CN114936324A - News recommendation method, device, terminal device and storage medium - Google Patents

Abstract

The application discloses a news recommendation method, a device, a terminal device and a storage medium, wherein the news recommendation method comprises the following steps: acquiring attribute information and historical behavior information of a user; acquiring the historical behavior quantity of the user according to the historical behavior information; classifying the users according to the historical behavior quantity to obtain a user classification result; and according to the user classification result, aiming at different user classifications, recommending news to the user by adopting a corresponding unsupervised clustering method based on the self attribute information or historical behavior information of the user. The method and the system provide the unsupervised clustering news recommendation scheme for the mobile device user without accumulating a large amount of user browsing data and without investing manpower to label the data, and solve the problem of high cost investment of news recommendation.

Description

News recommendation method, device, terminal device and storage medium

technical field

The present application relates to the field of mobile Internet technologies, and in particular, to a news recommendation method, apparatus, terminal device, and storage medium.

Background technique

With the popularity of mobile devices and the increasing number of news readers, mobile news recommendation has become one of the hot spots in the field of mobile recommendation. How to recommend news according to user characteristics and news characteristics to improve recommendation performance and user satisfaction has become the main task of news recommendation system. Traditional news recommendation methods first need to accumulate a large amount of user browsing data, invest manpower to label the data, and then train a supervised learning model to achieve news recommendation. During the process of realizing this application, the inventor found that this method requires a lot of data cost and labor cost, and it is difficult to implement news recommendation for a small-volume user such as a mobile device, which makes the input-output ratio seriously out of balance. Traditional supervised news recommendation methods have been difficult to adapt to new business models.

Therefore, it is necessary to propose a low-cost news recommendation solution.

SUMMARY OF THE INVENTION

The main purpose of this application is to provide a news recommendation method, device, terminal device and storage medium, which aims to provide mobile device users with a news recommendation solution that does not need to accumulate a large amount of user browsing data, and does not need to invest manpower to label the data, so as to solve the problem of news recommendation. the high cost of investment.

In order to achieve the above purpose, the present application provides a method for recommending news, the method for recommending news includes:

Obtain the user's own attribute information and historical behavior information;

Obtain the number of historical behaviors of the user according to the historical behavior information;

Classify users according to the number of historical behaviors to obtain user classification results;

According to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is adopted to recommend news to the user.

Optionally, the step of classifying users according to the number of historical behaviors, and obtaining a user classification result includes:

comparing the number of historical behaviors with a preset threshold;

Divide users whose number of historical behaviors is greater than or equal to the preset threshold as Class A users;

Divide the users whose number of historical behaviors is greater than zero and less than the preset threshold into Class B users;

The users whose number of historical behaviors is equal to zero are divided into C-type users.

Optionally, according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, the step of adopting the corresponding unsupervised clustering method to recommend news to the user includes:

If the user classification result is a C-type user, according to the user's own attribute information, and adopting a recommendation strategy based on user information, the corresponding news is recommended to the user.

Optionally, according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, the step of adopting the corresponding unsupervised clustering method to recommend news to the user includes:

If the user classification result is a B-type user, according to the user's own attribute information or historical behavior information, and adopt the recommendation strategy based on user information, the recommendation strategy based on the user's attention to the company, and the recommendation strategy based on the user's browsing news, The user recommends the corresponding news.

Optionally, according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, the step of adopting the corresponding unsupervised clustering method to recommend news to the user includes:

If the user classification result is a category A user, then according to the user's historical behavior information, a preset recommendation algorithm is adopted and a recommendation strategy based on the user's news browsing is used to recommend corresponding news to the user.

Optionally, the user's own attribute information includes user region information and department information, and the recommendation strategy based on user information includes: recommending corresponding popular news recommendations according to the user region information and/or department information. Strategy.

Optionally, the historical behavior information includes information about the user's following companies, and the recommendation strategy based on the user's following companies includes: recommending a recommendation strategy for the current day's news of the corresponding company according to the following company information.

Optionally, the historical behavior information includes the user's browsing news information, collecting news information, and searching for news information, and the recommendation strategy based on the user's browsing news includes:

If the user classification result is a B-type user, recommending a recommendation strategy for other news strongly associated with the browsing news information according to the browsing news information;

If the user classification result is a category A user, a recommendation algorithm is used to perform a recommendation strategy for news recommendation according to the browsing news information, saving news information and searching for news information.

Optionally, according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, the step of adopting the corresponding unsupervised clustering method to recommend news to the user further includes:

Get candidate recommendation news;

Obtain corresponding news factors according to the candidate recommended news;

Set corresponding weights according to the news factors to obtain news factors with different weights;

According to the news factors of different weights, comprehensively sort the candidate recommended news according to the weights;

Perform a deduplication operation on the candidate recommended news according to the user's historical behavior information to obtain a news recommendation list;

Recommend news to the user according to the news recommendation list.

Optionally, after the step of recommending news to the user according to the news recommendation list, the method further includes:

Obtain the news click ratio and corresponding browsing duration of the user from the recommended news portal;

The recommendation quality is monitored according to the news click ratio and the corresponding browsing time.

Optionally, after the step of recommending news to the user according to the news recommendation list, the method further includes:

Obtain the user's operation behavior information on the recommended news list;

setting a punishment mechanism according to the operation behavior information;

According to the penalty mechanism, the weight of the news factor of the candidate recommended news of the user is adjusted.

