Article recommendation method and device, electronic device, and storage medium

Abstract

The embodiment of the application provides an article recommendation method and device, electronic equipment and storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: obtaining an article data set, and performing feature processing on the article data set to obtain article feature data; wherein the number of article feature data is multiple, and each article feature data corresponds to an article; obtaining a user data set, and performing feature processing on the user data set to obtain user feature data; wherein the number of user feature data is multiple, and each user feature data corresponds to a user; constructing an article recommendation model based on the article feature data and the user feature data; for each user, determining a target recommended article from the multiple article feature data through the article recommendation model; and recommending the target recommended article to the user. The embodiment of the application can establish the connection between multiple users, realize article recommendation according to the correlation between users, and improve the accuracy and correlation of article recommendation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an article recommendation method, apparatus, electronic device, and storage medium. Background Technology

[0002] Currently, article recommendation methods typically employ machine learning or deep learning models. However, in practical applications, these models often overlook the potential connections between users when optimizing recommendation algorithms. This lack of effective utilization of the correlations between different users prevents them from recommending articles based on user relationships, thus reducing the accuracy and relevance of the recommendations.

[0003] Therefore, improving the accuracy of article recommendations has become an urgent technical problem to be solved. Summary of the Invention

[0004] The main objective of this application is to provide an article recommendation method, apparatus, electronic device, and storage medium, which aims to improve the accuracy of article recommendations.

[0005] To achieve the above objectives, a first aspect of this application proposes an article recommendation method, the method comprising:

[0006] Obtain an article dataset and perform feature processing on the article dataset to obtain article feature data; wherein, there are multiple articles with feature data, and each article feature data corresponds to an article;

[0007] A user dataset is acquired, and feature processing is performed on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user;

[0008] An article recommendation model is constructed based on the article feature data and the user feature data;

[0009] For each user, the target recommended article is determined from multiple article feature data using the article recommendation model;

[0010] The target recommended article is recommended to the user.

[0011] In some embodiments, determining the target recommended article from multiple article feature data using the article recommendation model includes:

[0012] Determine the Gaussian distribution data for the article recommendation model;

[0013] The Gaussian distribution data is sampled to obtain distribution characteristic data;

[0014] Calculate the upper bound of the confidence level for each of the article feature data based on the distribution feature data;

[0015] The target recommended article is determined from multiple article feature data based on the upper bound of the confidence level.

[0016] In some embodiments, the distribution feature data includes distribution mean feature data and distribution variance feature data; calculating the upper bound of the confidence level for each of the article feature data based on the distribution feature data includes:

[0017] Based on the article feature data and the distribution mean feature data, a first feature calculation is performed to obtain the first feature data;

[0018] The second feature is calculated based on the preset confidence factor, the article feature data, and the distribution variance feature data to obtain the second feature data;

[0019] The confidence upper bound is obtained by performing aggregation calculations based on the first feature data and the second feature data.

[0020] In some embodiments, after recommending the target article to the user, the method further includes:

[0021] Obtain feedback information from the user; wherein the feedback information is used to characterize the user's viewing behavior of the target recommended article;

[0022] Recommendation reward information is constructed based on the feedback information and the user feature data;

[0023] The article recommendation record is updated based on the target recommended article and the recommendation reward information; wherein, the article recommendation record includes the historical recommended articles of each user and the recommendation reward information corresponding to the historical recommended articles, and the historical recommended articles are determined by the article recommendation model;

[0024] The Gaussian distribution data is updated based on the updated article recommendation records to update the article recommendation model; wherein, the updated article recommendation model is used for the next article recommendation.

[0025] In some embodiments, determining the target recommended article from multiple article feature data based on the upper bound of the confidence level includes:

[0026] The confidence upper bounds corresponding to multiple article feature data are sorted to obtain a first confidence sequence;

[0027] Select the upper bound of the confidence level with the largest value from the first confidence level sequence and determine it as the first target confidence level data;

[0028] Obtain the article feature data corresponding to the first target confidence data, and determine it as the target recommended article.

[0029] In some embodiments, determining the target recommended article from multiple article feature data based on the upper bound of the confidence level includes:

[0030] The multiple article feature data are grouped to obtain at least one group of article data;

[0031] For each group of article data, the upper bound of the confidence score of each article feature data in the group of article data is summed to obtain the total confidence score data;

[0032] The sum of confidence scores corresponding to multiple grouped article data is sorted to obtain a second confidence score sequence;

[0033] Select the data with the largest sum of confidence scores from the second confidence score sequence and determine it as the second target confidence score data;

[0034] Obtain the grouped article data corresponding to the second target confidence data, and determine it as the target recommended article.

[0035] In some embodiments, the feature processing of the article dataset to obtain article feature data includes:

[0036] The article dataset is preprocessed to obtain the target article data;

[0037] Feature extraction is performed on the target article data to obtain the article feature data.

[0038] To achieve the above objectives, a second aspect of this application provides an article recommendation device, the device comprising:

[0039] The article data acquisition module is used to acquire an article dataset and perform feature processing on the article dataset to obtain article feature data; wherein, there are multiple articles with feature data, and each article feature data corresponds to an article;

[0040] The user data acquisition module is used to acquire a user dataset and perform feature processing on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user;

[0041] The recommendation model building module is used to build an article recommendation model based on the article feature data and the user feature data;

[0042] The recommended article determination module is used to determine a target recommended article for each user from multiple article feature data using the article recommendation model;

[0043] The article recommendation module is used to recommend the target articles to the user.

