Analyzing reading behavior by blog mining

This paper presents a study of the various aspects of blog reading behavior. The analyzed data are obtained from a Japanese weblog hosting service, Doblog. Four kinds of social networks are generated and analyzed: citation, comment, trackback, and blogroll networks. In addition, the user log data are used to identify readership relations among bloggers. After analysis of more than 50,000 users for about two years, we reveal some interactions between social relations and readership relations. We first show that bloggers read other weblogs on a regular basis (50% of weblogs that are read at least three times are read every five times a user logs in). We call this relation a regular reading relation (RR relation). Then, prediction of RR relations is done using features from the four kinds of social networks. Lastly, information diffusion on RR relations is analyzed and characterized. Results of this study show that the blogs in RR relations have an important role in bloggers' activities. We find the features which have a correlation with RR relations.