The embodiment of the present application further provides a news recommendation device, and the news recommendation device includes:

The information acquisition module is used to acquire the user's own attribute information and historical behavior information;

A quantity acquisition module, used for acquiring the historical behavior quantity of the user according to the historical behavior information;

A user classification module, configured to classify users according to the number of historical behaviors to obtain user classification results;

The news recommendation module is configured to, according to the user classification result, for different user classifications, and based on the user's own attribute information or historical behavior information, adopt a corresponding unsupervised clustering method to recommend news to the user.

An embodiment of the present application further provides a terminal device, the terminal device includes a memory, a processor, and a news recommendation program stored on the memory and running on the processor, where the news recommendation program is processed by the The steps of implementing the above-mentioned news recommendation method when the server is executed.

Embodiments of the present application further provide a computer-readable storage medium, where a news recommendation program is stored on the computer-readable storage medium, and when the news recommendation program is executed by a processor, the steps of the above-mentioned news recommendation method are implemented.

The news recommendation method, device, terminal device and storage medium proposed in the embodiments of the present application obtain the user's own attribute information and historical behavior information; acquire the user's historical behavior quantity according to the historical behavior information; Classification is performed to obtain a user classification result; according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is adopted to recommend news to the user. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Description of drawings

FIG. 1 is a schematic diagram of functional modules of a terminal device to which the news recommendation apparatus of the application belongs;

FIG. 2 is a schematic flowchart of a first exemplary embodiment of a news recommendation method of the present application;

3 is a schematic flowchart of a second exemplary embodiment of a news recommendation method of the present application;

FIG. 4 is a schematic flowchart of a seventh exemplary embodiment of the news recommendation method of the present application;

FIG. 5 is a schematic flowchart of an eighth exemplary embodiment of the news recommendation method of the present application;

6 is a schematic flowchart of a ninth exemplary embodiment of a news recommendation method of the present application;

FIG. 7 is a schematic overall flow diagram of a tenth exemplary embodiment of a news recommendation method of the present application.

The realization, functional characteristics and advantages of the purpose of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed ways

It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

The main solutions of the embodiments of the present application are: obtaining the user's own attribute information and historical behavior information; obtaining the user's historical behavior quantity according to the historical behavior information; classifying the user according to the historical behavior quantity, and obtaining the user classification result; According to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is adopted to recommend news to the user. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Technical terms involved in the embodiments of this application:

Supervised learning, Supervised Learning;

Unsupervised learning, Unsupervised Learning;

Association Rules Algorithm, The Association Rules Algorithm;

Collaborative Filtering Recommendation, Collaborative Filtering Recommendation;

Content-Based Recommendations;

Deep Learning, Deep Learning.

Among them, supervised learning, also known as supervised learning, supervised training or teacher learning, is a machine learning task that infers a function from labeled training data. The process of adjusting the parameters of the classifier to achieve the required performance using a set of samples of known classes. According to the existing dataset, know the relationship between the input and output results. According to this known relationship, an optimal model is obtained by training. That is to say, in supervised learning, the training data has both features and labels. Through training, the machine can find the connection between features and labels by itself. When faced with data with only features and no labels, Labels can be determined. Supervised learning has an application scenario where the output of a trained model can be a continuous value (called regression analysis), or predict a categorical label (called classification). In practical applications, the acquisition of classification labels often requires enormous manual workload, sometimes even very difficult.

Unsupervised learning is a method of machine learning that automatically classifies or groups input data without given pre-labeled training examples. The process of unsupervised learning can be understood as: we do not know the relationship between data and features in the data set, but obtain the relationship between data according to a cluster or a certain model. It can be said that compared to supervised learning, unsupervised learning is more like self-learning, letting machines learn to do things by themselves, without labels. Among them, as an important method of unsupervised learning, the idea of clustering is to classify samples with similar attributes into one category. For each data point, we can assign it to a specific class, and all data points between each class have some commonality, such as spatial proximity and other characteristics.

Association Rules Algorithm (The Association Rules Algorithm), an association rule is a rule whose support and confidence satisfy the user's given threshold. The so-called association reflects the knowledge of dependencies or associations between an event and other events. An association rule is an implication of the form X→Y, where X and Y are called the predecessor (LHS, left-hand-side) and successor (RHS, right-hand-side) of the association rule, respectively. Association rules are defined as: Suppose I is a set of items. Given a set consisting of items, called itemset D, where each transaction T(Transaction) is a non-empty subset of I, that is, each item corresponds to a unique identifier TID(TransactionID). The support of the association rule in D is the percentage of transactions in D that contain both X and Y, that is, the probability; the confidence is the percentage of Y that is included in the transaction in D when X is already included, that is, the conditional probability. The support is a percentage, which refers to the ratio between the number of occurrences of the X and Y combinations and the total number of times. The higher the support, the greater the frequency of the combination. Confidence is a conditional probability, which refers to how likely it is that Y will occur when X occurs.

Collaborative filtering recommendation (Collaborative Filtering Recommendation) analyzes user interests, finds similar (interested) users of the specified user in the user group, and integrates the evaluation of these similar users for a certain information to form the system's preference for the specified user. predict.

Content-Based Recommendations (Content-Based Recommendations) is to build a recommendation algorithm model based on the target-related information and user-related information, that is, the user's operation behavior on the target, and provide recommendation services for users. The target-related information may be metadata information, tags, user comments, manually annotated information, and the like describing the text of the target. User-related information refers to demographic information (eg, age, gender, preferences, region, income, etc.). The user's operation behavior on the target object may be comments, favorites, likes, browsing, clicks, etc. Content-based recommendation algorithms generally only rely on the user's own behavior to provide recommendations for users, and do not involve the behavior of other users.