[0044] To achieve the above objectives, a third aspect of the present application provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect.

[0045] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0046] The article recommendation method, apparatus, electronic device, and storage medium proposed in this application obtain article feature data by acquiring an article dataset and performing feature processing on the article dataset to obtain article feature data; wherein there are multiple article feature data, and each article feature data corresponds to an article; acquire a user dataset and perform feature processing on the user dataset to obtain user feature data; wherein there are multiple user feature data, and each user feature data corresponds to a user; construct an article recommendation model based on the article feature data and user feature data, thereby establishing connections between multiple users and realizing article recommendation based on the relevance between users; for each user, determine the target recommendation article from multiple article feature data through the article recommendation model; and recommend the target recommendation article to the user, thereby improving the accuracy and relevance of article recommendation. Attached Figure Description

[0047] Figure 1 This is a flowchart of the article recommendation method provided in the embodiments of this application;

[0048] Figure 2 yes Figure 1 The flowchart of step S101 in the text;

[0049] Figure 3 yes Figure 1 The flowchart of step S104 in the process;

[0050] Figure 4 yes Figure 3 The flowchart of step S303 in the process;

[0051] Figure 5 yes Figure 3 A flowchart of step S304 in the process;

[0052] Figure 6 yes Figure 3 Another flowchart of step S304 in the process;

[0053] Figure 7 This is a flowchart of an article recommendation method provided in another embodiment of this application;

[0054] Figure 8 This is a schematic diagram of the structure of the article recommendation device provided in the embodiments of this application;

[0055] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0059] First, let's analyze some of the terms used in this application:

[0060] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0061] The Bandit Algorithm, also known as the Multi-Armed Bandit Algorithm (MAB), is a reinforcement learning framework used to solve the problem of balancing exploration and exploitation with limited resources. Each arm represents a selectable action or decision, and each arm has an unknown reward distribution. The algorithm estimates the expected reward of each arm by continuously exploring and trying different arms (i.e., different actions or decisions) within a finite number of trials. Based on these estimates, it selects the arm that maximizes the cumulative reward. In this way, the Bandit Algorithm can effectively learn and make optimal decisions in uncertain environments.

[0062] The semi-slot algorithm is a generalized framework for multi-armed slot algorithms. In a semi-slot algorithm, the decision-maker can choose multiple arms in each round and attempts to maximize the cumulative reward within a finite number of attempts. Similar to traditional multi-armed slot algorithms, the semi-slot algorithm also requires a trade-off between exploring new arms and utilizing known high-reward arms.

[0063] The Multi-Task Hierarchical Bayesian algorithm is a machine learning algorithm that combines multi-task learning with the characteristics of hierarchical Bayesian models. By simultaneously learning multiple related tasks, it leverages the correlations between tasks to improve the performance of each task. Furthermore, by utilizing the structured hierarchy of the hierarchical Bayesian model, it constructs a hierarchical prior distribution for the parameters of each task, effectively avoiding overfitting caused by an excessive number of parameters. The Multi-Task Hierarchical Bayesian algorithm not only enhances the model's predictive performance and generalization ability but also estimates the parameters of the posterior distribution using Bayesian methods, providing a powerful solution for complex statistical problems.

[0064] The upper confidence bound (UCB), or the upper limit of a confidence interval, is an important component of the overall confidence interval. In statistics, when conducting sampling surveys or data analysis, since only a portion of the sample data is available and the population data is not, it is necessary to use the sample data to estimate the population. To measure the reliability of the estimate, a confidence interval is used to represent the range of estimated values ​​for the population parameter.

[0065] In the slot machine problem, reward typically refers to the payout or gain obtained after each attempt at a slot arm (action or decision). These rewards are randomly generated, and each arm has its own reward distribution. The goal of slot machine algorithms is to find and continuously select the arms that produce the highest expected reward within a finite number of attempts, using a certain strategy or algorithm, to maximize the cumulative total reward.

[0066] The slot machine problem is a sequential decision problem. In each round of decision-making, an action is selected from the action space, and a random reward (the reward distribution function is unknown) is received. After multiple rounds of decision-making, the difference between the cumulative reward obtained from the series of optimal actions and the cumulative reward actually obtained by the model is defined as "regret". The goal of slot machine learning is to minimize regret or the upper bound of regret by exploring as many different actions as possible and making the most efficient use of past actions that yielded high rewards.

[0067] When multiple slot machine tasks need to be solved, machine learning researchers employ multi-task learning / meta-learning paradigms to achieve task adaptation and knowledge transfer. Current research on multi-task slot machine problems can be divided into three main categories:

[0068] The first approach attempts to learn a low-dimensional representation shared by different slot tasks, thereby obtaining a tighter upper bound on cumulative regret than learning each slot task independently.

[0069] The second type leverages the similarity of context (e.g., features of actions) in slot machine tasks to improve the model's ability to predict rewards in new tasks.

[0070] The third type chooses to maintain a high-level distribution function of the hyperparameters of the slot machine algorithm (such as Tsallis-INF, OFUL, and Thompson sampling) and extract useful hyperparameters from this high-level distribution for new slot machine tasks to achieve efficient regret minimization.

[0071] The article recommendation method provided in this application belongs to the third category, which uses a hierarchical Bayesian model to illustrate the problem of learning similar tasks.

