CN104281882B - The method and system of prediction social network information stream row degree based on user characteristics - Google Patents

Abstract

The invention provides a method for predicting the popularity of social network information based on user characteristics. The method includes: obtaining user data and information data in the social network; extracting part of user attribute characteristics and user behavior characteristics from user data; Classify the user data according to the user behavior characteristics; obtain the user communication characteristics corresponding to the information data according to the information data and user categories; obtain the social network information popularity prediction model according to the user communication characteristics, and use the prediction model to predict the information popularity. The system for predicting the popularity of social network information based on user characteristics provided by the present invention includes an acquisition module, a feature extraction module, a classification module, a processing module, a prediction model module and a prediction model. The invention combines the characteristics of user behavior characteristics to more accurately predict the information dissemination of the social network, and solves the problems of lagging hotspot discovery, information push, and real-time network public opinion monitoring.

Description

Method and system for predicting social network information popularity based on user characteristics

technical field

The invention relates to the technical field of network security, in particular to a method and system for predicting the popularity of social network information based on user characteristics.

Background technique

At present, the Internet has become an important way to obtain information, especially with the rapid rise of various social networking sites, more convenient information acquisition and faster information dissemination follow. Social networks have formed a huge online social group and built close online interpersonal relationships. Information dissemination on social networks is different from traditional information dissemination such as letters, oral, newspapers, etc. Information dissemination in social networks has the following outstanding features. First, it has strong real-time performance. It is easy for readers to spread the major events they see in the fastest time; second, with a strong group nature, information release on social networks becomes arbitrary, and different people will publish some inflammatory messages for some purposes. information, and the widespread dissemination of this information will lead to group behavior. Third, the periodicity of information updates becomes smaller. Due to the massive release of information, the sources of information become wider and wider, so the information is gradually replaced by new information in the process of dissemination, and the periodicity becomes smaller.

The popularity prediction of information dissemination combined with the characteristics of information dissemination on social networks will effectively solve many problems. Being able to detect changes in information dissemination early and predict the popularity of information dissemination as early as possible has become the main part of real-time information push and social network public opinion monitoring. At present, the monitoring method is adopted for information push and public opinion dissemination. A threshold is set. When some parameters of the information are greater than this threshold, it will be defined as push information or public opinion information. These methods are relatively rough, and it is difficult to guarantee the real-time characteristics of the information.

Contents of the invention

Aiming at the defects of the prior art, the method for predicting the popularity of social network information based on user characteristics provided by the present invention, combined with the characteristics of user behavior characteristics, can more accurately predict the dissemination of information on social networks, and solve hot spot discovery lag, information push and network It is difficult to guarantee the real-time performance of public opinion monitoring.

In a first aspect, the present invention provides a method for predicting the popularity of social network information based on user characteristics, the method comprising:

Acquiring information data in the social network within a preset time and user data corresponding to the information data, the user data including a plurality of user attribute characteristics;

Extracting some user attribute features from the user data, and acquiring user behavior features of the user data according to the user data;

Classify the user data according to the user attribute characteristics and the user behavior characteristics to obtain the user category in the user data;

Acquiring user propagation characteristics corresponding to the information data according to the information data and the category of users in the user data;

Determine a predictive model of social network information popularity according to the user propagation characteristics;

The prediction model is used to analyze the information data generated within a period of time, and predict the popularity of the information.

Preferably, after the step of obtaining the information data in the social network within the preset time and the user data corresponding to the information data, the method further includes:

The user data and the information data are stored in a database.

Preferably, the acquisition of information data in the social network within a preset time period and user data corresponding to the information data includes:

Use web crawlers to obtain user data and information data of forum-like social networks;

Use application programming interface API to obtain user data and information data of Weibo social network;

A web crawler is used to obtain user data of a community social network, and a user's clipboard is used to obtain information data of a community social network.

Preferably, the user data is classified according to the user attribute characteristics and the user behavior characteristics to obtain the user category in the user data, including:

Performing normalization processing on the user attribute features and the user behavior features to obtain user features;

According to the user characteristics, a clustering algorithm is used to classify the user data to obtain user categories in the user data.

