Information recommendation method and device, storage medium and program product
By using the local information recommendation model of the behavior sequence Transformer and gradient boosting model on the client, the recommended information is reordered in real time, which solves the delay problem caused by cloud deployment and improves the immediacy and accuracy of the recommendation system.
Patent Information
- Application Number
- CN202510674988.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-23
- Publication Date
- 2025-09-05
AI Technical Summary
In existing service recommendation systems, the recommendation model is deployed in the cloud, which results in user interaction behavior being unable to affect the recommendation list in real time, leading to delayed recommendation results and reduced accuracy.
The local information recommendation model is implemented on the client side, and the behavior sequence Transformer model and gradient boosting model are used for training. The recommended information is re-sorted in real time based on the user's browsing information and pushed to the recommendation page.
It improves the immediacy and accuracy of recommended information, eliminates the impact of network transmission delays, and provides a smooth and personalized user experience.
Smart Images
Figure CN120596738A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of Internet software technology, and in particular to an information recommendation method, device, storage medium, and program product. Background Art
[0002] In existing service-based recommendation systems, recommendation models are typically deployed on cloud servers. Clients send requests to the cloud servers to retrieve recommendation lists and present them to users. However, this system architecture has some significant limitations: Client-side user interactions cannot be used in real-time for model predictions, immediately affecting the currently displayed recommendation list. This results in recommendations that fail to reflect users' latest interests and experience a certain delay in results.
[0003] The limitations of existing service recommendation systems directly affect the timeliness and accuracy of recommendations, and reduce user experience.
[0004] The above content is only used to assist in understanding the technical solution of this application and does not constitute an admission that the above content is prior art. Summary of the Invention
[0005] The main purpose of this application is to provide an information recommendation method, device, storage medium and program product, aiming to solve the technical problem of how to improve the timeliness and accuracy of information recommendation.
[0006] To achieve the above objectives, the present application proposes an information recommendation method, which is applied to a client and includes:
[0007] When a preset trigger condition is met, the recommended information is reordered according to the user's browsing information through a local information recommendation model to obtain reordered recommended information, wherein the local information recommendation model is trained based on the behavior sequence Transformer model and the gradient boosting model;
[0008] The reordered recommendation information is pushed to the recommendation page for the user to browse.
[0009] In one embodiment, before the step of reordering the recommended information based on the user's browsing information using the local information recommendation model when a preset trigger condition is met and obtaining the reordered recommended information, the method includes:
[0010] Sending a data acquisition request to a cloud service platform, so that the cloud service platform sends recommendation information, associated feature data, and model configuration information of a cloud information recommendation model to the client according to the data acquisition request;
[0011] Receive the recommendation information, associated feature data, and model configuration information of the cloud information recommendation model; push the recommendation information to a recommendation page for the user to browse; and / or
[0012] Determining whether the local information recommendation model needs to be updated based on the model configuration information;
[0013] When it is confirmed that an update is required, the cloud information recommendation model is asynchronously downloaded from the database.
[0014] In one embodiment, the local information recommendation model is obtained by the cloud service platform using a gradient boosting model to construct a model training sample set based on system users' browsing information, system users' feature data, and recommended information feature data, and training a behavior sequence Transformer model; and / or
[0015] The preset trigger condition includes at least one click of the user on the recommendation information on the recommendation page.
[0016] In one embodiment, the step of reordering the recommended information according to the user's browsing information using the local information recommendation model to obtain the reordered recommended information includes:
[0017] Obtaining user behavior characteristics based on the user's browsing information;
[0018] Based on the local information recommendation model, and according to the user behavior characteristics and the associated feature data, performing click prediction on the unexposed recommended information in the recommended information to obtain a predicted click rate of the unexposed recommended information;
[0019] reordering the unexposed recommendation information according to the predicted click-through rate to obtain reordered unexposed recommendation information;
[0020] The step of pushing the reordered recommendation information to a recommendation page for user browsing includes:
[0021] The reordered recommendation information is pushed to the unexposed area of the recommendation page for user browsing.
[0022] In addition, to achieve the above-mentioned purpose, the present application also proposes an information recommendation method, which is applied to a cloud service platform and includes:
[0023] According to the model download request of the client, an information recommendation model is sent to the client, so that when the preset trigger condition is met, the client reorders the recommended information according to the user's browsing information through the information recommendation model to obtain the reordered recommended information, wherein the information recommendation model is trained by the cloud service platform based on the behavior sequence Transformer model and the gradient boosting model.