[0072] Typical applications of slot machine algorithms include, but are not limited to, article recommendation, computational advertising, dynamic pricing, autonomous driving, and logistics scheduling.

[0073] In the field of article recommendation, current methods treat each user as an independent task and use traditional single-task algorithms to recommend articles to each user. This approach processes different users individually, ignoring the correlation between them (e.g., users from the same workplace or residential area are likely to have similar reading interests). It fails to facilitate the transfer of beneficial information between different tasks, resulting in low efficiency and reduced accuracy and relevance in article recommendation due to the inability to base recommendations on user correlations.

[0074] Based on this, embodiments of this application provide an article recommendation method, apparatus, electronic device, and storage medium, aiming to improve the accuracy of article recommendations.

[0075] The article recommendation method, apparatus, electronic device, and storage medium provided in this application are specifically described through the following embodiments. First, the article recommendation method in this application is described.

[0076] The article recommendation method provided in this application relates to the field of artificial intelligence technology. The article recommendation method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the article recommendation method, but is not limited to the above forms.

[0077] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0078] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0079] Figure 1This is an optional flowchart of the article recommendation method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0080] Step S101: Obtain the article dataset and perform feature processing on the article dataset to obtain article feature data; wherein, there are multiple article feature data, and each article feature data corresponds to an article;

[0081] Step S102: Obtain the user dataset and perform feature processing on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user;

[0082] Step S103: Construct an article recommendation model based on article feature data and user feature data;

[0083] Step S104: For each user, determine the target recommended article from multiple article feature data using an article recommendation model;

[0084] Step S105: Recommend the target article to the user.

[0085] Steps S101 to S105, as illustrated in this embodiment, involve obtaining an article dataset and performing feature processing on the dataset to obtain article feature data; wherein there are multiple article feature data sets, each corresponding to one article; obtaining a user dataset and performing feature processing on the dataset to obtain user feature data; wherein there are multiple user feature data sets, each corresponding to one user; constructing an article recommendation model based on the article feature data and user feature data, thereby establishing connections between multiple users and implementing article recommendations based on the relevance between users; for each user, determining a target recommended article from multiple article feature data sets using the article recommendation model; and recommending the target recommended article to the user, thereby improving the accuracy and relevance of article recommendations.

[0086] In some embodiments, the article dataset can be article data from online websites, such as news articles, experience-sharing articles, short reviews, etc., but is not limited to this.

[0087] Furthermore, the article dataset contains multiple articles, each with information including, but not limited to, article title, author, publication date, content, and type. Specifically, the required article information needs to be determined based on the actual application scenario, and is not limited to this.

[0088] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S202:

[0089] Step S201: Perform data preprocessing on the article dataset to obtain the target article data;

[0090] Step S202: Extract features from the target article data to obtain article feature data.

[0091] Steps S201 to S202 shown in the embodiments of this application, by preprocessing the article dataset to obtain the target article data, can ensure the quality and effectiveness of the dataset; by extracting features from the target article data and converting it into article feature data suitable for the model, the efficiency and accuracy of the article recommendation model can be significantly improved.

[0092] In step S201 of some embodiments, data preprocessing operations include, but are not limited to, data cleaning and data noise reduction. Specifically, the appropriate preprocessing measures need to be selected based on the actual application scenario, and are not limited to these.

[0093] In step S202 of some embodiments, feature extraction can be performed using deep neural network models, such as CNN, RNN, and LSTM models; or it can be performed using the TF-IDF feature representation method. Specifically, the appropriate feature extraction method needs to be selected based on the actual application scenario, and is not limited to these methods.

[0094] In some embodiments, the user dataset can be user registration information of an online website, and the user dataset contains multiple users. The relevant information of each user includes, but is not limited to, user age, user gender, user region, user personal preferences, etc. Specifically, it needs to be set according to the actual application scenario, and is not limited to this.

[0095] In some embodiments, step S102 may include, but is not limited to, the following steps;

[0096] Preprocess the user dataset to obtain the target user data;

[0097] Feature extraction is performed on the target user data to obtain user feature data.

[0098] By preprocessing the user dataset to obtain the target user data, the quality and validity of the dataset can be ensured. Feature extraction from the target user data and conversion into user feature data suitable for the model can significantly improve the efficiency and accuracy of the article recommendation model.

[0099] Furthermore, the specific implementation of step S102 is basically the same as the specific embodiments of steps S201 to S202 described above, and will not be repeated here.

[0100] In step S103 of some embodiments, an article recommendation model is constructed based on article feature data and user feature data. Specifically, the article recommendation model is a multi-task linear / semi-linear model under hierarchical Gaussian sampling, which can solve the multi-task linear / semi-linear problem under hierarchical Gaussian sampling. By establishing connections between multiple users, article recommendation is achieved based on the correlation between users, thereby improving the accuracy and relevance of article recommendation.

[0101] It should be noted that, in the embodiments of this application, the model must determine one or more target recommended articles for each user, which is referred to as an article recommendation task. The action is the selection of articles, and the reward corresponds to the user's viewing behavior of the target recommended articles.

[0102] In some examples, a detailed theoretical explanation of the multi-task linear slot / semi-slot model under hierarchical Gaussian sampling is provided, as follows:

[0103] First, the multi-task Bayesian linear slot / semi-slot problem assumes that each task s∈[m] (where [m]={1,2,…,m}) is defined by a task parameter θ. s,* To depict.