Deep learning is a new branch of machine learning and an important class of methods. Its main idea is to learn multi-layer artificial neural networks to obtain automatic feature extraction and pattern recognition capabilities. Deep learning combines low-level features to form more abstract high-level representation attributes or features to discover distributed feature representations of data. Deep learning can characterize massive data related to users and items by learning a deep nonlinear network structure. Through automatic feature learning from multi-source heterogeneous data, deep learning can map different data to the same latent space, and can obtain a unified representation of the data. Applying deep learning to recommender systems can better understand user needs, item characteristics, and historical interactions between them.

Specifically, referring to FIG. 1 , FIG. 1 is a schematic diagram of functional modules of a terminal device to which the news recommendation apparatus of the present application belongs. The news recommendation apparatus may be an apparatus independent of the terminal device, capable of performing news recommendation and network model training, and may be carried on the terminal device in the form of hardware or software. The terminal device may be an intelligent mobile terminal with a data processing function, such as a mobile phone or a tablet computer, or a fixed terminal device or a server with a data processing function.

In this embodiment, the terminal device to which the news recommendation apparatus belongs includes at least an output module 110 , a processor 120 , a memory 130 and a communication module 140 .

An operating system and a news recommendation program are stored in the memory 130, and the news recommendation device can obtain the obtained user's own attribute information and historical behavior information, the user's historical behavior quantity obtained through the historical behavior information, and classify the user according to the historical behavior quantity. The user classification results and the corresponding unsupervised clustering algorithms used for different user classifications based on the user's own attribute information or historical behavior information are stored in the memory 130; the output module 110 may be a display screen or the like. The communication module 140 may include a WIFI module, a mobile communication module, a Bluetooth module, etc., and communicate with an external device or a server through the communication module 140 .

The following steps are implemented when the news recommendation program in the memory 130 is executed by the processor:

Obtain the user's own attribute information and historical behavior information;

Obtain the number of historical behaviors of the user according to the historical behavior information;

Classify users according to the number of historical behaviors to obtain user classification results;

According to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is adopted to recommend news to the user.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

comparing the number of historical behaviors with a preset threshold;

Divide users whose number of historical behaviors is greater than or equal to the preset threshold as Class A users;

Divide the users whose number of historical behaviors is greater than zero and less than the preset threshold into Class B users;

The users whose number of historical behaviors is equal to zero are divided into C-type users.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

If the user classification result is a C-type user, according to the user's own attribute information, and adopting a recommendation strategy based on user information, the corresponding news is recommended to the user.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

If the user classification result is a B-type user, according to the user's own attribute information or historical behavior information, and adopt the recommendation strategy based on user information, the recommendation strategy based on the user's attention to the company, and the recommendation strategy based on the user's browsing news, The user recommends the corresponding news.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

If the user classification result is a category A user, then according to the user's historical behavior information, a preset recommendation algorithm is adopted and a recommendation strategy based on the user's news browsing is used to recommend corresponding news to the user.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

Get candidate recommendation news;

Obtain corresponding news factors according to the candidate recommended news;

Set corresponding weights according to the news factors to obtain news factors with different weights;

According to the news factors of different weights, comprehensively sort the candidate recommended news according to the weights;

Perform a deduplication operation on the candidate recommended news according to the user's historical behavior information to obtain a news recommendation list;

Recommend news to the user according to the news recommendation list.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

Obtain the news click ratio and corresponding browsing duration of the user from the recommended news portal;

The recommendation quality is monitored according to the news click ratio and the corresponding browsing time.

Further, when the news recommendation program in the memory 130 is executed by the processor, the following steps are also implemented:

Obtain the user's operation behavior information on the recommended news list;

setting a punishment mechanism according to the operation behavior information;

According to the penalty mechanism, the weight of the news factor of the candidate recommended news of the user is adjusted.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Based on the above-mentioned terminal device architecture but not limited to the above-mentioned architecture, the method embodiments of the present application are proposed.

Referring to FIG. 2 , FIG. 2 is a schematic flowchart of a first exemplary embodiment of a news recommendation method of the present application. The news recommendation method includes:

Step S101, acquiring the user's own attribute information and historical behavior information.

Specifically, obtain the user's own attribute information and historical behavior information, wherein the self-attribute information is the information associated with the user's recommended news, including but not limited to the user's regional information and department information; the historical behavior information is related to the user's news operation Behavior-related information, including but not limited to users' browsing news information, bookmarking news information, following company information, and searching for news information. When a user registers and logs in for the first time, a user information table is generated for the user. The user information table can be used to store the user's own attribute information and historical behavior information, and can be used to obtain the user's own attribute information and historical behavior information. .

Step S102: Acquire the historical behavior quantity of the user according to the historical behavior information.

Specifically, the user's historical behavior quantity is obtained according to the user's historical behavior information, wherein the historical behavior quantity is the number of news operation behaviors of the user in the past N days, including but not limited to the user's browsing news quantity, favorite news quantity, and following company quantity and the number of searches for news. The number of historical behaviors is stored in the user's user information table, and can be obtained according to the user's historical behavior information statistics. More specifically, compared with the traditional supervised news recommendation method that recommends news based on user browsing data, this embodiment uses the number of historical behaviors of users to divide multiple ranges of historical behaviors for users, and for different historical behaviors. Users with a range of behaviors can recommend news based on the shared characteristics of their users, which not only eliminates the need for manpower to label data, solves the problem of mobile device users' news recommendation without training data, but also satisfies the news recommendation needs of different users.