Preferably, the use of a clustering algorithm to classify user data includes:

Divide the user data into two categories, and calculate the distance between the category centers, if the distance between the category centers is less than the preset value, then merge the two categories into one category;

Continue to classify the user data of each category, and calculate the distance between the centers of each category, and stop the classification until the user data of three categories are merged into one category, and the category of the user is obtained.

Preferably, the determining the prediction model of social network information popularity according to the user communication characteristics includes:

Establish a multivariate linear model based on user characteristics;

The user dissemination information is used as a training set, and the linear model is trained to obtain a social network information popularity prediction model.

In a second aspect, the present invention provides a system for predicting the popularity of social network information based on user characteristics, the system comprising:

An acquisition module, configured to acquire information data in a social network within a preset time period and user data corresponding to the information data, the user data including a plurality of user attribute characteristics;

A feature extraction module, configured to extract part of user attribute features from the user data, and obtain user behavior features of the user data according to the user data;

A classification module, configured to classify user data according to the user attribute characteristics and the user behavior characteristics, to obtain the user category in the user data;

A processing module, configured to acquire user propagation characteristics corresponding to the information data according to the information data and the category of users in the user data;

A predictive model module, configured to determine a predictive model of social network information popularity according to the user communication characteristics;

The prediction module is used to analyze the information data generated within a period of time by using the prediction model, and predict the popularity of the information.

Preferably, the system also includes:

A storage module, configured to store the user data and the information data in a database.

Preferably, the classification module includes:

A normalization sub-module, configured to perform normalization processing on the user attribute features and the user behavior features to obtain user features;

The division sub-module is used to classify the user data by using a clustering algorithm according to the user characteristics, and obtain user categories in the user data.

Preferably, the predictive model module includes:

Establish a model sub-module for establishing a multivariate linear model based on user characteristics;

The training sub-module is used to use the user dissemination information as a training set to train the linear model to obtain a social network information popularity prediction model.

Based on the above technical solution, the method for predicting the popularity of social network information based on user characteristics provided by the present invention fully considers the real-timeness of social network information and the influence of user characteristics on information dissemination, and adopts the method of information popularity prediction to perform information The description of the communication mode can predict the information dissemination as early as possible, reducing the lag of traditional methods, and providing help for timely information push and timely control of public opinion on social networks. At the same time, the system of the present invention has low memory cost during operation, high efficiency, independence and portability. In general, the present invention can make an early prediction on the popularity of information dissemination, and is of great help to the timely push of information and the timely control of public opinion on the network.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

1 is a schematic flow diagram of a method for predicting social network information popularity based on user characteristics provided by an embodiment of the present invention;

Fig. 2 is a schematic flowchart of a method for acquiring user data and information data provided by another embodiment of the present invention;

3 is a structural diagram of a system for predicting social network information popularity based on user characteristics provided by an embodiment of the present invention;

Fig. 4 is a schematic structural diagram of a classification module provided by another embodiment of the present invention;

Fig. 5 is a schematic structural diagram of a prediction model module provided by another embodiment of the present invention.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in Figure 1, Figure 1 shows a method for predicting social network information popularity based on user characteristics provided by an embodiment of the present invention, the method includes the following steps:

Step 101: Obtain information data and user data corresponding to the information data in the social network within a preset time. Wherein, the user data includes multiple user attribute features.

In this embodiment, the acquired user data and information data in the social network are stored in the database.

According to different types of social networks, different data acquisition methods are adopted. Forum social networking sites use posts as information carriers, so it is suitable to use web crawlers to obtain post data.

For microblog-type social networks, the short-text microblogs are used to disseminate news, and the application programming interface (Application Programming Interface, API for short) provided by the microblog platform can be used to obtain information data and user data.

For community-based social networks, user data can be obtained through web crawlers, and information data can be obtained through the clipboards of these users.