[0104] In a multi-task linear slot / semi-slot problem, the model will encounter a set of tasks in the t-th iteration. Interact in each task Select action A s,t and receive a random reward Y s,t , where Y s,t Follows distribution

[0105] In a multi-task Bayesian linear slot machine, A s,t It is an action vector, Y s,t It is a random variable;

[0106] In a multi-task Bayesian linear semi-slot machine It is a subset of the action space [K]. It is a series of random variables, among which It is the feature vector corresponding to K actions (known in advance), η s,t It is a K-dimensional random noise.

[0107] The multi-task linear slot / semi-slot problem under the hierarchical Bayesian framework further assumes the task parameter θ. s,* It is independently sampled from the distribution And assume the hyperparameter μ * It is independently sampled from the higher-order prior distribution function Q(μ).

[0108] Therefore, we will take the multi-task linear slot machine problem under hierarchical Gaussian sampling as an example of hierarchical Bayesian slot machine / semi-slot machine problem and illustrate the specific algorithm.

[0109] Specifically, suppose the multi-task linear slot machine problem under hierarchical Gaussian sampling is as follows:

[0110]

[0111]

[0112]

[0113] in, This represents a multivariate Gaussian distribution with mean vector μ and variance matrix Σ, i.e., Gaussian distributed data.

[0114] The form of the multi-task linear semi-slot machine problem under hierarchical Gaussian sampling is very similar to the form described above, the only difference being that formula (3) is... Where <,> represents the inner product of two vectors.

[0115] Simultaneously define article recommendation records H represents the article recommendation records for the model and task s∈[m] before the t-th round. t =(H s,t ) s∈[m] This is a record of article recommendations for all tasks prior to round t.

[0116] Assume the higher-order posterior distribution (Gaussian distribution data) for round t. It is easy to know The corresponding mean vector is:

[0117]

[0118] The corresponding covariance matrices are as follows:

[0119]

[0120] In the multi-task linear slot machine problem under hierarchical Gaussian sampling:

[0121]

[0122]

[0123] In the multi-task linear semi-slot machine problem under hierarchical Gaussian sampling:

[0124]

[0125]

[0126] Where Φ a It is the transpose of the row vector of the a-th row of matrix Φ.

[0127] Therefore, from the higher posterior distribution Q t The hyperparameter μ is obtained by sampling from (μ). t Then, sample the task parameters for task s.

[0128] in It is the mean of the posterior distribution. It is the posterior distribution covariance matrix.

[0129] Finally, through the complete variance decomposition technique, we can see that:

[0130] The Gaussian distribution data is as follows:

[0131] in,

[0132] Please see Figure 3 In some embodiments, step S104 may include, but is not limited to, steps S301 to S304:

[0133] Step S301: Determine the Gaussian distribution data for the article recommendation model;

[0134] Step S302: Sample the Gaussian distribution data to obtain distribution characteristic data;

[0135] Step S303: Calculate the upper bound of the confidence level for each article's feature data based on the distribution feature data;

[0136] Step S304: Determine the target recommended article from multiple article feature data based on the upper bound of the confidence level.

[0137] Steps S301 to S304, as illustrated in this embodiment, calculate the upper bound of the confidence score for each article feature data using the distribution feature data of the high-level distribution data, and then determine the target recommended article from multiple article feature data based on the upper bound of the confidence score, thereby improving the accuracy of article recommendation. Compared with the currently commonly used hierarchical Bayes-Thompson sampling, this application does not require calculating the posterior probability distribution function for the task parameters, which can greatly simplify the program's running complexity and achieve less regret.

[0138] It should be noted that if the article recommendation model is performing the article recommendation task for the first time, the Gaussian distribution function is a pre-set high-level prior Gaussian distribution function; if the article recommendation model is not performing the article recommendation task for the first time, the Gaussian distribution function is a high-level posterior Gaussian distribution function adjusted based on the recommendation results of previous articles, thereby achieving task adaptation and knowledge transfer.

[0139] In step S302 of some embodiments, the distribution feature data includes distribution mean feature data and distribution variance feature data.

[0140] Specifically, from the Gaussian distribution data Q t The hyperparameter μ is obtained by sampling from (μ). t Then, we obtained:

[0141] Distribution mean characteristic data:

[0142] Distribution variance characteristic data:

[0143] Please see Figure 4 In some embodiments, step S303 may include, but is not limited to, steps S401 to S403:

[0144] Step S401: Calculate the first feature based on the article feature data and the distribution mean feature data to obtain the first feature data;

[0145] Step S402: Calculate the second feature based on the preset confidence factor, article feature data, and distribution variance feature data to obtain the second feature data;

[0146] Step S403: Perform aggregation calculation based on the first feature data and the second feature data to obtain the upper bound of the confidence level.

[0147] Steps S401 to S403 of this embodiment involve calculating a first feature based on article feature data and distribution mean feature data to obtain first feature data; calculating a second feature based on a preset confidence factor, article feature data, and distribution variance feature data to obtain second feature data; and performing aggregation calculation based on the first and second feature data to obtain the upper bound of the confidence score. Compared to the commonly used hierarchical Bayes-Thompson sampling, this application does not require calculating the posterior probability distribution function for the task parameters, which greatly simplifies the program's operational complexity and achieves less regret.

[0148] For a multi-task linear slot / semi-slot model under hierarchical Gaussian sampling, in the t-th iteration, for each user's corresponding article recommendation task... Calculate each action The upper bound of the confidence level:

[0149]

[0150] Where a represents the article's feature data, and δ∈(0,1) is the confidence factor.

