PhD Dissertation Proposal by CHONG Wen Haw

Context Recovery in Location-based Social Networks

This proposal seeks to recover two types of contextual information in LBSN: venue and semantics. We focus on flexible-usage and coarse-grained LBSN platform such as Twitter. In Twitter, the venue context is often missing. Although users can geocode their tweets with location coordinates, such geocoded tweets do not uniquely identify the posting venue, e.g. a restaurant, especially in cities where venues are densely packed. Furthermore, many users do not geocode their tweets in the first place. Therefore, we study the problem of fine-grained tweet geolocation, which links tweets to their specific posting venues.

Besides the venue context, it is also useful to understand what users post about in their content. This leads to the problem of recovering the semantic context. In concrete terms, we seek to associate the right semantic concepts to the content. We cast this as the entity linking problem. Given mentions of named entities, we seek to link the mentions to the correct referent entity in some knowledge base. Entity linking in LBSN content is a highly challenging problem due to the extremely short documents involved. To mitigate this challenge, we shall exploit certain properties of LBSN, e.g. collective exploitation of content that are posted close in space and time.