Step 102: Extract part of the user attribute features from the user data, and obtain the user behavior features of the user data according to the user data.

Specifically, the acquired user data is specifically divided into user attribute features and user behavior features.

The user attribute feature is the information provided by the user when registering a social network account, such as: name, age, gender, etc. For user attribute features, valid features that may affect information dissemination should be retained, and invalid features that will not affect information dissemination should be removed, such as phone numbers, zip codes, etc.

User behavior characteristics refer to the characteristics generated when users perform activities on social networks, such as the number of friends, the number of replies, and so on. However, there are some user behavior data that cannot be obtained directly through APIs and web crawlers, and need to be obtained indirectly through calculations, for example: using social network age, aggregation coefficient, etc. The following user characteristics are composed of effective user attribute characteristics and complete user behavior characteristics.

Step 103: Classify the user data according to the user attribute features and the user behavior features to obtain the user categories in the user data.

In this embodiment, this step includes:

Performing normalization processing on the user attribute features and the user behavior features to obtain user features;

According to the user characteristics, a clustering algorithm is used to classify the user data to obtain user categories in the user data.

Specifically, the CLA algorithm is a clustering algorithm. The CLA algorithm does not need to specify the number of classifications in advance, but finds the appropriate number of classifications through certain conditions. First, the CLA algorithm divides users into two categories and calculates the distance between the category centers. When the distance between the category centers is less than a certain value, the two categories are considered to belong to one category, and the two categories are fused into one category. Then increase the number of categories and reclassify users according to the above method, until the first time users of three categories merge into one category, stop the algorithm. This way users are classified into appropriate categories.

Step 104: Obtain user propagation characteristics corresponding to the information data according to the information data and user categories in the user data.

According to the classification of the above users, the users participating in information dissemination in the information data in the database are counted according to their categories. The number of users of various types is counted as the user dissemination feature of this information.

Step 105: Determine a prediction model of social network information popularity according to the user communication characteristics.

Specifically, this step includes: establishing a multivariate linear model based on user characteristics; using the user dissemination information as a training set to train the linear model to obtain a social network information popularity prediction model.

In this embodiment, the user dissemination characteristics of information are used as a training set, and a model for predicting information popularity can be obtained by using a linear regression method. In the calculation process, the gradient descent algorithm can be used to quickly obtain the influence weight of each category of users on information dissemination.

Step 106: Using the obtained prediction model to analyze the information data generated within a period of time, and predict the popularity of the information.

The method for predicting the popularity of social network information based on user characteristics provided in this embodiment fully considers the real-timeness of social network information and the influence of user characteristics on information dissemination, and uses the information popularity prediction method to describe the information dissemination mode. It can predict information dissemination as early as possible, reducing the lag of traditional methods, and providing help for timely information push and timely control of public opinion on social networks. At the same time, the system of the present invention has low memory cost during operation, high efficiency, independence and portability. In general, the present invention can make an early prediction on the popularity of information dissemination, and is of great help to the timely push of information and the timely control of public opinion on the network.

Next, another embodiment of the present invention takes Tianya Forum as an example to illustrate a method for predicting the popularity of social network information based on user characteristics. The method includes:

Step 1: Obtain information data and user data.

The specific process of this step is as follows:

Since Tianya Forum does not provide an API for effectively obtaining data, in this embodiment, a method of writing a web crawler is used to obtain information data and user data.

Platform environment: Install and configure Microsoft SQL Server2008 database under 32-bit windows7 platform, and use Microsoft Visual Studio 2010 to write the web crawler program PostCrawler. As shown in FIG. 2 , it is a flow chart of the method for obtaining user data and information data provided in this embodiment. For the specific process of running the program, refer to the following web crawler programs postCrawler and UserCrawler.

1) Execute the web crawler program postCrawler by setting a Uniform Resource Locator (Uniform Resource Locator, abbreviated as URL) URL pool on the host. Each post has a corresponding unique post ID. The URL of the post can be obtained through the post ID, and a URL pool can be established through continuous post IDs, so that information data and user data can be obtained continuously. However, because some posts are officially deleted, and some post URLs cannot return the post information normally, regular expressions must be used to filter valid information data before obtaining data.