[0151] In some other embodiments, the upper bound of the confidence level can also be set to The form is given by , where c is a variable parameter that can vary depending on the specific settings, thus giving the hierarchical Bayesian UCB algorithm greater flexibility and applicability.

[0152] Please see Figure 5 In some embodiments, step S304 may include, but is not limited to, steps S501 to S503:

[0153] Step S501: Sort the upper bounds of the confidence scores corresponding to multiple article feature data to obtain the first confidence score sequence;

[0154] Step S502: Select the upper bound of the confidence level with the largest value from the first confidence level sequence and determine it as the first target confidence level data;

[0155] Step S503: Obtain the article feature data corresponding to the first target confidence data and determine it as the target recommended article.

[0156] Steps S501 to S503, as illustrated in this embodiment, are for a multi-task linear slot machine model under hierarchical Gaussian sampling. By sorting the upper bounds of confidence scores corresponding to multiple article feature data, a first confidence score sequence is obtained. The upper bound with the largest value is selected from the first confidence score sequence and determined as the first target confidence score data. The article feature data corresponding to the first target confidence score data is obtained and determined as the target recommended article. This accurately selects the target recommended article for each user, improving the accuracy of article recommendation. Furthermore, since the Gaussian distribution data (high-order Gaussian distribution function) of the hyperparameters of the multi-task linear slot machine model under hierarchical Gaussian sampling combines the recommended articles and rewards for each user, a connection between multiple users can be established, enabling article recommendation based on the correlation between users, thus improving the relevance of article recommendations.

[0157] Specifically, each article feature data has a corresponding upper bound for confidence. After sorting the upper bounds of confidence for multiple article feature data according to their numerical values, a first confidence sequence is obtained. Then, the upper bound with the largest value is selected from the first confidence sequence. Use this as the first target confidence data; obtain the article feature data corresponding to the first target confidence data, and determine it as the target recommended article A. s,t .

[0158] It should be noted that sorting can be done by numerical value from largest to smallest or from smallest to largest. The choice should be made based on the specific application scenario and is not limited to this.

[0159] Please see Figure 6 In some embodiments, step S304 may also include, but is not limited to, steps S601 to S605:

[0160] Step S601: Group the multiple article feature data to obtain at least one group of article data;

[0161] Step S602: For each group of article data, sum the upper bound of the confidence level of each article feature data in the group of article data to obtain the total confidence level data;

[0162] Step S603: Sort the total confidence scores of multiple grouped articles to obtain the second confidence score sequence;

[0163] Step S604: Select the data with the largest total confidence score from the second confidence score sequence and determine it as the second target confidence score data;

[0164] Step S605: Obtain the grouped article data corresponding to the second target confidence data and determine it as the target recommended article.

[0165] Steps S601 to S605 of this embodiment, for a multi-task linear semi-slot machine model under hierarchical Gaussian sampling, involve grouping multiple article feature data to obtain at least one group of article data. For each group of article data, the upper bound of the confidence score of each article feature data in the group of article data is summed to obtain a total confidence score. The total confidence scores corresponding to multiple grouped article data are sorted to obtain a second confidence score sequence. The total confidence score with the largest value is selected from the second confidence score sequence and determined as the second target confidence score. The grouped article data corresponding to the second target confidence score is obtained and determined as the target recommended article. This accurately selects multiple target recommended articles for each user, improving the accuracy of article recommendation. In addition, since the Gaussian distribution data (high-order Gaussian distribution function) of the hyperparameters of the multi-task linear semi-slot machine model under hierarchical Gaussian sampling combines the recommended articles and rewards for each user, it is possible to establish connections between multiple users and realize article recommendation based on the correlation between users, thus improving the relevance of article recommendation.

[0166] In step S601 of some embodiments, multiple article feature data are grouped according to tags such as article type / domain to form at least one group of article data. The number of article features contained in each group of article data may not be the same. The specific tags used for classification need to be selected based on the actual application scenario, and are not limited thereto.

[0167] Then, for each group of article data, the upper bound of the confidence score corresponding to each article feature data in the group of article data is summed to obtain the total confidence score data.

[0168] The total confidence score can be represented as: Where A represents the grouped article data;

[0169] Next, the sum of confidence scores for the multiple grouped articles is sorted by numerical value to obtain a second confidence score sequence. Then, the sum of confidence scores with the largest value is selected from the second confidence score sequence. And it was determined as the confidence level data for the second objective;

[0170] Finally, the grouped article data corresponding to the second target confidence data is obtained and determined as the target recommendation article set A. s,t .

[0171] It should be noted that sorting can be done by numerical value from largest to smallest or from smallest to largest. The choice should be made based on the specific application scenario and is not limited to this.

[0172] This application's embodiments extend the confidence upper bound calculation method of the hierarchical Bayesian UCB algorithm from "the upper bound of a single article feature data" to "the sum of the upper bounds of multiple article feature data in grouped article data," thus extending the hierarchical Bayesian UCB algorithm to a hierarchical Bayesian UCB semi-automatic setting. This allows the article recommendation method based on the hierarchical Bayesian algorithm to have a wider range of application scenarios.