[0024] In one embodiment, before the step of sending the information recommendation model to the client according to the model download request of the client, the following steps are included:
[0025] Collect browsing information of system users through the client;
[0026] Obtain system user feature data and recommended information feature data from the database;
[0027] Constructing a model training sample set based on the browsing information, system user feature data and recommendation information feature data;
[0028] Using the gradient boosting model, feature pruning is performed on several sample features of the model training sample set to obtain a pruned model training sample set;
[0029] Based on the behavior sequence Transformer model, model training is performed according to the trimmed model training sample set to obtain an information recommendation model.
[0030] In one embodiment, the step of using the gradient boosting model to perform feature pruning on several sample features of the model training sample set to obtain a pruned model training sample set includes:
[0031] Using the gradient boosting model, a decision tree is constructed based on the model training sample set, and feature gains of the sample features are determined;
[0032] Determining a feature gain ranking corresponding to the model training sample set according to the feature gain of the sample feature;
[0033] The model training sample set is trimmed according to the feature gain sorting to obtain a trimmed model training sample set.
[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes an information recommendation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the information recommendation method described above.
[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium, and when the computer program is executed by a processor, the steps of the information recommendation method described above are implemented.
[0036] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the information recommendation method described above.
[0037] One or more technical solutions proposed in this application have at least the following technical effects:
[0038] The information recommendation method, device, storage medium and program product proposed in the embodiments of the present application specifically obtain reordered recommendation information by reordering the recommended information according to the user's browsing information through a local information recommendation model when a preset trigger condition is met, wherein the local information recommendation model is trained based on a behavior sequence Transformer model and a gradient boosting model; and the reordered recommendation information is pushed to a recommendation page for the user to browse.
[0039] This application is applied to the client. When the preset trigger conditions are met, the client's local information recommendation model is used to combine the user's browsing information in real time, re-sort the recommendation order of the recommended information, and push the re-sorted recommended information to the user for browsing. The above method can be combined with the user's interactive behavior in real time, and use the client's local recommendation information model to directly complete the sorting reasoning and dynamic update of the recommendation results without the need for data interaction with the cloud server, thereby eliminating the impact of network transmission delay on the timeliness of the recommendation, and realizing the seamless optimization of the recommendation strategy. While protecting user privacy, it significantly improves the immediacy and accuracy of information recommendations, thereby providing users with a smoother and more personalized service experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0042] Figure 1 A flowchart of the first embodiment of the information recommendation method for this application is provided;
[0043] Figure 2 A flowchart of Example 3 of the information recommendation method for this application is provided;
[0044] Figure 3 A flowchart of Example 4 of the information recommendation method of this application is provided;
[0045] Figure 4 A schematic diagram of the device structure of the hardware operating environment involved in the information recommendation method in the embodiment of the present application;
[0046] Figure 5 This is an example diagram of the recommendation system architecture involved in Example 2 of this application;
[0047] Figure 6 This is an example diagram of the information recommendation optimization sorting effect involved in Example 4 of this application.
[0048] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0049] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.
[0050] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.
[0051] The main solution of the embodiment of the present application is: when the preset trigger conditions are met, the recommended information is reordered according to the user's browsing information through the local information recommendation model to obtain the reordered recommended information, wherein the local information recommendation model is trained based on the behavior sequence Transformer model and the gradient boosting model; the reordered recommended information is pushed to the recommendation page for the user to browse.
[0052] Technical terms involved in this application:
[0053] BST (Behavior Sequence Transformer) model: The BST model is a deep learning model based on the self-attention mechanism. It utilizes the Transformer architecture to process user behavior sequence data. It effectively captures long-term dependencies within user behavior sequences, thereby providing more accurate information recommendation predictions. Its core concept is to capture the dynamic relationships between different behaviors in a sequence through a multi-head self-attention mechanism and to enhance sequence order information by introducing timestamps or position encodings.
[0054] Gradient Boosting Model: The gradient boosting model is an ensemble learning algorithm based on the boosting framework. It gradually corrects prediction errors by iteratively training weak learners (such as decision trees). It is used for regression and classification problems. Its core principle is residual learning. In each round of training, a new weak learner is trained to fit the negative gradient (i.e., residual) of the current model, gradually approaching the true target through weighted superposition.
[0055] In this embodiment, for ease of description, the following description is made with the information recommendation device as the execution subject.
[0056] In existing service-based recommendation systems, recommendation models are typically deployed on cloud servers, with clients sending requests to the cloud servers to retrieve recommendation lists and present them to users. However, this system architecture has some significant limitations.