PostCrawler is defined as follows:

2) The crawler UserCrawler can be set to obtain user data through the reply user ID of the post. Each user has a corresponding user ID, and the URL of the corresponding user page can be found through the user ID. You can access the user page through the URL to view the user's basic information and historical activities and store them in the database.

UserCrawler is defined as follows:

3) SQL Server 2008 database design. The design fields for info data and user data are as follows:

Information data: ID (post ID), hostID (post user ID), click (click volume), reply (reply volume), time (post time), userIDList (reply user ID list)

Reply user ID list: userID (reply user ID), replyTime (reply time)

User data: ID (user ID), fans (number of fans), follows (number of followers), posts (number of posts), replyPosts (number of replies), registerDate (date of registration), lastLoginDate (time of last login), score (community Points), logins (number of logins), topic (number of participating boards), age (age of using Tianya Forum), clusteringCoefficient (aggregation coefficient), reciprocity (reciprocity coefficient), userType (user category).

Here, the functions PostCrawler ( ) and UserCrawler ( ) use any existing programming language on the existing operating system platform to realize the function of the function and obtain information data and user data.

Step 2: extract effective user attribute features from the user data, and calculate user behavior features.

User attributes are basic user information that users are required to fill in when registering an account. Some of this information can be used as user characteristics for user classification. It is not necessary to fill in these information when registering an account on Tianya Forum, but for other social networks, the data can be obtained through API and web crawlers. Invalid user attribute features can be deleted before being stored in the database, or only valid user features can be selected for classification after storing all user attribute features in the database before performing the classification step.

User behavior characteristics refer to the characteristics generated when users conduct activities on social networks. Some user behavior characteristics can be obtained directly through web crawlers and APIs, while others need to be obtained through calculations.

For the users of Tianya Forum, we can obtain the user's age (the age of using Tianya Forum), clusteringCoefficient (aggregation coefficient), and reciprocity (reciprocity coefficient) by calculation.

Age refers to the user behavior characteristics of the length of time a user is active on the Tianya Forum, that is, the time from the user's registration to the last login to the Tianya Forum. The calculation method is:

age=registerDate-lastLoginDate

ClusteringCoefficient is a user behavior characteristic used to measure the interconnectivity between users and their neighbors. If user A follows user B and user C, then clusteringCoefficient is the probability that there is a following relationship between user B and user C. Use C to represent the clusteringCoefficient, G _Δ to represent that user A, user B, and user C all have a following relationship, and G _Λ to represent that only user A, user B, and user C have a following relationship, and the calculation formula can be obtained:

Reciprocity is used to indicate the probability of users following each other, that is, the ratio of the number of users who follow each other with user i to the number of users who follow user i. Use R to represent reciprocity, use A to represent the number of mutual followers with user i, and B to represent the number of all users followed by user i, and the calculation formula can be obtained:

R=A/B

In this way, all user characteristics are obtained, and these user characteristics are stored in the database.

Step 3: Use the CLA algorithm to classify users according to user characteristics. The expression of the CLA algorithm is as follows:

In the CLA algorithm, the K-means algorithm is the most widely used partition-based clustering algorithm, which classifies users by calculating the Euclidean distance of user features. d(C ₁ ,C ₂ ) represents the distance between the cluster center C ₁ of category 1 and the cluster center C ₂ of category 2, namely where F is the number of selected user features. T in the algorithm is a threshold parameter that needs to be set. If the distance between two cluster centers is less than the threshold, the categories of the two cluster centers will be merged together.

Through the CLA algorithm, we can find the appropriate number of categories K and K cluster centers. Through these cluster centers, we can divide users into K categories according to their user characteristics.

