Identifying local events by using microblogs as social sensors

Local search is increasingly attracting more demand, whereby the users are interested to find out about places or events in their local vicinity. In this paper, we propose to use the Twitter microblogging platform to detect and rank local events of interest in real-time. We present a novel event retrieval framework, where both the contents of the tweets and the volume of the microblogging activity are exploited to locate an event happening in a certain area within a city that matches the user's interests as expressed in the form of a query. In particular, the framework measures unusual microblogging activities in a certain area and uses that as an indication of the occurrence of an event which is then used by the ranking function. Since the proposed event retrieval task is a new Information Retrieval (IR) task, we devise a methodology that is inspired by the conceptually similar IR problem of video segmentation to thoroughly evaluate our approach. Our evaluation is conducted on a set of tweets collected over a period of twelve days from different areas of London, as well as two sets of local events collected within the same period using crowdsourcing and local news sources in London. In addition to new insights on the factors that influence the development of an effective event ranking model, our empirical results show the promise and effectiveness of our proposed approach in identifying and ranking local events in real-time.