[0057] Existing service-based recommendation technologies suffer from two main drawbacks: First, latency in recommendation lists. Clients lack the ability to adjust the order of recommendations in real time, so user interactions don't immediately affect the currently displayed recommendation list. Consequently, recommendation results fail to reflect the user's latest interests. Users must swipe the screen multiple times to trigger new recommendation requests before their interactions are considered for model inference, thus affecting the recommendation list.
[0058] Second, there's the latency of processing user interaction data. Client-side user interaction data is typically uploaded to cloud servers via tracking logs. After data processing and feature calculation, it can be used for subsequent model predictions. This entire process involves multiple data processing steps, making it impossible for data to drive real-time updates to information recommendation devices.
[0059] Therefore, these limitations of existing service recommendation systems directly affect the timeliness and accuracy of recommendation results and reduce user experience.
[0060] This application provides a solution. When the preset trigger conditions are met, the client's local information recommendation model is used to combine the user's browsing information in real time, re-sort the recommendation order of the recommended information, and push the re-sorted recommended information to the user for browsing. The above method can be combined with the user's interactive behavior in real time, and the local recommendation information model of the client is used to directly complete the sorting reasoning and dynamic update of the recommendation results without the need for data interaction with the cloud server, thereby eliminating the impact of network transmission delay on the timeliness of the recommendation, and realizing the seamless optimization of the recommendation strategy. While protecting user privacy, it significantly improves the immediacy and accuracy of information recommendations, thereby providing users with a smoother and more personalized service experience.
[0061] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, mobile phone, etc., or an electronic device or information recommendation device capable of implementing the above functions. The following uses the information recommendation device as an example to illustrate this embodiment and the following embodiments.
[0062] Based on this, the embodiment of the present application provides an information recommendation method, referring to Figure 1 , Figure 1 This is a flowchart of the first embodiment of the information recommendation method of this application.
[0063] In this embodiment, the information recommendation method is applied to a client, and the information recommendation method includes steps S110 to S120:
[0064] Step S110: When a preset trigger condition is met, the recommended information is reordered according to the user's browsing information by using a local information recommendation model to obtain reordered recommended information, wherein the local information recommendation model is trained based on a behavior sequence Transformer model and a gradient boosting model;
[0065] It should be noted that the preset trigger condition refers to the condition under which the client triggers the local information recommendation model to re-sort when the user performs a specific operation on the recommendation page. It can be a click on the recommended information, the user's content search behavior, the user's stay time on the recommendation page, etc. In this application, recommended information refers to the content recommended by the client to the user, including but not limited to various types of goods, services, stores, etc. The client refers to the front-end part of the mobile device or application used by the user, which is responsible for interacting with the user and performing local computing tasks.
[0066] It is understood that since this application is applied to the information recommendation scenario, the information recommendation device needs to execute step S110. When the preset trigger condition is met, the client calls the local information recommendation model to determine the user's behavior sequence based on the user's browsing information, and re-orders the recommended information based on the user's historical interests and store information to obtain the re-ordered recommended information. The user's browsing information includes behavioral data such as clicks, browsing, searches, and dwell time on the recommended page.
[0067] The re-ranked recommendation information is combined with the user's interactive behavior in real time, and the client's local recommendation information model is used to directly complete the ranking reasoning and dynamic update of the recommendation results. It will be more in line with the user's current interests and preferences, effectively improving the immediacy and accuracy of information recommendations, thereby enhancing the user's browsing experience.
[0068] Step S120: Push the reordered recommendation information to a recommendation page for user browsing.
[0069] It should be understood that in order to avoid duplicate recommendations to users, the information recommendation device needs to remove the recommended information that has already been viewed by the user from the reordered recommended information before pushing it to the recommendation page, and directly push the reordered and unviewed recommended information to the user for browsing, effectively improving the timeliness of information recommendations and thus enhancing the user's browsing experience. The recommendation page is the interface in the client for users to browse recommended information.
[0070] Through the above scheme, in this embodiment, when the preset trigger conditions are met, the information recommendation device uses the client's local information recommendation model to combine the user's browsing information in real time, re-sort the recommendation order of the recommended information, and push the re-sorted recommended information to the user for browsing. The above method can be combined with the user's interactive behavior in real time, and use the client's local recommendation information model to directly complete the sorting reasoning and dynamic update of the recommendation results without the need for data interaction with the cloud server, thereby eliminating the impact of network transmission delay on the timeliness of the recommendation, and realizing the seamless optimization of the recommendation strategy. While protecting user privacy, it significantly improves the immediacy and accuracy of information recommendations, thereby providing users with a smoother and more personalized service experience.
