In Twitter, a user follows other users for various purposes, e.g., for
information gathering, for personal communication, and for reading a
chat by celebrities. As a result, the intention of followers behind
follow links is different from case to case. We classify follow links
in Twitter based on the followers' intention along with three
classification axes: user-orientation, content-orientation, and
mutuality. The combination of these three axes covers most major
types of followers' intention found in Twitter. We collected 1760
Twitter follow links through a questionnaire and we found that (1)
user-orientation and content-orientation have weak positive
correlation, and user-orientation also has weak positive correlation
with mutuality, but content-orientation has no correlation with
mutuality, (2) content-oriented follows are more frequent than
user-oriented follows even among communication-oriented users, and (3)
the users have no clear intention for more than 20% of their follow
links. We then constructed classifiers with various features of the
followee, the follower, and their relationship. We also developed a
method of classifying “lists” in Twitter into information lists and
community lists, and used the types of lists including the followee.
Our experiments show that (1) no single property is a prominent
discriminator, and (2) classification accuracy for follow links by
information-gathering users are higher than that for
communication-oriented users.
/ Fax: +81(Japan) 75-753-5978 / Office: Research Bldg. #7, room 404