Step S103: Classify users according to the number of historical behaviors to obtain a user classification result.

Specifically, users are classified according to their historical behavior quantities, and a user classification result is obtained, wherein users whose historical behavior quantity values fall within the same quantity range are divided into one category. According to different needs of the actual situation, users can be divided into multiple categories by setting different preset thresholds for the number of historical behaviors of users. For example, users can be divided into two categories, three categories, and four categories according to different preset thresholds. class, five classes or more. After the users are classified, they have the same number of classification features, and the classification results of users with the same number of features are obtained.

Step S104, according to the user classification result, for different user classifications, and based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is used to recommend news to the user.

Specifically, according to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is used to recommend news to users. Quantitative classification features, and based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. In other words, users with different numbers of historical behaviors are classified according to their own attribute information, such as user geographic information and/or department information, or according to their users' historical behavior information, such as users' browsing news information, favorite news information , pay attention to company information and search for news information, and use the corresponding unsupervised clustering method to recommend news for users who belong to the same user category.

The execution body of the method in this embodiment may be a news recommendation device, or a news recommendation terminal device or server. In this embodiment, a news recommendation device is used as an example, and the news recommendation device may be integrated in an intelligent device with data processing function. on terminal devices such as mobile phones and tablet computers.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Further, referring to FIG. 3 , FIG. 3 is a schematic flowchart of a second exemplary embodiment of the news recommendation method of the present application. Based on the above embodiment shown in FIG. 2, in this embodiment, step S103, classifying users according to the number of historical behaviors, and obtaining a user classification result includes:

Step S1031, comparing the number of historical behaviors with a preset threshold;

Step S1032, classifying users whose number of historical behaviors is greater than or equal to the preset threshold as Class A users;

Step S1033, classifying the users whose number of historical behaviors is greater than zero and less than the preset threshold as B-type users;

In step S1034, the users whose historical behavior quantity is equal to zero are divided into C-type users.

Compared with the embodiment shown in FIG. 2 above, this embodiment provides a method for classifying users according to the number of historical user behaviors.

Specifically, after obtaining the number of historical behaviors of the user in the past N days, it is compared with a preset threshold. Based on the actual requirement of dividing users into new users or old users, this embodiment sets two preset thresholds according to the number of historical behaviors of users, and divides users into three categories. The first preset threshold is zero, and the second preset threshold is any value greater than zero. When the number of historical behaviors of the user is greater than or equal to the preset threshold, the user is classified as a class A user; when the number of historical behaviors of the user is greater than zero and less than the preset threshold, the user is classified as a class B user ; When the number of historical behaviors of the user is equal to zero, that is, the user has no news operation behavior before, the user is classified as a C-type user.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Further, based on the above-mentioned embodiment shown in FIG. 2 or FIG. 3 , a third exemplary embodiment of the present invention is proposed. In this embodiment, in the above step S104, according to the user classification result, for different user classifications, and based on the user's own attribute information or historical behavior information, recommending news to the user by using the corresponding unsupervised clustering method may include:

Step S1041, if the user classification result is a C-type user, according to the user's own attribute information, and adopting a recommendation strategy based on user information, recommend corresponding news to the user.

Specifically, according to the comparison result between the user's historical behavior quantity and the preset threshold, the users are divided into A-type users, B-type users and C-type users. If the classification result of the user is a C-type user, that is, the C-type user has no news operation behavior before, then according to the C-type user's own attribute information, including but not limited to the user's geographical information or the department information, the information based on the user's information will be used. recommends the corresponding news to users. Among them, the recommendation strategy based on user information is a strategy that can recommend news to users according to the user's own attribute information and adopt the corresponding unsupervised clustering method.

Step S1042, if the user classification result is a B-type user, according to the user's own attribute information or historical behavior information, and adopt a recommendation strategy based on user information, a recommendation strategy based on a user's attention to a company, and a recommendation based on the user's browsing news strategy to recommend corresponding news to users.

Specifically, if the classification result of the user is a B-type user, that is, the B-type user has the number of historical behaviors less than the preset threshold, then according to the B-type user's own attribute information or historical behavior information, and use the user information-based Recommendation strategy, recommendation strategy based on the user's attention to the company, and recommendation strategy based on the user's browsing news, recommend the corresponding news to the user. Among them, the self attribute information includes but is not limited to the user's geographical information and the department information; historical behavior information includes but is not limited to the user's browsing news information and following company information. According to the above information of the user, a corresponding recommendation strategy is adopted to recommend corresponding news to the user. Among them, the recommendation strategy based on user information is a strategy that can recommend news to users according to the user's own attribute information and adopts the corresponding unsupervised clustering method; the recommendation strategy based on the user's attention to the company is that the user can follow the company information according to the user's own information, and use the corresponding unsupervised clustering method. The supervised clustering method is a strategy for recommending news to users; the recommendation strategy based on user browsing news is a strategy that can recommend news to users by using the corresponding unsupervised clustering method according to the news information browsed by the user.

Step S1043, if the user classification result is a category A user, according to the user's historical behavior information, using a preset recommendation algorithm and a recommendation strategy based on the user's browsing news, recommend corresponding news to the user.