[0071] The second embodiment of the present application provides another information recommendation method. In this embodiment, the information recommendation method is applied to a cloud service platform, including step S210:
[0072] In step S210, according to the model download request of the client, an information recommendation model is sent to the client, so that when the preset trigger condition is met, the client reorders the recommended information according to the user's browsing information through the information recommendation model to obtain the reordered recommended information, wherein the information recommendation model is trained by the cloud service platform based on the behavior sequence Transformer model and the gradient boosting model.
[0073] Specifically, before the client calls the local information recommendation model, it needs to send a model download request to the cloud service platform. The cloud service platform confirms the version information of the information recommendation model required by the client based on the client's model download request, and sends the information recommendation model to the client, so that the client has the information recommendation model locally, that is, the local information recommendation model.
[0074] It should be understood that the client is connected to the cloud service platform and the database to form a recommendation system.
[0075] For example, the information recommendation method can be applied to Figure 5In the client-side or cloud-side architecture shown, the recommendation system includes a service application (APP), i.e., the client. The client includes a recommendation page that implements data display, user interaction, and local reordering. The client also implements model resource management, feature preprocessing, and inference engine functions.
[0076] The cloud platform includes a model platform and a recommendation service platform. The recall module in the recommendation service platform retrieves relevant data for selected recommendations, for example, by region or price. The sorting module uses the cloud-based information recommendation model to sort recommendations, while the reordering module reorders recommendations based on pre-set business metrics. The model platform manages versioning, training, and prediction of the information recommendation model. Data storage utilizes multiple databases, including Redis, Hive, ECS, and MySQL.
[0077] Furthermore, before step S210, the information recommendation method further includes steps A01 to A05:
[0078] Step A01, collecting browsing information of system users through the client;
[0079] Step A02: Obtain system user characteristic data and recommendation information characteristic data from the database;
[0080] Step A03: constructing a model training sample set based on the browsing information, system user characteristic data, and recommendation information characteristic data;
[0081] It should be noted that the cloud service platform refers to the software system deployed on cloud servers, responsible for processing data uploaded by clients, training and storing models, and providing service support to clients. A database refers to a structured or unstructured data storage system that stores system user data, recommendation information data, and model data, such as a relational database (MySQL, PostgreSQL, etc.) or a non-relational database (MongoDB, HBase, etc.). System users are those who use the recommendation services provided by the recommendation system.
[0082] The cloud service platform needs to collect the browsing information of system users on the recommendation page through the client. Specifically, through the monitoring and recording functions of the client, it captures various behavioral data of users on the recommendation page and transmits it to the cloud service platform or stores it in the database for subsequent processing.
[0083] The cloud service platform also needs to obtain system user characteristic data and recommendation information characteristic data from the database. System user characteristic data includes user personal information, historical behavior data, interest tags, and other data that can reflect user characteristics. Recommendation information characteristic data includes attribute information of recommended content, such as the store's location, dish type, price range, and user reviews.
[0084] Then, the cloud service platform first extracts features from the browsing information of system users and confirms the recent behavioral feature data of system users, including but not limited to the recommended information, search content and total browsing time clicked by users in the past day, as well as the recommended information list, search content, browsing time, number of swipes and other data clicked by users on the current recommendation page.
[0085] Then, the system user's recent behavior feature data, system user feature data and recommended information feature data are integrated and structured, and a model training sample set is constructed with <system user identifier, recommended information identifier> as the sample and the system user's recent behavior feature data, system user feature data and recommended information feature data as the sample features, so as to obtain a model training sample set including several sample features, so that the model training sample can be used to train the information recommendation model in the future.
[0086] Step A04: using the gradient boosting model, perform feature pruning on several sample features of the model training sample set to obtain a pruned model training sample set;
[0087] Specifically, since users often access the client through the user-side terminal, in order to reduce the resource usage of the information recommendation model, the cloud service platform can use the gradient boosting model to perform feature importance analysis on several sample features of the model training sample set, reduce the features that contribute less to the prediction ability of the information recommendation model, and thus perform feature trimming on the model training sample set including several sample features to obtain the trimmed model training sample set, thereby reducing the complexity of the information recommendation model and improving the model training efficiency and the generalization ability of the model.
