real-time operation data of the target user to adjust the prediction result obtained based on the historical operation data of the target user, so that the finally obtained recommendation result more conforms to the current preferences and needs of the target user, and improves the accuracy of the recommendation result.

[0104] In an alternative embodiment of the present invention, the step of inputting the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user in step 403 includes:

[0105] Step S51: Access the query interface for the feature embedding vector data table in the server, and query the perception layer input parameters corresponding to the target user according to the feature information of the target user and the product feature information;

[0106] Step S52: Input the perception layer input parameters into the perception layer model for prediction processing to obtain the products to be recommended for the target user.

[0107] In an embodiment of the present invention, the server splits the first recommendation model into an input layer model and a perception layer model. Among them, the input data of the input layer model is the commodity feature information and the user feature information of the target user, and the output data is the input data of the perception layer model, that is, the perception layer input parameters in the present invention. The perception layer model analyzes and calculates the perception layer input parameters and outputs a prediction result for the target user.

[0108] Before the splitting, for the first recommendation model, its input data is the commodity feature information and the user feature information of the target user, and the output data is the commodity name / commodity identifier. After the server splits the first recommendation model, only the perception layer model is sent to the client, and the client cannot directly process the commodity feature information and the user feature information through the perception layer model to obtain the recommended commodities for the target user.

[0109] Therefore, to solve this problem and improve the prediction efficiency of the client for the recommended commodities for the target user, the present invention configures a query interface for the feature embedding vector data table in the server. The feature embedding vector data table records the vector information of the user feature information and the vector information of the commodity feature information. When determining the recommended commodities for the target user, the client can access the query interface configured for the feature embedding vector data table in the server, query the perception layer input parameters corresponding to the target user according to the user feature information and the commodity feature information, and thus there is no need to further process the user feature information and the commodity feature information. By inputting the queried perception layer input parameters into the perception layer model, the recommended commodities for the target user can be determined.

[0110] In summary, in an embodiment of the present invention, the perception layer model running in the client is lightweight. The perception layer model obtained by splitting the first recommendation model has low requirements for the operating environment, and its model performance is the same as that of the first recommendation model and can run normally in the client, which is beneficial to improving the recommendation efficiency of the client. Moreover, the present invention predicts the preferred commodities of the target user twice through the perception layer model and the second recommendation model respectively, and adjusts the prediction result obtained based on the historical operation data of the target user with the prediction result obtained based on the real-time operation data of the target user, so that the finally obtained recommendation result more conforms to the current preferences and needs of the target user and improves the accuracy of the recommendation result.

[0111] Figure 5 It is a block diagram of a recommendation device provided by an embodiment of the present invention, which is applied to the server, as Figure 5 shown. The device 50 may include:

[0112] The model splitting module 501 is used to split the first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client, so that the client determines the products to be recommended for the target user according to the perception layer model;

[0113] The data acquisition module 502 is used to acquire the historical operation data of the target user from the client;

[0114] The candidate product determination module 503 is used to input the historical operation data and the product feature information in the database into the second recommendation model for prediction processing, and obtain the set of candidate products corresponding to the target user;

[0115] The candidate product sending module 504 is used to send the set of candidate products to the client, so that the client determines the recommendation result according to the set of candidate products and the products to be recommended.

[0116] Optionally, the splitting of the first recommendation model into an input layer model and a perception layer model, and sending the perception layer model to the client includes:

[0117] Split the first recommendation model into an input layer model and a perception layer model;

[0118] Perform compression processing on the perception layer model to obtain the compressed perception layer model;

[0119] Send the compressed perception layer model to the client.

[0120] Optionally, the device further includes:

[0121] The data table generation module is used to generate and store the feature embedding vector data table according to the input layer model. The feature embedding vector data table includes the vector information of the user feature information and the product feature information. The input layer model is used to preprocess the user feature information and the product feature information input by the client, and obtain the corresponding vector information according to the preprocessed user feature information and product feature information, and construct the input parameters of the perception layer according to the obtained vector information;

[0122] The interface configuration module is used to configure the query interface for the feature embedding vector data table, so that the client can obtain the input parameters of the perception layer based on the user feature information and the product feature information, and determine the products to be recommended for the target user according to the input parameters of the perception layer.

