Socio Textual Mapping

Abstract

Location-based social networks are a source of geo-spatial data enriched by textual information, such as news, travel blogs, tweets and user recommendations. Such data may describe an event, an experience or a point of interest that is relevant to a user. In this vision paper we propose to describe a spatial region by the thoughts, ideas and emotions frequently and recently expressed by people in that region. For this purpose, we envision to extract features from geo-textual data, which capture not only the vocabulary, but also current topics and current general interests. We formally define the problem of drawing a socio textual map using geo-textual data and identify the necessary steps towards this vision: We represent each region as a stream of text messages such as tweets. In each region, we maintain a feature representation of text messages. We define a dissimilarity measure between such collections to assess the similarity between two regions. Using this measure, we utilize a metric clustering approach to obtain a social map of similar regions. We present a proof of concept by implementing the aforementioned steps with initial solutions. This proof of concept shows that an initial solution, which clusters the feature representations of regions, also yields clusters having regions that are spatially close. We theoretically explain this proof of concept by Tobler's first law of geography.