[0173] Please see Figure 7 In some embodiments, step S106 may include, but is not limited to, steps S701 to S704:

[0174] Step S701: Obtain user feedback information; wherein, the feedback information is used to characterize the user's viewing behavior of the target recommended article;

[0175] Step S702: Construct recommendation reward information based on feedback information and user characteristic data;

[0176] Step S703: Update the article recommendation record based on the target recommended article and recommendation reward information; wherein, the article recommendation record includes each user's historical recommended articles and the corresponding recommendation reward information for the historical recommended articles, and the historical recommended articles are determined through the article recommendation model;

[0177] Step S704: Update the Gaussian distribution data based on the updated article recommendation records to update the article recommendation model; wherein, the updated article recommendation model is used for the next article recommendation.

[0178] Steps S701 to S704, as illustrated in the embodiments of this application, whether in a multi-task linear model under hierarchical Gaussian sampling or a multi-task linear semi-linear model under hierarchical Gaussian sampling, collect user viewing behavior on one or more target recommended articles as feedback information. Combined with user feature data to construct recommendation reward information, the recommendation effect can be evaluated in real time, and the recommendation strategy can be adjusted. This allows for continuous self-optimization and updating of article recommendation records, thereby improving the accuracy and personalization of recommendations. Furthermore, by updating the Gaussian distribution data in the article recommendation model, user preferences and behavioral patterns can be continuously learned, providing more accurate and efficient candidate articles for the next article recommendation. This closed-loop feedback learning mechanism helps to build a continuously evolving article recommendation model, thereby improving the accuracy of article recommendations.

[0179] In some embodiments, the referral reward information is denoted as Y. s,t ;

[0180] Since the multi-task linear semi-semi ...

[0181]

[0182] For a given user, the article recommendation record is updated based on the target recommended articles and recommendation reward information determined for that user. Specifically:

[0183] in, This refers to all article recommendation records for this user prior to the t-th article recommendation.

[0184] After all the article recommendation records for all users have been updated, the total number of article recommendation records H is obtained. t+1 =(H s,t+1 ) s∈[n] Specifically, this refers to the article recommendation records of all users prior to the t-th article recommendation.

[0185] Next, based on the updated article recommendation record Ht+1 For Gaussian distributed data Q t An update is performed to update the article recommendation model; specifically, the updated Gaussian distribution data is denoted as...

[0186] It should be noted that the updated article recommendation model is used for the next article recommendation. In the next article recommendation, if there are new users, knowledge transfer can be performed based on the advanced posterior distribution function built from the previously executed article recommendation task. This can alleviate the cold start problem of the model and improve the accuracy and relevance of article recommendations, effectively improving the recommendation quality for new users.

[0187] In some embodiments, the regret bound verification is performed on the multi-task linear slot / semi-slot model under hierarchical Gaussian sampling proposed in this application. The specific verification process is as follows:

[0188] First, define For any ε > 0, then perform a multi-task Bayesian regret. Decomposed into three parts:

[0189]

[0190] For the last two terms 1{Δ s,t <∈,E s,t}and It can be used To control.

[0191] For the first item, the confidence upper bound calculation method provided in the embodiments of this application can be used to control the upper bound.

[0192] Then through algebraic control Size;

[0193] We can obtain a logarithmic regret bound of order of magnitude O(mlog(mn)logn) (the logarithmic regret bound for the number of iterations n in each task, in the sense of the average regret bound).

[0194] For any iteration t, let |S t |=1, assuming the action space A is finite, then in the setting of a multi-task Gaussian linear slot machine, for any δ∈(0,1), ∈>0, the Bayesian regret of the hierarchical Bayesian UCB algorithm. The upper bound is as follows:

[0195]

[0196] Where c1 and c2 are some constants. λ1(M) represents the largest eigenvalue of matrix M. Let δ = 1 / (mn), ∈ = 1 / (mn), then the order of magnitude of the aforementioned regret bound is O(mlog(mn)logn).

[0197] Furthermore, for any iteration t, let |S t |=1, assuming the action space A is finite, in the setting of a multi-task Gaussian linear semi-slot machine, for any δ∈(0,1), ∈>0, the Bayesian regret of the hierarchical Bayesian UCB algorithm. The upper bound is as follows:

[0198]

[0199] Where c4 is a certain constant.

[0200] In some embodiments, the regret bounds of Bayesian slot machine algorithms at different levels are compared. The Bayesian regret bounds of Bayesian slot machine algorithms at different levels under the multi-task d-dimensional linear (or K-arm) slot machine problem are shown in Table 1 below.

[0201] Where m is the number of tasks, n is the number of iterations for each task, and A is the action space. The Bayesian regret bound = bound I + bound II + negligible terms, where bound I is the regret bound for solving m tasks, and bound II is the hyperparameter μ. * The regret boundary, K is the size of the action space A.

[0202]

[0203] Table 1

[0204] In addition, the regret bounds of Bayesian semi-semi ...

[0205] Where m is the number of tasks, n is the number of iterations for each task, and A is the action space. The Bayesian regret bound = bound I + bound II + negligible terms, where bound I is the regret bound for solving m tasks, and bound II is the hyperparameter μ. * The regret bound is defined by K, where K is the size of the action space A, and L is the number of actions selected in each iteration (1≤L≤K).

[0206]

[0207] Table 2

[0208] The multi-task linear / semi-slot machine model with hierarchical Gaussian sampling proposed in this application can achieve a better upper bound on confidence. In the hierarchical Bayesian UCB slot machine algorithm, it utilizes the high-level posterior distribution Q. t mean and variance Calculate each action upper bound of confidence Then select the action with the largest upper bound of confidence. Compared to the Hierarchical Thompson sampling algorithm (HierTS), it first calculates the task parameters θ using Bayes' theorem. s,t The posterior probability distribution function, from which θ is sampled. s,t Then select an action. Compared to hierarchical Bayesian Thompson sampling, the hierarchical sampling proposed in this application's embodiments...