[0123] Optionally, the device further includes:

[0124] The model training module is used to train the second recommendation model based on the pre-acquired training sample set to obtain the trained second recommendation model;

[0125] A model compression module, configured to compress the trained second recommendation model to obtain a first recommendation model.

[0126] Optionally, the model splitting module includes:

[0127] A historical data acquisition sub-module, configured to acquire the historical recommendation data of the target user;

[0128] A model adjustment sub-module, configured to adjust the first recommendation model according to the historical recommendation data to obtain an adjusted first recommendation model;

[0129] A model splitting sub-module, configured to split the adjusted first recommendation model into an input layer model and a perception layer model.

[0130] Figure 6 is a block diagram of another recommendation device provided by an embodiment of the present invention, which is applied to a client, such as Figure 6 shown, the device 60 may include:

[0131] A feature determination module 601, configured to determine the user feature information of the target user according to the real-time operation data of the target user;

[0132] A data reception module 602, configured to receive the perception layer model and the candidate commodity set sent by the server, where the perception layer model is obtained by the server splitting the first recommendation model; the candidate commodity set is obtained by the server performing prediction processing on the historical operation data of the target user and the commodity feature information in the database;

[0133] A recommended commodity determination module 603, configured to input the user feature information and the commodity feature information into the perception layer model for prediction processing to obtain the commodities to be recommended for the target user;

[0134] An update module 604, configured to update the candidate commodity set according to the commodities to be recommended to obtain a recommendation result and display it.

[0135] Optionally, the recommended commodity determination module includes:

[0136] A parameter query sub-module, configured to access the query interface for the feature embedding vector data table in the server, and query the perception layer input parameters corresponding to the target user according to the feature information of the target user and the commodity feature information;

[0137] A recommended commodity determination sub-module, configured to input the perception layer input parameters into the perception layer model for prediction processing to obtain the commodities to be recommended for the target user.

[0138] Optionally, the operation data includes at least one of user click data, user browsing data, user order data, and user comment data.

[0139] For the above device embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For related parts, please refer to the partial description of the method embodiments.

[0140] In addition, an embodiment of the present invention further provides a terminal, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements each process of the above method embodiment and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0141] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it implements each process of the above method embodiment and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0142] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other.

[0143] It is easy for those skilled in the art to think that any combination application of the above embodiments is feasible. Therefore, any combination among the above embodiments is an implementation scheme of the present invention. However, due to space limitations, this specification will not elaborate on each one here.

[0144] The recommended method provided here is not inherently related to any specific computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. It is obvious from the above description how to construct a system with the solution of the present invention. In addition, the present invention is not directed to any specific programming language. It should be understood that the content of the present invention described herein can be implemented using various programming languages, and the description of the specific language above is to disclose the best implementation mode of the present invention.

[0145] In the specification provided here, a large number of specific details are described. However, it can be understood that the embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures, and technologies are not shown in detail so as not to obscure the understanding of this specification.

[0146] Similarly, it should be understood that, in order to streamline the present invention and assist in understanding one or more of the various inventive aspects, in the foregoing description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected by the claims, the inventive aspects lie in less than all of the features of the single foregoing disclosed embodiment. Thus, the claims following the detailed description hereby expressly incorporate the detailed description, where each claim itself serves as a separate embodiment of the present invention.

[0147] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from those of the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that provides the same, equivalent, or similar purpose.

[0148] In addition, those skilled in the art will be able to understand that, although some of the embodiments described herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments is meant to be within the scope of the present invention and forms different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0149] Each component embodiment of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the method for performing operations according to the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (for example, a computer program and a computer program product) for performing part or all of the methods described herein. Such a program for implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0150] It should be noted that the above embodiments illustrate the present invention rather than limit the present invention, and those skilled in the art can design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names.

Claims

1. A recommendation method, characterized in that, Applied to the server side, the method includes: Split the first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client so that the client determines the to-be-recommended products of the target user according to the perception layer model, including: splitting the first recommendation model into an input layer model and a perception layer model; performing compression processing on the perception layer model to obtain a compressed perception layer model; sending the compressed perception layer model to the client; Obtain the historical operation data of the target user from the client; Input the historical operation data and the product feature information in the database into the second recommendation model for prediction processing to obtain a set of candidate products corresponding to the target user; Send the set of candidate products to the client so that the client determines the recommendation result according to the set of candidate products and the to-be-recommended products.