[0088] In a feasible implementation, the step A04: using the gradient boosting model to perform feature pruning on several sample features of the model training sample set to obtain a pruned model training sample set, includes steps B01 to B03:
[0089] Step B01, using the gradient boosting model, constructing a decision tree based on the model training sample set, and determining the feature gain of the sample features;
[0090] Specifically, the cloud service platform uses a complete model training sample set (including all sample features) to train the gradient boosting model to construct a decision tree, wherein a decision tree refers to a tree structure model, each internal node represents a judgment on a sample feature, each branch represents the result of the judgment, and each leaf node corresponds to a prediction result. Among them, the gradient boosting model can be GBDT (Gradient Boosting Decision Tree), XGBoost, LightGBM (Light Gradient Boosting Machine), NGBoost (Natural Gradient Boosting). In this embodiment, the gradient boosting model is preferably a LightGBM model.
[0091] Specifically, during the decision tree construction process, the gradient boosting model traverses all sample features in the model training sample set and calculates the information gain of each sample feature at each possible split point (i.e., at an internal node) based on the split effect. The information gain can be calculated based on indicators such as the Gini coefficient, information entropy, and the reduction in impurity before and after the split.
[0092] By comparing the information gain of different sample features, the optimal feature at the current node is selected as the basis for splitting, forming a branch of the decision tree. This branch construction process is continuously recursively carried out until the preset stopping conditions, such as tree depth and sample number threshold, are reached, thus constructing a complete decision tree. Through multiple rounds of iteration and the combination of multiple trees, the gradient boosting model can comprehensively evaluate the importance of each feature in different contexts, thereby providing a reliable feature importance score, namely feature gain, for feature pruning.
[0093] Step B02: determining a feature gain ranking corresponding to the model training sample set based on the feature gain of the sample features;
[0094] Step B03: trimming the model training sample set according to the feature gain sorting to obtain a trimmed model training sample set.
[0095] Specifically, the cloud service platform sorts all sample features according to their feature gain to obtain the feature gain ranking corresponding to the model training sample set.
[0096] Then, in order to reduce the features that contribute less to the predictive ability of the information recommendation model, sample features below the preset gain threshold can be eliminated, or the sample features ranked in the top N by feature gain can be selected based on the maximum number N of model features, or the model performance of different feature subsets can be evaluated based on cross-validation, and the feature subset with the best performance can be selected to achieve feature trimming of the model training sample set, so as to use the trimmed model training sample set to train a smaller BST model that is more suitable for mobile device inference, avoiding the repeated attempts to permutate and combine sample features one by one to consume a large amount of training resources and reducing model training time.
[0097] Step A05: Based on the behavior sequence Transformer model, model training is performed according to the trimmed model training sample set to obtain an information recommendation model.
[0098] Specifically, the cloud service platform uses the trimmed model training sample set to train the behavior sequence Transformer model to obtain a trained information recommendation model, and stores the trained information recommendation model in the database.
[0099] The BST model structure primarily consists of a feature embedding layer, a Transformer layer, and a multi-layer perceptron (MLP) layer. The specific model training process is as follows: First, the feature embedding layer processes the pruned model training sample set, converting it into an acceptable input format for the model, such as representing sample features as vectors while constructing user behavior sequences. Second, the parameters of the behavior sequence Transformer model are initialized, including the weights of the feature embedding layer, the Transformer layer, and the multi-layer perceptron (MLP) layer. Then, the Transformer layer's self-attention mechanism is used to model the user's historical behavior sequence, capturing the dynamic changes and long-term dependencies of user interests. It automatically learns the correlation between the recommendation target and the user's behavior sequence, and outputs the Transformer result. The Transformer result is then concatenated with other feature representations, and the MLP layer outputs the click-through rate prediction results for the recommended information. The cloud service platform can also adjust the model parameters through a backpropagation algorithm to minimize the loss function between the predicted results and the true labels. Finally, the trained model is validated and tested to evaluate its accuracy and generalization ability in recommendation tasks.
[0100] Furthermore, in order to improve the training efficiency and performance of the model, model training optimization strategies can be adopted, such as learning rate adjustment, batch normalization, regularization, etc.
[0101] This embodiment uses the above scheme to send an information recommendation model to the client according to the client's model download request, so that when the preset trigger conditions are met, the client re-sorts the recommended information according to the user's browsing information through the information recommendation model to obtain the re-sorted recommended information, wherein the information recommendation model is trained by the cloud service platform based on the behavior sequence Transformer model and the gradient boosting model. The above method enables the client to combine the user's interactive behavior in real time and use the client's local recommendation information model to directly complete the sorting reasoning and dynamic update of the recommendation results without data interaction with the cloud server, thereby eliminating the impact of network transmission delay on the timeliness of recommendation, and realizing the seamless optimization of the recommendation strategy, while protecting user privacy. It significantly improves the immediacy and accuracy of information recommendations, thereby providing users with a smoother and more personalized service experience.