Step 4: For Tianya Forum, the users who participate in information dissemination are the users who reply to the posts. Through the reply user ID list of the posts in the database, the types of all reply users can be obtained:

usertype:＝argmin _j ｜｜user _i -C _j ｜｜ ² (1)

Among them, user _i refers to the user, C _j represents the cluster center of user category j, and the cluster center with the shortest distance between the user characteristics and the center point corresponds to the user's userType.

Count the number of posting users in each category, and use these values as the user propagation characteristics of posts.

Step 5: First define the model to predict the reply volume of all posts in t _t (t _t >t _r ) days through the historical records of user reply volume in t _r days of a post v The new model takes into account not only the volume of replies within t _r days, but also which users replied to those replies. Through the previous step, all users who replied to the post can be clustered by calculating the user's tags. We build a multivariate linear model based on user characteristics to show that there is a strong linear relationship between early and late replies and the impact of user behavior on post replies. We can define x _i (v, t _r ) as the sum of the number of replies of users whose user category label userType is i within t _r days, so that we can get the feature vector X _k (v, t _r ), which is expressed as X _k (v,t _r )=(x ₁ (v,t _r ),x ₂ (v,t _r ),...,x _k (v,t _r )). In this way, we get the reply volume of all posts in _t days predictive model:

where parameters Is the parameter vector of this model, where γ _i is the influence parameter of the number of replies of users of category i on the prediction. This parameter vector depends on t _r , t _t and k.

In order to calculate the parameter vector of the model, we introduce mean Relative Squared Error (mRSE), which is an important indicator for evaluating the prediction model. When the minimum value of mRSE is smaller, the performance of the model is better, and mRSE is expressed as:

Substituting our formula (2) into formula (3), we can get:

By giving the training sample set C and setting t _r and t _t , the gradient descent method can be used to train the parameter vector when mRSE takes the minimum value value of . The problem thus becomes:

This can be done by adding The prediction model described by formula (2) is used to predict the future reply volume of the post.

As shown in Figure 3, it is a structural diagram of a system for predicting social network information popularity based on user characteristics provided by an embodiment of the present invention, the system includes: an acquisition module 301, a feature extraction module 302, a classification module 303, and a processing module 304 , a predictive model module 305 and a predictive model 306 .

The obtaining module 301 is configured to obtain information data in a social network within a preset time period and user data corresponding to the information data, where the user data includes a plurality of user attribute features.

The feature extraction module 302 is configured to extract part of user attribute features from the user data, and obtain user behavior features of the user data according to the user data.

The classification module 303 is configured to classify the user data according to the user attribute characteristics and the user behavior characteristics to obtain the user category in the user data.

The processing module 304 is configured to acquire user propagation characteristics corresponding to the information data according to the information data and user categories in the user data.

The prediction model module 305 is configured to determine a prediction model of social network information popularity according to the user communication characteristics.

The prediction module 306 is configured to use the prediction model to analyze the information data generated within a period of time, and predict the popularity of the information.

Further, the system further includes: a storage module, configured to store the user data and the information data in a database.

Specifically, as shown in FIG. 4 , the classification module 303 includes a normalization submodule 401 and a molecularization module 402 .

A normalization submodule 401, configured to perform normalization processing on the user attribute features and the user behavior features to obtain user features;

The division sub-module 402 is configured to classify the user data by using a clustering algorithm according to the user characteristics, and obtain user categories in the user data.

Specifically, as shown in FIG. 5 , the predictive model module 305 includes a model building submodule 501 and a training submodule 502 .

Establishing a model submodule 501, used to establish a multivariate linear model based on user characteristics;

The training sub-module 502 is configured to use the user dissemination information as a training set to train the linear model to obtain a social network information popularity prediction model.

