HomeResearchSocial NetworksVisualization of Social Network based on User Levels

Visualization of Social Network based on User Levels

1 Introduction

These years, social network has attracted considerable attention by researchers. Social network visualization could help us to recognize the structure of social network and the behaviors of people. This work researches the visualization of online social network and presents two visualization applications which visualize the local network and the overall network respectively.
The original data of social network is obtained from Sina Weibo, which is one of the biggest microblog website in China. In Sina Weibo, the relation between users is directed. Therefore, the relation network could be recognized as a directed graph. The visualization is based on user levels, which could be the indegree or Pagerank of the users. By partitioning users in different levels, we can stress the high level users and reduce the data volume.

2 Local Network Visualization

This work visualizes a target person and his/her follows. In the visualization, the target person is laid out in the center and the layout of follows is calculated by their indegree. Follows with large indegree has large scale and will be laid out in the outside of whole users. In contrast, follows with few indegree are small and will be placed in the inside. User could adjust the scale of the visualization, which will move all follows from inside to outside and display follows in different levels.

Figure 1: The original display of the local network visualization.

Figure 2: The zooming display of the local network visualization.

The visualization results are shown in Fig.1 and Fig.2. Fig.1 is the original display of the visualization, which stress the follows with large indegree. After adjust the scale, the display changes to Fig.2, which stress the follows with few indegree. Since the target person (in the center of Fig.2) has few indegree, he/she could be found with follows of same level.

3 Overall Network Visualization

3.1 The Layout of Stars

This work visualizes the overall social network. The data set contains 80.8 million users and their relations of Sina Weibo, which covers about 16% of all users. Since the data set is large, sampling is a necessary step of visualization. In this work, we sample top users with the N highest indegree and call them stars. Compare with other users, stars have more features (their fans set) which could be used to decide their layout.
In the visualization, we choose N=1000 and calculate the similarity between the top 1000 stars by their fans set. Then spectral clustering is used to cluster the stars to different category. Spectral clustering uses the Laplacian matrix to cluster data by their similarity. Each item of eigenvectors of the Laplacian matrix could be seen as a dimension reduction coordinate of an original data. This coordinate could be used to calculate the layout of stars. First, each coordinate of stars could be linear represented by the centers of each cluster. Then we place the centers on the two-dimensional plane which is the visualization interface. The layout of stars could finally be calculated by the linear representation.

Figure 3: The original visualization result of stars.

Fig.3 shows the original visualization result of stars. In Fig.3, the color of stars stands for the probability of which cluster they belong to. In this result, some areas have high density and difficult to distinguish each stars. The virtual force algorithm is used to solve this problem. Each of two stars are repulsed when they are close to each other and attracted when they away from each other. Each star is also attracted by his/her original position. Fig.4 show the adjusted visualization result of stars which is adjusted by the virtual force algorithm. It has lower highest density then Fig.3.

Figure 4: The adjusted visualization result of stars.

3.2 The Layout of Other Users

The layout of other users could be determined by the layout of stars. The position and color are both calculated by the mean of the stars they follow. Since the user count is huge, we cannot draw all users in the visualization. Therefore, we use layout picture to visualize other users. The layout picture is partitioned by its pixel. The brightness of each pixel is proportionate to the logarithm of the number of users in this pixel. And the color of each pixel is the mean of the color of users in this pixel. Fig.5 shows the visualization result of other users.

Figure 5: The visualization result of other users.

The final overall network visualization result is the combination of stars’ visualization and other users’ visualization, which is shown in Fig.6.

Figure 6: The overall network visualization result.

5 Online Visualization Address

Local network visualization (need an account of Sina Weibo):

6 Contact the authors

For any questions and comment about the website, please contact Xiangqian Che.