TV program detection in tweets

Abstract

Posting comments on social networks using second screen devices (e.g., tablets) while watching TV is becoming very common. The simplicity of microblogs makes Twitter among the preferred social services used by the TV audience to share messages about TV shows and movies. Thus, users' comments about TV shows are considered a valuable indicator of the TV audience preferences. However, eliciting preferences from a tweet requires to understand if the tweet refers to a specific TV program, a task particularly challenging due to the nature of tweets - e.g., the limited length and the massive use of slangs and abbreviations. In this paper, we present a solution to identify whether a tweet posted by a user refers to one among a set of known TV programs. In such process, referred to as item detection, we assume the system is given a set of items (e.g., the TV shows or movies) together with some features (e.g., the title of the TV show). We tested the solution on a dataset composed by approximately 32000 tweets, where the optimal configuration reached a precision of about 92% with a recall equals to about 65%.