2. The method according to claim 1, wherein The method further includes: Generate and store a feature embedding vector data table according to the input layer model. The feature embedding vector data table includes vector information of user feature information and product feature information. The input layer model is used to preprocess the user feature information and product feature information input by the client, obtain corresponding vector information according to the preprocessed user feature information and product feature information, and construct perception layer input parameters according to the obtained vector information; Configure a query interface for the feature embedding vector data table for the client to obtain perception layer input parameters based on user feature information and product feature information, and determine the to-be-recommended products of the target user according to the perception layer input parameters.

3. The method according to claim 1, characterized in that, Before splitting the first recommendation model into an input layer model and a perception layer model, the method further includes: Train the second recommendation model based on a pre-obtained training sample set to obtain a trained second recommendation model; Perform compression processing on the trained second recommendation model to obtain the first recommendation model.

4. The method according to claim 1, wherein The splitting of the first recommendation model into an input layer model and a perception layer model includes: Obtain the historical recommendation data of the target user; Adjust the first recommendation model according to the historical recommendation data to obtain an adjusted first recommendation model; Split the adjusted first recommendation model into an input layer model and a perception layer model.

5. A recommendation method, characterized in that, Applied to the client side, the method includes: Determine the user feature information of the target user according to the real-time operation data of the target user; Receive the perception layer model and the set of candidate products sent by the server. The perception layer model is obtained by the server splitting the first recommendation model, including: after splitting the first recommendation model into an input layer model and a perception layer model, performing compression processing on the perception layer model to obtain a compressed perception layer model; the set of candidate products is obtained by the server performing prediction processing on the historical operation data of the target user and the product feature information in the database based on the second recommendation model; Input the user feature information and the product feature information into the perception layer model for prediction processing to obtain the to-be-recommended products of the target user; Update the set of candidate products according to the to-be-recommended products to obtain a recommendation result and display it.

6. The method according to claim 5, wherein Inputting the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user includes: Accessing the query interface for the feature embedding vector data table in the server, and querying the perception layer input parameters corresponding to the target user according to the user feature information and the product feature information of the target user; Inputting the perception layer input parameters into the perception layer model for prediction processing to obtain the products to be recommended for the target user.

7. A recommendation device, characterized in that, Applied to the server, the device includes: A model splitting module, configured to split a first recommendation model into an input layer model and a perception layer model, and send the perception layer model to the client, so that the client determines the products to be recommended for the target user according to the perception layer model, including: splitting the first recommendation model into an input layer model and a perception layer model; performing compression processing on the perception layer model to obtain a compressed perception layer model; sending the compressed perception layer model to the client; A data acquisition module, configured to acquire the historical operation data of the target user from the client; A candidate product determination module, configured to input the historical operation data and the product feature information in the database into a second recommendation model for prediction processing to obtain a set of candidate products corresponding to the target user; A candidate product sending module, configured to send the set of candidate products to the client, so that the client determines a recommendation result according to the set of candidate products and the products to be recommended.

8. A recommendation device, characterized in that, Applied to the client, the device includes: A feature determination module, configured to determine the user feature information of the target user according to the real-time operation data of the target user; A data receiving module, configured to receive the perception layer model and the set of candidate products sent by the server, where the perception layer model is obtained by the server splitting the first recommendation model, including: after splitting the first recommendation model into an input layer model and a perception layer model, performing compression processing on the perception layer model to obtain a compressed perception layer model; the set of candidate products is obtained by the server performing prediction processing on the historical operation data of the target user and the product feature information in the database based on the second recommendation model; A recommended product determination module, configured to input the user feature information and the product feature information into the perception layer model for prediction processing to obtain the products to be recommended for the target user; An update module, configured to update the set of candidate products according to the products to be recommended to obtain a recommendation result and display it.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the recommendation method as described in any one of claims 1 to 6.

10. An electronic device, characterized in that, Including: A processor and a memory, where the processor is configured to execute the data processing program stored in the memory to implement the recommendation method as described in any one of claims 1 to 6.

Images