Since its foundation in 2006, Twitter has enjoyed a meteoricudrise in popularity, currently boasting over 500udmillion users. Its short text nature means that the serviceudis open to a variety of different usage patterns, whichudhave evolved rapidly in terms of user base and utilization.udPrior work has categorized Twitter users, as well asudstudied the use of lists and re-tweets and how these canudbe used to infer user profiles and interests. The focus ofudthis article is on studying why and how Twitter usersudmark tweets as “favorites”—a functionality with currentlyudpoorly understood usage, but strong relevance forudpersonalization and information access applications.udFirstly, manual analysis and classification are carriedudout on a randomly chosen set of favorited tweets, whichudreveal different approaches to using this functionalityud(i.e., bookmarks, thanks, like, conversational, and selfpromotion).udSecondly, an automatic favorites classificationudapproach is proposed, based on the categoriesudestablished in the previous step. Our machine learningudexperiments demonstrate a high degree of success inudmatching human judgments in classifying favoritesudaccording to usage type. In conclusion, we discuss theudpurposes to which these data could be put, in theudcontext of identifying users’ patterns of interests.
展开▼