[0209] The Bayesian UCB algorithm does not require computation of the task parameter θ. s,t The posterior probability distribution function can greatly simplify the program's execution complexity. Furthermore, theoretical verification shows that the hierarchical Bayesian UCB algorithm achieves less regret than the hierarchical Thompson sampling algorithm.

[0210] Furthermore, in the theoretical analysis of the Hierarchical Bayesian UCB slot machine algorithm, this application's embodiments decompose regret into three parts: regret when the upper bound of confidence holds and the slot machine task distribution interval is greater than a certain threshold, regret when the upper bound of confidence holds and the slot machine task distribution interval is not greater than a certain threshold, and regret when the upper bound of confidence does not hold. Using this regret bound analysis method, this application provides an improved regret bound for the Hierarchical Bayesian UCB slot machine algorithm.

[0211] Furthermore, the embodiments of this application can extend the confidence upper bound calculation method of the hierarchical Bayesian UCB algorithm from "the upper bound of a single article feature data" to "the sum of the upper bounds of multiple article feature data in grouped article data", thus extending the hierarchical Bayesian UCB algorithm to the hierarchical Bayesian UCB semi-engine setting, thereby enabling the article recommendation method based on the hierarchical Bayesian algorithm to have a wider range of application scenarios.

[0212] Furthermore, the hierarchical Bayesian algorithm in the multi-task semi-computer setting provided in this application embodiment can include not only the Bayesian UCB algorithm, but also the Thompson sampling algorithm or other Bayesian algorithm. For example, the Hierarchical Thompson sampling algorithm (HierTS) can be extended to the semi-computer setting: firstly, the task parameter θ is calculated using Bayes' rule. s,t The posterior probability distribution function, from which θ is sampled. s,t Then, use offline tools to calculate the optimal set of actions. Thus, the hierarchical Thompson sampling algorithm under the multi-task semi-slot machine setting is obtained.

[0213] It should be noted that the multi-task linear slot machine / semi-slot machine model under hierarchical Gaussian sampling provided in this application embodiment can be applied not only to the article recommendation application scenario proposed in this application, but also to application scenarios such as clinical trials and dynamic product pricing.

[0214] For example, in clinical trial applications:

[0215] Expected goals: To improve the accuracy of evaluating the effectiveness of treatment regimens in clinical trials, promote effective drug development, and improve patient recovery efficiency.

[0216] Input: Representation vectors (i.e., action vectors) of different treatment options (such as different drugs or different drug dosages), and hierarchical Bayesian modeling of the relationships between different patients.

[0217] Process: By modeling the relationships between different patients as a hierarchical Bayesian model, the hierarchical Bayesian UCB algorithm is used to select different treatment plans for different patients, record the patients' feedback on the treatment plans, and update the statistical characteristics of the hierarchical Bayesian model (such as the mean and variance of the high-level prior distribution) in real time based on historical records.

[0218] Results: By using a hierarchical Bayesian model to facilitate the transfer of beneficial information among different patients (especially those with similar diseases), the accuracy of evaluating the effectiveness of treatment regimens in clinical trials can be improved, thereby enhancing patient recovery efficiency.

[0219] In dynamic product pricing application scenarios:

[0220] Expected goals: To improve the accuracy of product pricing in response to market demand, thereby boosting product sales and increasing the economic benefits for product manufacturers.

[0221] Input: Different pricing for different products and corresponding sales feedback, feature vector representations of different products.

[0222] Process: By modeling the relationships between different products as a hierarchical Bayesian model, the hierarchical Bayesian UCB algorithm is used to dynamically price different products, record market feedback on the pricing of various products, and update the statistical characteristics of the hierarchical Bayesian model (such as the mean and variance of the high-level prior distribution) in real time based on historical records.

[0223] Results: The hierarchical Bayesian model enables the transfer of useful information between different products (especially similar products), improves the accuracy of product pricing, promotes product sales, and enhances the economic benefits for product manufacturers.

[0224] Please see Figure 8 This application also provides an article recommendation device that can implement the above-described article recommendation method. The device includes:

[0225] The article data acquisition module 801 is used to acquire an article dataset and perform feature processing on the article dataset to obtain article feature data; there are multiple article feature data, and each article feature data corresponds to an article;

[0226] The user data acquisition module 802 is used to acquire a user dataset and perform feature processing on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user;

[0227] Recommendation model building module 803 is used to build an article recommendation model based on article feature data and user feature data;

[0228] The article recommendation determination module 804 is used to determine the target recommended article for each user from multiple article feature data through an article recommendation model;

[0229] The article recommendation module 805 is used to recommend target articles to users.

[0230] The specific implementation of the device recommended in this article is basically the same as the specific embodiment of the method recommended in the above article, and will not be repeated here.

[0231] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method recommended in the article above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0232] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0233] The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0234] The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the method recommended in the embodiments of this application.

[0235] The input / output interface 903 is used to implement information input and output;

[0236] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0237] Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904);

[0238] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0239] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method recommended in the article described above.

