E Constant

In an ongoing research project, we consider Twitter, the social network service, as a market, in which people produce and consume information by tweeting and following others. Let’s say user A and user B are two frequent tweeters. One interesting question would be “how similar” is the information contained in A’s and B’s tweets. One way to solve this question is of course do a semantic analysis of the two’s tweets. Another way, as we propose, is look at their respective followers. Presumably everyone has a preference on what kind of information she consumes, so her following someone should tell that, to some extent, she likes the one’s tweets. Therefore we can predict that if A’s and B’s tweets are similar, then they should have followers of similar preferences; inversely if A’s and B’s tweets are quite different, then the followers they attract should have quite different preferences.
We define Followers-Similarity-Index (FSI) of A and B as

For fun, I computed the FSIs for the following Chinese twitter users @flypig, @junyu, @turingbook, @mozhixu, @glif, @williamlong, @DashHuang, @Stefsunyanzi, @virushuo, @WangShuo, @xiaolai, @zuola, @wglxh, @wangpei, @gaojiamin, @ag108lau, @arthur369, @mranti, @songshinan, @hecaitou, @duanzi, @isaac, @shizhao, @luoyonghao, @jaqi, @jason5ng32, @maoz, @izlmichael, @roseluqiu, @livid, @onlyswan, @aiww, @fzhenghu, @zhangfacai. There are 34 people and hence combinations. I used the data collected at 0:00 Feb 17 CST. Below are the 561 computed FSIs sorted.