[0102] Based on the above embodiments of the present application, in the third embodiment of the present application, the same or similar contents as those of the first embodiment and / or the second embodiment can be referred to the above introduction and will not be described in detail later. Figure 2 , before step S110, steps S310 to S350 are included:
[0103] Step S310: Sending a data acquisition request to a cloud service platform, so that the cloud service platform sends recommendation information, associated feature data, and model configuration information of a cloud information recommendation model to the client according to the data acquisition request;
[0104] Step S320, receiving the recommendation information, associated feature data, and model configuration information of the cloud information recommendation model;
[0105] Step S330: Push the recommendation information to a recommendation page for the user to browse; and / or
[0106] It should be noted that model configuration information refers to configuration information such as the version number, parameter settings, and input and output formats of the cloud-based information recommendation model. Associated feature data refers to various feature data related to recommended information, such as basic user characteristics, historical behavior characteristics, and attribute characteristics of recommended information. This feature data is used as input to the model to achieve personalized recommendations. In this embodiment, the associated feature data includes at least system user feature data and recommended information feature data.
[0107] Specifically, when a user uses the information recommendation service on a client, the client sends a data retrieval request to the cloud service platform via a network request. The client then receives the recommended information, associated feature data, and model configuration information for the cloud information recommendation model from the cloud service platform in real time. After receiving the data retrieval request, the cloud service platform retrieves the corresponding data from the database and packages it into a response, returning it to the client.
[0108] Then, the client pushes the recommendation information to the associated recommendation page, allowing the user to browse the recommendation information.
[0109] Step S340: confirming whether the local information recommendation model needs to be updated based on the model configuration information;
[0110] Step S350: When it is confirmed that an update is required, the cloud information recommendation model is asynchronously downloaded from the database.
[0111] It should be understood that the cloud information recommendation model and the local information recommendation model are essentially information recommendation models, and the only difference is the location where the models are stored.
[0112] The local information recommendation model or the cloud information recommendation model is obtained by training the behavior sequence Transformer model using a gradient boosting model based on the system user's browsing information, system user feature data, and recommended information feature data to construct a model training sample set.
[0113] To ensure consistency in the inference and prediction capabilities of the client's local information recommendation model and the cloud service platform's cloud information recommendation model, before the client uses the local information recommendation model to re-rank recommended information, it must confirm whether the local information recommendation model needs to be updated based on the model configuration information. If a discrepancy is confirmed between the local information recommendation model and the received model configuration information, it is confirmed that the local information recommendation model needs to be updated. The client then asynchronously downloads the cloud information recommendation model from the database to synchronize the cloud and client models.
[0114] In this embodiment, the client obtains the model configuration information, recommendation information, and associated feature data of the cloud information recommendation model from the cloud service platform, so as to push the recommendation information to the recommendation page for the user to browse. At the same time, it determines whether the client's local information recommendation model needs to be updated based on the model configuration information. When an update is required, it is updated through asynchronous download, thereby ensuring the consistency of the cloud and client models. After the user triggers the reordering of the recommendation information on the client, the computing power of the user's mobile device can be quickly used to perform model prediction, thereby effectively improving the timeliness and accuracy of information recommendations.
[0115] Based on the above embodiments of the present application, in the fourth embodiment of the present application, the same or similar contents as those of the first embodiment and / or the second embodiment and / or the third embodiment can be referred to above and will not be described in detail. Figure 3 The step S110 of reordering the recommended information according to the user's browsing information by using the local information recommendation model to obtain the reordered recommended information includes steps S410 to S430:
[0116] Step S410, obtaining user behavior characteristics based on the user's browsing information;
[0117] Specifically, the browsing information collected by the client includes exposure logs and click logs. The exposure log records multiple recommended information in the exposure area of the recommendation page, and the click log records the number of clicks by the user on the recommendation page. Preferably, the preset trigger condition can be at least one click by the user on the recommendation information on the recommendation page.
[0118] When a user clicks on the recommended information in the exposure area on the client's recommendation page at least once, the information recommendation device needs to extract user behavior characteristics based on the browsing information, specifically the recommended information list clicked by the user on the current recommendation page, search content, browsing time, number of swipes, and other data.