Specifically, if the classification result of the user is a Class A user, that is, the Class A user has a number of historical behaviors greater than or equal to the preset threshold, then according to the historical behavior information of the Class A user, including but not limited to the user's browsing news Information, collecting news information, paying attention to company information and searching for news information, using preset recommendation algorithms and recommendation strategies based on users' browsing news, recommend corresponding news to users. Among them, the recommendation strategy based on the user's browsing news is a strategy that can construct a news recommendation system by using a preset unsupervised clustering recommendation algorithm according to the user's historical behavior information, and recommend corresponding news to the user based on the news recommendation system.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Further, based on the above step S1041 or step S1042, a fourth exemplary embodiment of the present invention is proposed. The user's own attribute information includes user region information and department information, and the recommendation strategy based on the user information includes a recommendation strategy for recommending corresponding popular news according to the user region information and/or department information.

Specifically, the user region information and/or the department information of the user is acquired, and the user's popular news within a corresponding range is calculated according to the user region information and/or the department information of the user. Among them, the calculation formula of popular news includes but is not limited to the following formula:

Among them: "Total Views" is the summary of the times the news has been viewed by all users since the news was released; "Operations" is the summary of the times the news has been manipulated by all users, including but not limited to clicking, forwarding, Favorites; "News Release Days" is the number of days since the news was released.

Top news is defined as the sum of the total number of views and operations for the news since the news was published, and then divided by the number of days since the news was published.

By calculating the user's popular news within the range corresponding to the user's user geographic information and/or the department information of the user, the corresponding popular news is acquired, and news recommendation is made to the user.

Through the above solution, this embodiment recommends corresponding hot news to users in different regions and departments by utilizing the user's regional characteristics and/or the characteristics of the department, so that the hot news can be recommended to the user in time.

Further, based on the above step S1042, a fifth exemplary embodiment of the present invention is proposed. The historical behavior information includes the user's following company information, and the recommendation strategy based on the user's following company includes: recommending a recommendation strategy for the current day's news of the corresponding company according to the following company information.

Specifically, the user's following company information is obtained, and according to the user's following company information, the daily news of the following company is pulled and recommended to the user through the Smart Eye interface. When the user has multiple following companies, obtain the following company information of the user on the multiple following companies, and calculate the user's attention to each company. The calculation formula of the company's attention includes but is not limited to the following formula:

func(follow, search times, operation time) (2)

Among them, "Follow" is the user's following behavior about the company, that is, the user clicks the company's follow button box to complete the following behavior; "Search times" is the number of times the user has searched the company and the news content released by the company; "Operation time" is the duration of the user's operation on the company and the news content published by the company, wherein the operation includes but is not limited to browsing.

First, assign a certain scoring rule to the user's operation behavior of the company concerned, in which the user can obtain different behavior assignment points through the three behaviors of following, search times, and operation time. Among them, the behaviors of different operation behaviors are assigned points. It is not static and can be adjusted according to the actual situation. Then, the user's attention to the concerned company is obtained by calculating the sum of the scores assigned by the user to all the operation behaviors of the concerned company.

By calculating the user's degree of attention to each company, and taking the user's degree of attention to the company as his degree of attention to the company's news, according to the user's degree of attention to the company's news, the corresponding weight is given to the news, according to Weights recommend to users news about companies to follow.

This embodiment obtains the user's following company information through the above solution, calculates and obtains the user's degree of attention to different companies of interest, and recommends the news of the following companies to the user according to the degree of attention, so that the recommended news is more in line with the user's interest preference.

Further, based on the above step S1042 or step S1043, a sixth exemplary embodiment of the present invention is proposed. The historical behavior information includes the user's browsing news information, collecting news information, and searching for news information, and the recommendation strategy based on the user's browsing news includes:

If the user classification result is a B-type user, recommending a recommendation strategy for other news strongly associated with the browsing news information according to the browsing news information;

Specifically, if the classification result of the user is a B-type user, that is, the B-type user has a number of historical behaviors less than a preset threshold, then according to the user's browsing news information, an association rule algorithm is used to obtain the user's browsing news information. Other news that is strongly related, and recommend news to the user.

If the user classification result is a category A user, a recommendation algorithm is used to perform a recommendation strategy for news recommendation according to the browsing news information, saving news information and searching for news information.

Specifically, if the classification result of the user is a Class A user, that is, the Class A user has a number of historical behaviors greater than or equal to the preset threshold, then according to the historical behavior information of the Class A user, including but not limited to the user's browsing news information, collect news information, pay attention to company information and search for news information, and use collaborative filtering recommendation algorithm, content-based recommendation algorithm and deep learning-based recommendation algorithm to build a multi-model recommendation system. news strategy.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

Further, referring to FIG. 4 , FIG. 4 is a schematic flowchart of a seventh exemplary embodiment of a news recommendation method of the present application. Based on the embodiment shown in FIG. 2, in this embodiment, in the above step S104, according to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is adopted Recommending news to users can also include:

Step S1044, obtain candidate recommendation news.

Specifically, according to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, a corresponding unsupervised clustering method is used to obtain candidate recommended news. So far, before recommending news to the user, further processing is performed on the obtained candidate recommended news.

Step S1045: Acquire corresponding news factors according to the candidate recommended news.