The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand; Modifications are made to the recorded technical solutions, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. a kind of method of the prediction social network information stream row degree based on user characteristics, it is characterised in that this method includes：

Obtain information data and user data corresponding with described information data in social networks, the user in preset time Data include multiple user property features；

The effective user property feature of influence Information Communication is extracted from the user data, and according to the number of users According to the user behavior feature of the acquisition user data；

User data is classified according to the user property feature and the user behavior feature of extraction, obtained in user data The classification of user；

According to the classification of user in described information data and the user data, user corresponding with described information data is obtained Propagation characteristic；

According to user's propagation characteristic, the forecast model of social network information stream row degree is determined；

The information data produced in a period of time is analyzed using the forecast model, information popularity is predicted；

The classification according to user in described information data and the user data, is obtained corresponding with described information data User's propagation characteristic, including：

According to the classification of user, the user that Information Communication is participated in the information data in database is carried out according to the classification of user Statistics, the number of users of each classification of statistics is used as the corresponding user's propagation characteristic of described information data；

It is described that the forecast model of social network information stream row degree is determined according to user's propagation characteristic, including：

Set up the multivariate linear model based on user characteristics；

Using user's propagation characteristic as training set, the linear model is trained, social network information stream row is obtained Spend forecast model；

Wherein, the user characteristics is normalized by the user property feature and the user behavior feature extracted Arrive.

2. according to the method described in claim 1, it is characterised in that the Information Number obtained in preset time in social networks After user data step corresponding with described information data, this method also includes：

By in the user data and described information data Cun Chudao databases.

3. according to the method described in claim 1, it is characterised in that the Information Number obtained in preset time in social networks According to user data corresponding with described information data, including：

The user data and information data of forum's class social networks are obtained using web crawlers；

The user data and information data of microblogging class social networks are obtained using application programming interface API；

The user data of community's class social networks is obtained using web crawlers, community's class social network is obtained using the clipbook of user The information data of network.

4. according to the method described in claim 1, it is characterised in that the user property feature according to extraction and the user Behavioural characteristic is classified user data, obtains the classification of user in user data, including：

User property feature and the user behavior feature to extraction are normalized, and obtain user characteristics；

According to the user characteristics, user data is classified using clustering algorithm, the classification of user in user data is obtained.

5. method according to claim 4, it is characterised in that the use clustering algorithm is classified user data, Including：

User data is divided into two classes, and calculates the distance of class center, if the distance of class center is less than preset value, by this Two classifications are fused into a classification；

User data of all categories is continued to classify, and calculates the distance at center of all categories, until there is the user of three classifications Data fusion obtains the classification of user into classification is stopped during a classification.

6. a kind of system of the prediction social network information stream row degree based on user characteristics, it is characterised in that the system includes：

Acquisition module, for obtaining information data and user corresponding with described information data in preset time in social networks Data, the user data includes multiple user property features；

Characteristic extracting module, the effective user property feature for extracting influence Information Communication from the user data, with And according to the user data, obtain the user behavior feature of the user data；

Sort module, is classified user data for the user property feature according to extraction and the user behavior feature, Obtain the classification of user in user data；

Processing module, for the classification according to user in described information data and the user data, is obtained and described information The corresponding user's propagation characteristic of data；

Forecast model module, for according to user's propagation characteristic, determining the forecast model of social network information stream row degree；

Prediction module, for being analyzed using the forecast model the information data produced in a period of time, to information flow Row degree is predicted；

The processing module, specifically for：

According to the classification of user, the user that Information Communication is participated in the information data in database is carried out according to the classification of user Statistics, the number of users of each classification of statistics is used as the corresponding user's propagation characteristic of described information data；

The forecast model module includes：

Model submodule is set up, for setting up the multivariate linear model based on user characteristics；

Submodule is trained, for user's propagation characteristic, as training set, to be trained to the linear model, obtain society Hand over network information prevalence degree forecast model；

Wherein, the user characteristics is normalized by the user property feature and the user behavior feature extracted Arrive.

7. system according to claim 6, it is characterised in that the system also includes：

Memory module, for by the user data and described information data Cun Chudao databases.

8. system according to claim 6, it is characterised in that the sort module includes：

Submodule is normalized, is normalized, obtains for the user property feature to extraction and the user behavior feature To user characteristics；

Submodule is divided, for according to the user characteristics, user data to be classified using clustering algorithm, number of users is obtained According to the classification of middle user.