Exploiting Temporal Locality to Determine User Bias in Microblogging Platforms

Bias is the human tendency to favor one side of a discussion in argumentation, lacking neutrality and balance. Determining user biases is  key to applications that process, interpret, and recommend content generated by those users in social media platforms. This paper addresses the problem of determining (in a supervised way) biases of microbloggers from the stream of messages. In this paper, we evaluate the use of a new criterion to identify user bias in social media systems: the temporal locality among users that have similar bias, i.e., the fact that people having similar biases express at about the same time. We show that this remarkable property indeed holds in some domains discussed in Twitter and may be explained mainly by the real-time use of the microblogging platform, i.e., users with similar biases react altogether to the outcome of events that are in accordance with their opinion (e.g., their favorite soccer teams scores a goal). Besides the precision of the computed biases, our proposal presents two major advantages that are consequences of not considering content at all (only temporal information is used). First, it is very efficient, i.e., a modest hardware can process on the fly the whole stream of messages about a popular topic commented in Twitter. Second, we believe that it may be applied to a wide range of domains regardless the language in which the messages are written. The experimental section of this paper reports the efficient learning of precise biases in both sportive and political contexts where the numerous messages are either written in English or in Portuguese.