Specifically, the corresponding news factors are obtained according to the candidate recommended news, wherein the news factors include but are not limited to news popularity, news sources, news release days, and user preference information. Among them, news popularity can be obtained by calculation; news sources include but are not limited to popular news related to the user's region, popular news related to the user's department, and news of the day the user is following the company; the number of days since the news was released; the number of days since the news was released; The preference information is to record the mapping of the user to the keyword table by maintaining the data set, wherein the keyword table is composed of all keywords, and the keywords include but are not limited to the companies concerned by the user, the field to which the user browses the news, and each keyword have corresponding weights.

Step S1046, setting corresponding weights according to the news factors, to obtain news factors with different weights.

Specifically, corresponding weights are set according to the news factors to obtain news factors with different weights. A corresponding weight is set for each news factor, including but not limited to news popularity weight, news source weight, news release days weight, and user preference information weight.

Step S1047, according to the news factors of different weights, comprehensively sort the candidate recommended news according to the weights.

Specifically, according to the news factors of different weights, comprehensively sort the candidate recommended news according to the weights. The weight of each candidate recommendation news is integrated, and the candidate recommendation news is sorted in turn according to the weight.

Step S1048: Perform a deduplication operation on the candidate recommended news according to the user's historical behavior information to obtain a news recommendation list.

Specifically, according to the historical behavior information of the user, a deduplication operation is performed on the candidate recommended news to obtain a news recommendation list. Obtain the user's historical behavior information, such as the news information that the user has browsed, compare it with the candidate recommended news that has been comprehensively sorted, filter the news that the user has browsed in the candidate recommended news, and get the candidate recommended news that has not been browsed. , and generate the final news recommendation list accordingly.

Step S1049, recommend news to the user according to the news recommendation list.

Specifically, recommend news to the user according to the news recommendation list. Push the weighted, sorted and deduplicated news recommendation list to the user's recommended news entry to complete the news recommendation to the user.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

In addition, compared with the prior art, in the process of news recommendation in this embodiment, a weighting mechanism is designed to add weights to each news factor of the candidate recommended news, comprehensively sort the candidate recommended news according to the weight, and filter the user's existing news through the deduplication operation. The browsed candidate recommended news realizes the purpose that breaking news can be recommended to users in a timely manner.

Further, referring to FIG. 5 , FIG. 5 is a schematic flowchart of an eighth exemplary embodiment of a news recommendation method of the present application. Based on the embodiment shown in FIG. 4, in this embodiment, the above step S1049, after the step of recommending news to the user according to the news recommendation list, further includes:

Step S1050, obtaining the news click ratio and corresponding browsing duration of the user from the recommended news portal;

Step S1051: Perform recommendation quality monitoring according to the news click ratio and the corresponding browsing duration.

Specifically, a recommendation quality monitoring mechanism is designed. After the news recommendation list is pushed to the user's recommended news entry, the user's operation behavior information from the recommended news entry is obtained, wherein the operation behavior information includes the news click ratio and the corresponding browsing time. According to the news click ratio and corresponding browsing time of the news recommendation list, the quality of the news recommendation and the user's interest transfer are evaluated.

Compared with the prior art, the present embodiment monitors the news click ratio and the corresponding browsing time of the recommended news list based on the recommendation quality monitoring mechanism through the above solution, and adds a time weight to the score of each recommended news, thereby improving the recommendation. The recommendation quality of the system improves the problem of user interest migration.

Further, referring to FIG. 6 , FIG. 6 is a schematic flowchart of a ninth exemplary embodiment of a news recommendation method of the present application. Based on the embodiment shown in FIG. 4, in this embodiment, the above step S1049, after the step of recommending news to the user according to the news recommendation list, further includes:

Step S1060: Obtain the operation behavior information of the user on the recommended news list.

Specifically, after the news recommendation list is pushed to the user's recommended news entry, if the user has an operation behavior, obtain the user's operation behavior information from the recommended news entry, where the operation behavior information includes but is not limited to browsing news information, Collect news information, follow company information and search news information.

Step S1061, setting a punishment mechanism according to the operation behavior information.

Specifically, a punishment mechanism is set according to the operation behavior information. The penalty mechanism is used to adjust the weights corresponding to news factors in the user's current candidate recommended news after detecting that the user operates on the recommended news list for users of different categories.

Step S1062, according to the punishment mechanism, adjust the weight of the news factor of the candidate recommended news of the user.

Specifically, according to the punishment mechanism, for B-type users and C-type users, after obtaining the operation behavior information of this type of users on the recommended news portal, adjust the weight of news sources in the candidate recommended news for this type of user; for A-type users , after obtaining the operation behavior information of this type of user on the recommended news entry, adjust the weight of the user's preference information in the candidate recommendation news of this type of user, wherein the weight of the user's preference information is adjusted by adjusting the weight of each keyword in the keyword table corresponding to the user. weight to achieve.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

In addition, compared with the prior art, this embodiment uses the above solution to provide feedback to the news recommendation algorithm according to the user's operation behavior information on the recommended news list, adjust the weights of relevant news factors in the recommended news, and improve the problem of user interest migration and recommendation. inaccurate question.

