Huan Sun

This semester for a class on HCI, I have been looking at dual screen experience that people interact with mobile devices while watching TV in China. I collected over 1 million Sina Weibo messages during ‘Chunwan’ (春晚), a state sponsored TV program to celebrate Chinese New Year, which is also a popular cultural event that almost everyone in China participates. While the social engagement aspect of this study is still ongoing, one of the side findings from this data is quite worth-noting: Within this data, over 60% users are female on Sina Weibo, but male users are almost twice likely to have a large number of followers(over 5000) than their female counterpart.

Sina Weibo public timeline API returns recently published posts, it also provides user’s gender, user’s number of followers and many other things, so I quickly checked a few descriptive statistics:

Out of 1,090,799 users in my data, 61.6% are female users and 38.4% are male users. In terms of their distribution in different categories of number of followers, we can see from the graph above that for both female and male, most users have followers less than 500, very few users have the number of followers over 5,000. It is probably because the influence of online users often reflects the power structure in real-life. Also Weibo’s realname verification system has made people with better professional positions, celebrities, government officials, scholars and journalists easier have larger number of followers.

It can be seen that there are more male users than female users in the 50,000 — 99,999 category even though the baseline tells us more female users. This made me wonder if it is more likely for men to perform better in the category of very large number of followers. So I decided to check the conditional probabilities of users in different number of followers categories given you are a man or a woman.

This result may confirm my guess that as a man there might be 0.5% chances to have over 100,000 followers, but this chance for women is only 0.3%. I am thinking about a more visually persuasive way presenting the inequality, so here comes this graph below:

The y-axis shows the percentage of men or women that deviates from their respective percentage baselines in different number of followers categories. It is clearer to see that men are more likely than women to occupy the most influential categories while they are also still likely to have smallest number of followers. For example, around 10% of users having over 100, 000 followers should have been women compared with baseline but they are men. These shifted numbers in the 10,000 and 5,000 categories are about 20% and 10% respectively.

Is this gender inequality a particular phenomenon on Sina Weibo? Is it also true on other social media platforms? I was fortunate enough to get help from Soshio, a company that assists me in acquiring data about Chunwan from both Tencent and Sina Weibo. Their data scraped from Sina Weibo confirms the 6:4 female/male user distribution, and the conditional probabilities of female and male users in these categories are list here:

What is the gender presence in Tencent Weibo like? Out of 76730 Tencent Weibo users, 56.3% are male, and 43.7% are female. This is contrary to Sina Weibo’s female dominated gender presence, see the graph below:

Even though there are more males on Tencent Weibo than females, males still occupy the categories with larger number of followers disproportionally compared with their 56.3% baseline:

Still on Tencent Weibo around 7% or 8% users in the 200–1000 followers category should have been women instead of men. I am not surprised on the fact that Tencent Weibo men occupy more influential positions as they do on Sina Weibo. Also there are significantly less users who have over 5000 followers on Tencent Weibo. One of the reasons might be that most celebrities who would have large number of followers have already chosen Sina Weibo as their primary social media site. Tencent QQ instant messenger’s legacy has attracted many QQ users to their Weibo platform but not necessarily those celebrities.

As I mentioned in the beginning, this result is a side finding from the data we need for another study. This blog post is more exploratory and I hope this preliminary finding could be useful for researchers interested in gender representation for future studies. To gather more representative data, one can randomize the dates when the data is scraped from these weibo platforms. Also if we get data through Weibo public timeline API, we as researchers are not sure how these messages returned to us are sampled. One way to deal with this issue is to have alternative source of data (as I got help from Soshio) to validate the findings.

 

Sina Weibo is one of the major micro-blogging services in China. According to a recent news report, Weibo has over 300 million users, and 100 million posts are generated everyday. In contrast with its counterpart in the West, a large number of studies on Twitter has been done, but the number of studies on Weibo is rather limited. Some researchers are constrained in studying Weibo because of language issues. For those who are sufficient in Chinese language, the research that has been done is mostly qualitative case studies. Thus, the question posed here is how to conduct large scale quantitative analysis of Chinese Weibo?
In comparison with qualitative case studies, large scale quantitative analysis on Chinese Weibo are limited because of the restrictions of Sina API to retrieve the data.The article on Chinese censorship by Bamman and others used the public timeline to retrieve the stream of Weibo posts. According to the Sina public timeline document, it returns the most recent 20 public posts. However, for researchers, it is still unknown if this 20 posts are representative to the larger Weibo posts set. Another approach is to fetch posts through users’ ID. WeiboScope, a project at Hong Kong University, has archives of posts from over 300 thousands users who have at least 1000 followers. During their presentation at the 10th Chinese Internet Research Conference, they showed their approach to the issue of censorship: they retrieve these users’ statuses every day, and compare to their archive to determine which posts are deleted. Another tool to retrieve Weibo data I tried is DiscoverText, in which users can set up their words to search, and searching time intervals, and the tool will return posts containing the words. Using both tools, I did a search of posts containing the word “Wukan (乌坎)” sent from March 23 to March 27. WeiboScope returned me around 200 items while DiscoverText fetched around 1000 items. I am not quite sure about the reasons for the distinctive results, but my guess is that one only looks at popular users’ posts while the other fetches from the general public timeline.

Despite these restrictions, I still think the chances of conducting large scale quantitative research on Chinese Weibo lie in the refinement of the research questions. The limited sample set can still be representative to the limited scope of the questions. Specifically, instead of asking about the frequencies of certain key words appearing on all the Weibo posts, researchers can shift their perspective to a limited scope that only focuses on an interesting group. For example, Weibo provides lists of users in different occupations, and if the research questions only relate to posts produced by a certain type of users, such as journalists, researchers can directly make use of the list.

Following this idea, Wei Wei and I have been working on a project mapping out the NGO networks on Weibo. The figure above shows who mentions whom among NGOs on Weibo. The node size is the in-degree centrality and the colors show different communities. The NGOs are largely clustered by their working domains such as saving animals and evironment protection. Large NGOs, for example, WWF, UNICEF and China Foundation Center are frequently mentioned by other NGOs on Weibo. We are still in the process of collecting data and especially welcome any suggestions on data scraping on Weibo and network analysis methods. The broader topic of NGO networks we are interested in is how the social media empowers grassroots NGOs. Resources for media coverage from mainstream media on grassroots NGOs are limited because they cannot compete with government affiliated NGOs for media attention, so we contend their integrated use of social media may amplify their voices and construct a network that challenges the offline hierarchical positions. Although the figure above shows large international organizations occupy key positions in the network, we might also look into dynamic data to see the change of the structures instead of this static picture.

The debate of whether the use of Internet can enhance the offline political participation has mostly drawn empirical materials from democratic countries, but this study directs our attention to the authoritarian regime, China. Instead of establishing direct link between Internet use and offline participation, this paper focuses on mechanisms in the more nuanced process from informational use of social networking sites by Chinese students to their political participation. Using data collected in spring 2010 from a random sample of students of a university in Beijing, this study builds up structural equation models to test the mechanisms in the process. Main findings include the positive relationship between time spent on SNS and their offline political participation through paths of political informational behavior and civic actions. Political attitude is proved as mediation between the amount of time spent on SNS and political informational actions. This article also provides evidence that the purposes of using SNS significantly influence the types of participation.

Full paper is available here. This paper has been presented at 2012 China Internet Research Conference.