Abstract: Due to the advance of many social network applications, social group feature analytics are attracting a lot of attention. In the meantime, microblogging, as a kind of social network application, attracts more and more people to use it. With the utilization of bigger and broader crowds over microblogging, surveying massive user features will be an important aspect of exploitation of crowd-sourced data. For better understanding microblogging user group features, in this study, a user classification approach was proposed by means of Boolean operations and it is easily find different microblogging user group features by this approach. In the experiment, some facts were discussed on the exploratory survey to exploit a great deal of microblogging data and how to analyze the features of the different user groups.
INTRODUCTION
In recent years, the emergence of microblogging website for encouraging crowds to share their information in almost real time across the open space; it is already an important communication and information exchange platform for internet users. However, how to make good use of mass of information brought by the new media is a new challenge.
To study marketing and recommendation based on microblogging, a top priority, it is necessary to study the characteristics of the user groups of the microblogging. Therefore, this study focuses on analysis of microblogging user group characteristics.
As a newly emerged type of social network, Microblogging is increasingly extending its role from communication way into an important platform for sharing real-time information (Hughes and Palen, 2009).
In the first applications area, microblogging user is generally envisioned as social sensors. In the picture, everyone in the microblogging can be viewed as a sensor. Sakaki et al. (2010) studied the real-time feature of microblogging user's posts and proposed an event notification system able to monitor posts and deliver notifications promptly. Demirbas et al. (2010) proved the precision of using Twitter to deduce information from two case studies.
In the second application area, microblogging user features is used to enhance recommendations. Diakopoulos and Shamma (2010) proposed an approach based on tweet data analysis, to help professionals better understand the dynamics change of people’s sentiment. Chen et al. (2010) presented an method to rate tweets mainly based on the popularity of embedded URLs. To support tweet ranking, Duan et al. (2010) classified tweet features into three categories: content relevance features, twitter specific features and account authority features.
The related work has proved the effectiveness and efficiency of leveraging microblogging to facilitate information dissemination and recommendation. However, according to the literature, the research on microblogging user groups feature analysis is just started. In this study, a novel user group features analytical method was proposed that helps companies to better understand microblogging users crowd characteristics.
MATERIALS AND METHODS
In this study, an efficient method for microblogging users classification based on Boolean algebra was proposed. K-means cluster approach was also applied to do further analysis on important individual user groups.
The data sample in the research comes from weibo-the largest Chinese microblogging website. By means of firefly (Fig. 1), independently developed software of microblogging data crawling by the authors, the researchers randomly collect data from weibo users. Considering the influence of “Zombie Fans” on analysis, data-filtering prerequisites was set in pre-processing to clear away “Zombie Fans” as follows: Fans Number >δ1, microblogging number >δ2, Friend Number >δ3 (δ1, δ2, δ3 are slightly larger than zero).
| Fig. 1: | Microblogging crawling software, firefly |
After the preprocessing, microblogging users matrix was established and do the classification based on Boolean algebra as follows:
| • | Establish n-tuple ui = {ai1,…,aij,…, ain} to denote all features j of user i, i = 1, 2,…, m; j = 1, 2,…, n. And the n-tuple λi = {λi1,…, λij,…, λn} is established to describe the corresponding attribute value of above-mentioned specific features of user i |
| • | Let θj is the median of the set λi = {λi1,…, λij,…, λn} i, i = 1, 2,…, m; j = 1, 2,…,n. Define a new variable 0 ηij: |
| (1) |
| Through Eq. 1, to map λij to ηij. And then, to use 0-1 variable to represent the features use i, ui = {ηi1,…, ηij,…, ηin}, ηij is 0 or 1, i = 1, 2,…, m | |
| • | The microblogging users features matrix is now a 0-1 matrix: U = {u1,…, ui,…, um}T, i = 1, 2,……, m |
| • | A Boolean operation classification algorithm (BOC) was designed for classifying microblogging users into different classes. Firstly, a set was defined as follows: |
| • | S = {Bf|Bf = {bf1,…, bfj, …, bfn}} And, f = 1, 2,…,g,…, h,…,p; j = 1, 2,…,n; And bfj is 0 or 1, the number of bf is 2p; To any Bg and Bh, ∃bgj, bhj⇒ bgi^bhj are not zero at the same time |
Secondly, based on the definition and explanation above, Pseudo code algorithm of BOC based on Boolean operation is as follows (Fig. 2).