Further, referring to FIG. 7, FIG. 7 is a schematic diagram of the overall flow of the tenth exemplary embodiment of the news recommendation method of the present application. The overall flow of this embodiment includes:

Based on the user, obtain the user's own attribute information and historical behavior information. Obtain the user's historical behavior quantity according to the user's historical behavior information. The user is classified according to the number of historical behaviors of the user, wherein the number of historical behaviors of the user is compared with a preset threshold, and the users whose number of historical behaviors is greater than or equal to the preset threshold are classified as Class A users; Users whose behavior quantity is greater than zero and less than the preset threshold are classified as B-type users; users whose historical behavior quantity is equal to zero are classified as C-type users, and the user classification result of the user is obtained. According to the user classification result of the user, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to the user, wherein, if the user classification result is a class C user, then According to the user's own attribute information, and adopt the recommendation strategy based on user information, recommend the corresponding news to the user; if the user classification result is a B-type user, then according to the user's own attribute information or historical behavior information, and adopt The recommendation strategy based on user information, the recommendation strategy based on the user's attention to the company, and the recommendation strategy based on the user's browsing news, recommend the corresponding news to the user; if the user classification result is a class A user, according to the user's historical behavior information, adopt The preset recommendation algorithm recommends the corresponding news to the user based on the recommendation strategy of the user browsing news. More specifically, the recommendation strategy based on user information includes: recommending a corresponding popular news recommendation strategy according to the user's regional information and/or department information; the recommendation strategy based on the user's following companies includes: according to the following company information , recommending the recommendation strategy of the news of the day corresponding to the company; the recommendation strategy based on the user's browsing news includes: if the user's classification result is a B-type user, then according to the browsing news information, recommend other news that is strongly related to the browsing news information. Recommendation strategy for news; if the user classification result is a category A user, a recommendation algorithm is used to perform a recommendation strategy for news recommendation according to the browsing news information, collecting news information and searching for news information. Further, according to the user classification result, for different user classifications, and based on the user's own attribute information or historical behavior information, the step of using the corresponding unsupervised clustering method to recommend news to the user further includes: obtaining candidate recommended news; Recommend news to obtain the corresponding news factors; set the corresponding weights according to the news factors to obtain news factors of different weights; according to the news factors of different weights, comprehensively sort the candidate recommended news according to the weights; According to the historical behavior information of the user, Perform a deduplication operation on the candidate recommended news to obtain a news recommendation list; recommend news to the user according to the news recommendation list. Further, after the user logs in to the recommended news portal, after the step of recommending news to the user with the news recommendation list, the method further includes: obtaining the news click ratio and corresponding browsing duration of the user from the recommended news portal; Quality monitoring is recommended. At the same time, after the step of recommending news to the user by the news recommendation list, the method further includes: acquiring the user's operation behavior information on the recommended news list; setting a punishment mechanism according to the operation behavior information; The weight of news factors is adjusted, and then the comprehensive ranking of recommended news is adjusted.

This embodiment adopts the above solution, specifically by acquiring the user's own attribute information and historical behavior information; acquiring the user's historical behavior quantity according to the historical behavior information; classifying the users according to the historical behavior quantity, and obtaining the user classification result; According to the user classification results, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is used to recommend news to users. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

In addition, an embodiment of the present application also proposes a news recommendation device, and the news recommendation device includes:

The information acquisition module is used to acquire the user's own attribute information and historical behavior information;

A quantity acquisition module, used for acquiring the historical behavior quantity of the user according to the historical behavior information;

A user classification module, configured to classify users according to the number of historical behaviors to obtain user classification results;

The news recommendation module is configured to, according to the user classification result, for different user classifications, and based on the user's own attribute information or historical behavior information, adopt a corresponding unsupervised clustering method to recommend news to the user.

For the principle and implementation process of implementing news recommendation in this embodiment, please refer to the above-mentioned embodiments, which will not be repeated here.

In addition, an embodiment of the present application also proposes a terminal device, the terminal device includes a memory, a processor, and a news recommendation program stored on the memory and running on the processor, the news recommendation program being The steps of implementing the above-mentioned news recommendation method when the processor is executed.

When the news recommendation program is executed by the processor, all the technical solutions of the foregoing embodiments are adopted, so at least all the beneficial effects brought by all the technical solutions of the foregoing embodiments are provided, which will not be repeated here.

In addition, an embodiment of the present application also proposes a computer-readable storage medium, where a news recommendation program is stored on the computer-readable storage medium, and when the news recommendation program is executed by a processor, the steps of the above-mentioned news recommendation method are implemented .

When the news recommendation program is executed by the processor, all the technical solutions of the foregoing embodiments are adopted, so at least all the beneficial effects brought by all the technical solutions of the foregoing embodiments are provided, which will not be repeated here.

Compared with the prior art, the news recommendation method, device, terminal device and storage medium proposed in the embodiments of the present application obtain the user's own attribute information and historical behavior information; obtain the user's historical behavior quantity according to the historical behavior information; According to the user classification result, for different user classifications, based on the user's own attribute information or historical behavior information, the corresponding unsupervised clustering method is adopted to recommend news to the user. By using the corresponding unsupervised clustering method for news recommendation after user classification, it can solve the high-cost investment problems such as accumulating a large amount of user browsing data, investing manpower to label the data, and training a supervised learning model for mobile device user news recommendation. Based on the solution of the present application, by classifying users and applying corresponding unsupervised clustering methods for different categories of users without accumulating a large amount of user browsing data and without investing manpower to label the data, the recommended news is provided. Finally, the solution is solved by the solution of the present application. It solves the problem of high cost of news recommendation for mobile device users.

It should be noted that, herein, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or system comprising a series of elements includes not only those elements, It also includes other elements not expressly listed or inherent to such a process, method, article or system. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article or system that includes the element.