[0119] Step S420: Based on the local information recommendation model and according to the user behavior characteristics and the associated feature data, predicting clicks on the unexposed recommended information in the recommended information to obtain a predicted click-through rate of the unexposed recommended information;
[0120] Step S430: reordering the unexposed recommendation information according to the predicted click-through rate to obtain reordered unexposed recommendation information;
[0121] Then, the information recommendation device needs to use the client's local information recommendation model, combined with user behavior characteristics and other related feature data, to predict the click-through rate of the recommended information in the unexposed area of the recommendation page (i.e., unexposed recommended information), and obtain the user's predicted click-through rate for multiple unexposed recommended information.
[0122] Finally, according to the predicted click-through rate of each unexposed recommendation information, the recommendation order of the unexposed recommendation information is reordered in descending order from large to small, thereby obtaining the reordered unexposed recommendation information.
[0123] Furthermore, the step of pushing the reordered recommendation information to a recommendation page for user browsing includes:
[0124] The reordered recommendation information is pushed to the unexposed area of the recommendation page for user browsing.
[0125] It is understandable that after obtaining the re-ordered unexposed recommendation information, the information recommendation device can push the re-ordered recommendation information to the unexposed area of the recommendation page, thereby realizing the user's imperceptible update of the information recommendation results and realizing the imperceptible optimization of the recommendation strategy.
[0126] For example, refer to Figure 6 The example diagram of the information recommendation sorting optimization effect shown in the figure shows that when the user clicks on the exposed area of the life service recommendation page, the recommendation order of the unexposed stores in the unexposed area is triggered to be re-arranged, thereby achieving seamless optimization of the recommendation strategy.
[0127] Furthermore, after step S430, the information recommendation method further includes:
[0128] According to preset business indicators, the re-sorted unexposed recommendation information is optimized to obtain optimized unexposed recommendation information.
[0129] Specifically, since the present application is applicable to information recommendation scenarios, information recommendation services usually have relevant business indicator requirements. For example, the recommendation order of recommended information with high business importance should be advanced, and the user's click rate or conversion rate on the recommendation page must be maximized. Therefore, after using the client's local information recommendation model to reorder the recommended information, the reordered unexposed recommended information can be further optimized according to the recommendation business rules preset by relevant personnel, so that the optimized unexposed recommended information can not only reflect the user's latest interests in a timely manner, but also ensure that it meets the relevant business indicator requirements.
[0130] Through the above solution, this embodiment specifically collects real-time user browsing information from the client, uses the client's local information recommendation model, and combines it with the associated feature data obtained in advance from the cloud service platform to re-rank the unexposed recommended information to obtain the re-ranked unexposed recommended information. By pushing the re-ranked unexposed recommended information to the user, it is possible to achieve real-time integration with the user's interactive behavior, directly complete the ranking reasoning and dynamic update of the recommendation results using the client's local recommendation information model, without the need for data interaction with the cloud server, thereby effectively improving the timeliness and accuracy of information recommendations, while also effectively protecting the user's privacy, achieving user-imperceptible optimization of information recommendations, and improving the user's browsing experience.
[0131] The present application provides an information recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the information recommendation method in the above-mentioned embodiment 1.
[0132] Reference below Figure 4 , which shows a schematic diagram of the structure of an information recommendation device suitable for implementing an embodiment of the present application. The information recommendation device in the embodiment of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The information recommendation device shown is only an example and should not limit the functions and scope of use of the embodiments of the present application.
[0133] like Figure 4 As shown, the information recommendation device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory 1002 or programs loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the information recommendation device. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the information recommendation device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an information recommendation device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems can be implemented or have alternatively.
[0134] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.
[0135] The information recommendation device provided in this application, utilizing the information recommendation method described in the aforementioned embodiment, can address the technical problem of improving the timeliness and accuracy of information recommendations. Compared to the prior art, the information recommendation device provided in this application achieves the same beneficial effects as the information recommendation method described in the aforementioned embodiment. Other technical features of this information recommendation device are the same as those disclosed in the aforementioned embodiment and are not further elaborated here.
[0136] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0137] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.
[0138] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer program) stored thereon, wherein the computer-readable program instructions are used to execute the information recommendation method in the above-mentioned embodiment.
[0139] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.
[0140] The computer-readable storage medium may be included in the information recommendation device, or may exist independently without being incorporated into the information recommendation device.
[0141] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the information recommendation device, the information recommendation device: when a preset trigger condition is met, reorders the recommended information according to the user's browsing information through the local information recommendation model to obtain reordered recommended information, wherein the local information recommendation model is trained based on the behavior sequence Transformer model and the gradient boosting model; and pushes the reordered recommended information to the recommendation page for the user to browse.