On the basis of the BOC classification, K-means method was used according to its specific item attribution by each class of users respectively and then extract the clusters with relatively larger amount of elements as the representatives for typical analysis, aiming to find out the typical characteristics of each class of users, which would provide better service for understanding microblogging users at the meantime.
| Fig. 2: | Pseudo code of BOC |
Through this method, it is easy to find the typical user cluster in one specific user group and then, sampling and descriptive statistical analysis can done from this typical user cluster and finally summarize the typical characteristics of these user groups.
EXPERIMENTS
Experiments study and analysis: Basing on the above theoretical model, an experimental analysis can be done on the features of weibo users. weibo is the largest Chinese microblogging website, it owns nearly 500 million users. By analyzing main features of weibo users, this study proposes a classification idea based on some users features as follow: friends number, fans number, posts number and their total online time. With the weibo API, information was collected from 42079 randomly selected users including information listed above, as well as some additional user information (e.g., verified user or non-verified user).
In addition, taking into account the influence of ‘zombie fans’ on integral mean by setting up the following filter conditions when calculating the integral mean: number of fans >δ1, microblog number >δ2, Friend Number >δ3. After data preprocessing, 34050 user records were met the filter conditions. Table 1 is an example; it shows the mean, mode and median of these four variables of some users meeting the filter conditions (only non-verified users included, part of the total population).
The Boolean operation classification algorithm mentioned in chapter was adopted, after classification, the proportions of each user class were shown in the following Fig. 3.
Deep analysis of the classification result: By taking the K-means method to cluster each type of users, aiming to find the most representative and typical group cluster of each type of users (the cluster which has the largest amount of element in one class ) and making analysis of these representative group cluster.
Each type of users was clustered into five clusters and choose the clusters with the largest number of element as a example of analysis, one of the example of the results can be seen as in Table 2.
Use this clustering method to analyze all 16 types of users group. Then sample among those clusters which contain most users of each type, followed by artificial check towards the sample.
According to the analytical result, it could be find out that there are obvious differences between non-verified and verified users:
| • | Non-verified users include organization users and individual users. Organization non-verified users mainly write micro-blog and gather information. Theirmicro-blogs are often related to a specific topic. Those who gather information are mainly small organizations and they have the demand for information from other users in order to do research on certain industry. Individual non-verified users are mainly normal users. These users mainly follow their interested users to get information they want, write micro-blog about their daily life and use it as a tool to communicate with friends in real world |
| • | The verified users are rather centered in term of user type, mainly focused on (0, 1, 1, 1) type and (1, 1, 1, 1) type. These two types of users share the same feature of having a huge quantity of fans. They are usually well-known organizations or famous persons in different field and thus be invited to join weibo. These users already have a great influence in real world, so they can easily attract a huge number of followers. They use it as a platform to release information to the public. They are the most social active users on weibo and they play the role as the central nodes in microblogging social network |
| Table 1: | Statistical characteristic of qualified
non-verified users |
| Table 2: | Clustering analysis example of a specific
type of non-verified users |
Circle represents the cluster with the
largest number of element as an example of analysis |
|
| Fig. 3: | Proportion of each user category |
CONCLUSION
This study introduced a novel approach to analyze microblogging users group features. Especially, this study also do experimental study to prove the validity of the approach based on a real dataset of weibo users. In the future work, much deeper user group features analysis approach will be explored and tested for big data.
ACKNOWLEDGMENT
This study is supported by a grant from Humanities and Social Science Fund of Chinese Ministry of Education (No. 11YJA870017).