Mining Events through Activity Title Extraction and Venue Coupling

In this paper, we discuss the challenges to construct an event/activity search engine through activity title extraction and venue recognition as well as relation coupling. While distant supervision is a common technique to speed up training data preparation, it does not always work on social network. For activity titles, they may contain other entities such as person, venues and temporal expressions, which could be much longer than general named entities (such as person names). Venue recognition is another challenge, as they could be an address, a specific point-of-interest like an restaurant or organization name, or across a neighborhood. Another problem is how to determine the venue and correct date of the event when multiple venues or temporal expressions are recognized in a message. In this paper, a sequential pattern mining approach is applied to discover rules for coupling the recognized place names with the activity title in discussion. The experimental result shows that our approach is beneficial and practical for locating social activity venues.