The above-mentioned serial numbers of the embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

From the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is better implementation. Based on such understanding, the technical solutions of the present application can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products are stored in the above storage medium (such as ROM/RAM, magnetic CD, CD), including several instructions to make a terminal device (which may be a mobile phone, a computer, a server, a controlled terminal, or a network device, etc.) to execute the method of each embodiment of the present application.

The above are only the preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present application, or directly or indirectly applied in other related technical fields , are similarly included within the scope of patent protection of this application.

Claims (14)

1. A news recommendation method is characterized by comprising the following steps:

acquiring self attribute information and historical behavior information of a user;

acquiring the historical behavior quantity of the user according to the historical behavior information;

classifying the users according to the historical behavior quantity to obtain a user classification result;

and according to the user classification result, aiming at different user classifications, recommending news to the user by adopting a corresponding unsupervised clustering method based on the self attribute information or historical behavior information of the user.

2. A news recommendation method according to claim 1, wherein said step of classifying the users according to the number of the historical behaviors to obtain the user classification result comprises:

comparing the historical behavior quantity with a preset threshold value;

dividing the users with the historical behavior number larger than or equal to the preset threshold value into A-type users;

dividing the users with the historical behavior quantity larger than zero and smaller than the preset threshold value into B-type users;

and dividing the users with the historical behavior quantity equal to zero into C-type users.

3. The news recommendation method according to claim 2, wherein the step of recommending news to the user by using a corresponding unsupervised clustering method based on user attribute information or historical behavior information for different user classifications according to the user classification results comprises:

and if the user classification result is a C-type user, recommending corresponding news to the user according to the attribute information of the user and by adopting a recommendation strategy based on the user information.

4. The news recommendation method according to claim 2, wherein the step of recommending news to the user by using a corresponding unsupervised clustering method based on user attribute information or historical behavior information for different user classifications according to the user classification results comprises:

and if the user classification result is a B-type user, recommending corresponding news to the user according to the attribute information or the historical behavior information of the user and by adopting a recommendation strategy based on user information, a recommendation strategy based on a user attention company and a recommendation strategy based on news browsing of the user.

5. The news recommendation method according to claim 2, wherein the step of recommending news to the user by adopting a corresponding unsupervised clustering method based on user attribute information or historical behavior information for different user classifications according to the user classification results comprises:

and if the user classification result is the class A user, recommending the corresponding news to the user by adopting a preset recommendation algorithm and based on a recommendation strategy for browsing the news by the user according to the historical behavior information of the user.

6. The news recommendation method according to claim 3 or 4, wherein the user's own attribute information includes user region information and department information, and the recommendation policy based on the user information includes: and recommending a corresponding hot news recommending strategy according to the user region information and/or the department information.

7. A news recommendation method according to claim 4, wherein the historical behavior information includes information about companies interested by the user, and the recommendation policy based on companies interested by the user includes: and recommending a recommendation strategy of the news of the corresponding company on the same day according to the concerned company information.

8. The news recommendation method according to claim 4 or 5, wherein the historical behavior information includes browsing news information, favorite news information, and search news information of the user, and the recommendation strategy based on browsing news of the user includes:

if the user classification result is a B-type user, recommending a recommendation strategy of other news strongly associated with the browsing news information according to the browsing news information;

and if the user classification result is the A-type user, adopting a recommendation algorithm to carry out a recommendation strategy of news recommendation according to the browsing news information, the collected news information and the searched news information.

9. The news recommendation method according to any one of claims 3, 4 or 5, wherein the step of recommending news to users by using corresponding unsupervised clustering methods based on user attribute information or historical behavior information according to the user classification results and for different user classifications further comprises:

acquiring candidate recommended news;

acquiring corresponding news factors according to the candidate recommended news;

setting corresponding weights according to the news factors to obtain news factors with different weights;

according to the news factors with different weights, comprehensively sequencing the candidate recommended news according to the weights;

according to the historical behavior information of the user, performing duplicate removal operation on the candidate recommended news to obtain a news recommendation list;

and recommending news to the user according to the news recommending list.

10. The news recommendation method of claim 9, wherein after the step of recommending news to a user according to the news recommendation list, further comprising:

acquiring a news click ratio and corresponding browsing duration of the user from the recommended news entry;

and monitoring the recommendation quality according to the news click ratio and the corresponding browsing duration.

11. The news recommendation method of claim 9, wherein after the step of recommending news to a user according to the news recommendation list, further comprising:

acquiring operation behavior information of the user on the recommended news list;

setting a punishment mechanism according to the operation behavior information;

and according to the punishment mechanism, carrying out weight adjustment on news factors of the candidate recommended news of the user.

12. A news recommender, the news recommender comprising:

the information acquisition module is used for acquiring the attribute information and the historical behavior information of the user;

the quantity obtaining module is used for obtaining the historical behavior quantity of the user according to the historical behavior information;

the user classification module is used for classifying the users according to the historical behavior quantity to obtain a user classification result;

and the news recommending module is used for recommending news to the user by adopting a corresponding unsupervised clustering method based on the attribute information or the historical behavior information of the user according to the user classification result and aiming at different user classifications.

13. A terminal device, characterized in that the terminal device comprises a memory, a processor and a news recommender stored on the memory and operable on the processor, which when executed by the processor implements the steps of the news recommendation method as claimed in any one of claims 1 to 11.

14. A computer-readable storage medium, having stored thereon a news recommender, which when executed by a processor implements the steps of the news recommendation method as claimed in any one of claims 1-11.

Images