Mining POI Alias from Microblog Conversations

In location-based analysis for microblogs, it is important to know if two toponyms refer to the same point-of-interest, i.e., alias. However, existing online knowledge bases are often incomplete or inaccurate for toponym alias data, especially for those used in informal conversations. In this paper, we propose a method for extracting compatible toponyms from microblog conversations. We first extract a number of coordinate-associated toponyms, then use compatibility measures to identify compatible toponyms. We propose three compatibility measures, namely, geographical closeness, surface name similarity, and association similarity. We show that by combining these measures and using particle swarm optimization for weight tuning, we can reach a high matching accuracy. The finding of this paper can be useful for improving location-based analysis as well as extending existing knowledge bases.