[0142] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).
[0143] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.
[0144] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.
[0145] The computer-readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned information recommendation method. This computer-readable storage medium can address the technical problem of improving the timeliness and accuracy of information recommendations. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the information recommendation method provided in the aforementioned embodiment, and are not further elaborated here.
[0146] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned information recommendation method when executed by a processor.
[0147] The computer program product provided in this application can solve the technical problem of how to improve the timeliness and accuracy of information recommendations. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the information recommendation method provided in the above embodiment, and will not be repeated here.
[0148] The above description is only part of the embodiments of the present application and does not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.
Claims
1. An information recommendation method, characterized in that: The information recommendation method is applied to a client, and the information recommendation method includes: When a preset trigger condition is met, the recommended information is reordered according to the user's browsing information through a local information recommendation model to obtain reordered recommended information, wherein the local information recommendation model is trained based on the behavior sequence Transformer model and the gradient boosting model; The reordered recommendation information is pushed to the recommendation page for the user to browse.
2. The information recommendation method according to claim 1, wherein: Before the step of reordering the recommended information according to the user's browsing information by using the local information recommendation model when the preset trigger condition is met to obtain the reordered recommended information, the method includes: Sending a data acquisition request to a cloud service platform, so that the cloud service platform sends recommendation information, associated feature data, and model configuration information of a cloud information recommendation model to the client according to the data acquisition request; receiving the recommendation information, associated feature data, and model configuration information of a cloud information recommendation model; Pushing the recommendation information to a recommendation page for the user to browse; and / or Determining whether the local information recommendation model needs to be updated based on the model configuration information; When it is confirmed that an update is required, the cloud information recommendation model is asynchronously downloaded from the database.
3. The information recommendation method according to claim 2, wherein: The local information recommendation model is obtained by the cloud service platform using a gradient boosting model to construct a model training sample set based on system users' browsing information, system users' feature data and recommended information feature data, and training a behavior sequence Transformer model; and / or The preset trigger condition includes at least one click of the user on the recommendation information on the recommendation page.
4. The information recommendation method according to claim 3, wherein: The step of reordering the recommended information according to the user's browsing information using the local information recommendation model to obtain the reordered recommended information includes: Obtaining user behavior characteristics based on the user's browsing information; Based on the local information recommendation model, and according to the user behavior characteristics and the associated feature data, performing click prediction on the unexposed recommended information in the recommended information to obtain a predicted click rate of the unexposed recommended information; reordering the unexposed recommendation information according to the predicted click-through rate to obtain reordered unexposed recommendation information; The step of pushing the reordered recommendation information to a recommendation page for user browsing includes: The reordered recommendation information is pushed to the unexposed area of the recommendation page for user browsing.
5. An information recommendation method, characterized in that: The information recommendation method is applied to a cloud service platform, and the information recommendation method includes: According to the model download request of the client, an information recommendation model is sent to the client, so that when the preset trigger condition is met, the client reorders the recommended information according to the user's browsing information through the information recommendation model to obtain the reordered recommended information, wherein the information recommendation model is trained by the cloud service platform based on the behavior sequence Transformer model and the gradient boosting model.
6. The information recommendation method according to claim 5, wherein: Before the step of sending information recommendation model to the client according to the model download request of the client, the method includes: Collect browsing information of system users through the client; Obtain system user feature data and recommended information feature data from the database; Constructing a model training sample set based on the browsing information, system user feature data and recommendation information feature data; Using the gradient boosting model, feature pruning is performed on several sample features of the model training sample set to obtain a pruned model training sample set; Based on the behavior sequence Transformer model, model training is performed according to the trimmed model training sample set to obtain an information recommendation model.
7. The information recommendation method according to claim 6, wherein: The step of using the gradient boosting model to perform feature pruning on several sample features of the model training sample set to obtain a pruned model training sample set includes: Using the gradient boosting model, a decision tree is constructed based on the model training sample set, and feature gains of the sample features are determined; Determining a feature gain ranking corresponding to the model training sample set according to the feature gain of the sample feature; The model training sample set is trimmed according to the feature gain sorting to obtain a trimmed model training sample set.
8. An information recommendation device, characterized in that: The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the information recommendation method according to any one of claims 1 to 7.
9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the information recommendation method according to any one of claims 1 to 7 are implemented.
10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the information recommendation method according to any one of claims 1 to 7 are implemented.