Revealing microblogger interests by analyzing contributions

Abstract

Personal blogs are online diaries. Bloggers share their comments, opinions, feelings and experiences on their blogs. People, who share similar interests but typically do not know each other in person, follow each other's updates through their blogs. Microblogging is a kind of blogging in which users' contributions consist of shorter messages. Microblogs may express what the microblogger is doing or thinking, or inform about something like news, entertainment, good deals, etc. Since microblogging is suitable for mobile use, and short microblog contributions do not require much attention as long, well-structured blog posts, microbloggers tend to post their updates more frequently than regular bloggers, which results in a larger number of microblog posts. As a result, the microblogosphere presents a vast amount of short messages that arrive at high speed. In microblogging systems, there is the problem of nding users of interest { as people are multifaceted and often escape notice. When deciding whether to follow a user who may be following us or followed by a friend, it would be useful to know something about them. Usually, a person who wants to get an opinion about a microblogger can look at the metadata supplied by the system, examine other microbloggers in communication with that particular microblogger, or read her contributions. A user's dynamic and continuously updated contributions reveal her interests in that particular system. However analyzing microblog posts is more di cult than analyzing blog posts. Compared to well written, structured blogger posts, microblogger posts are restricted in size and plenty. The sheer volume and fragmented nature of microblogs make it di cult to assess the characteristics and interests of a user. Manually analyzing microblog contributions would be overwhelming due to their quantity and fragmented nature. In this study, a model for automatically revealing microbloggers' characteristics and interests is proposed. Proposed approach supports the following: analyze all signi cant words uttered in posts, analyze external references existing in posts, analyze internal references existing in posts, and examine user meta information in the microblogging system An implementation of this model, which uses the API of the highly successful and widely used microblogging service Twitter is presented. The results of this work are discussed in terms of determining the speci c characteristics of particular groups of users as well as the comparison of individual microblogger contributions. Such information could be utilized in deciding who to follow for what purpose.