[0240] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0241] The article recommendation method, apparatus, electronic device, and storage medium provided in this application embodiment acquire an article dataset and perform feature processing on the article dataset to obtain article feature data; wherein the number of article feature data is multiple, and each article feature data corresponds to an article; acquire a user dataset and perform feature processing on the user dataset to obtain user feature data; wherein the number of user feature data is multiple, and each user feature data corresponds to a user; construct an article recommendation model based on the article feature data and user feature data, thereby establishing connections between multiple users and realizing article recommendation based on the relevance between users; for each user, determine the target recommendation article from multiple article feature data through the article recommendation model; recommend the target recommendation article to the user, thereby improving the accuracy and relevance of article recommendation.

[0242] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0243] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0244] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0245] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0246] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0247] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0248] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The coupling or direct coupling or communication connection between the shown or discussed units may be through some interfaces, or indirect coupling or communication connection between the apparatus or units, and may be electrical, mechanical, or other forms.

[0249] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0250] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0251] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0252] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An article recommendation method, characterized in that, The method includes: Obtain an article dataset and perform feature processing on the article dataset to obtain article feature data; wherein, there are multiple articles with feature data, and each article feature data corresponds to an article; A user dataset is acquired, and feature processing is performed on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user; An article recommendation model is constructed based on the article feature data and the user feature data; For each user, the target recommended article is determined from multiple article feature data using the article recommendation model; The target recommended article is recommended to the user; The step of determining the target recommended article from multiple article feature data using the article recommendation model includes: Determine the Gaussian distribution data for the article recommendation model; The Gaussian distribution data is sampled to obtain distribution characteristic data; Calculate the upper bound of the confidence level for each of the article feature data based on the distribution feature data; The target recommended article is determined from multiple article feature data based on the upper bound of the confidence level; The distribution feature data includes distribution mean feature data and distribution variance feature data; the step of calculating the upper bound of the confidence level for each article feature data based on the distribution feature data includes: Based on the article feature data and the distribution mean feature data, a first feature calculation is performed to obtain the first feature data; The second feature is calculated based on the preset confidence factor, the article feature data, and the distribution variance feature data to obtain the second feature data; The confidence upper bound is obtained by performing aggregation calculations based on the first feature data and the second feature data.

2. The method according to claim 1, characterized in that, After recommending the target article to the user, the method further includes: Obtain feedback information from the user; wherein the feedback information is used to characterize the user's viewing behavior of the target recommended article; Recommendation reward information is constructed based on the feedback information and the user feature data; The article recommendation record is updated based on the target recommended article and the recommendation reward information; wherein, the article recommendation record includes the historical recommended articles of each user and the recommendation reward information corresponding to the historical recommended articles, and the historical recommended articles are determined by the article recommendation model; The Gaussian distribution data is updated based on the updated article recommendation records to update the article recommendation model; wherein, the updated article recommendation model is used for the next article recommendation.

3. The method according to claim 1, characterized in that, The step of determining the target recommended article from multiple article feature data based on the upper bound of the confidence score includes: The confidence upper bounds corresponding to multiple article feature data are sorted to obtain a first confidence sequence; Select the upper bound of the confidence level with the largest value from the first confidence level sequence and determine it as the first target confidence level data; Obtain the article feature data corresponding to the first target confidence data, and determine it as the target recommended article.

4. The method according to claim 1, characterized in that, The step of determining the target recommended article from multiple article feature data based on the upper bound of the confidence score includes: The multiple article feature data are grouped to obtain at least one group of article data; For each group of article data, the upper bound of the confidence score of each article feature data in the group of article data is summed to obtain the total confidence score data; The sum of confidence scores corresponding to multiple grouped article data is sorted to obtain a second confidence score sequence; Select the data with the largest sum of confidence scores from the second confidence score sequence and determine it as the second target confidence score data; Obtain the grouped article data corresponding to the second target confidence data, and determine it as the target recommended article.

5. The method according to any one of claims 1 to 4, characterized in that, The step of performing feature processing on the article dataset to obtain article feature data includes: The article dataset is preprocessed to obtain the target article data; Feature extraction is performed on the target article data to obtain the article feature data.

6. An article recommendation device, characterized in that, The device includes: The article data acquisition module is used to acquire an article dataset and perform feature processing on the article dataset to obtain article feature data; wherein, there are multiple articles with feature data, and each article feature data corresponds to an article; The user data acquisition module is used to acquire a user dataset and perform feature processing on the user dataset to obtain user feature data; wherein, there are multiple user feature data, and each user feature data corresponds to one user; The recommendation model building module is used to build an article recommendation model based on the article feature data and the user feature data; The recommended article determination module is used to determine a target recommended article for each user from multiple article feature data using the article recommendation model; The article recommendation module is used to recommend the target articles to the user; The step of determining the target recommended article from multiple article feature data using the article recommendation model includes: Determine the Gaussian distribution data for the article recommendation model; The Gaussian distribution data is sampled to obtain distribution characteristic data; Calculate the upper bound of the confidence level for each of the article feature data based on the distribution feature data; The target recommended article is determined from multiple article feature data based on the upper bound of the confidence level; The distribution feature data includes distribution mean feature data and distribution variance feature data; the step of calculating the upper bound of the confidence level for each article feature data based on the distribution feature data includes: Based on the article feature data and the distribution mean feature data, a first feature calculation is performed to obtain the first feature data; The second feature is calculated based on the preset confidence factor, the article feature data, and the distribution variance feature data to obtain the second feature data; The confidence upper bound is obtained by performing aggregation calculations based on the first feature data and the second feature data.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 5.

Images