Analyzing Relationship of Words Using Biased LexRank from Geotagged Tweets

A place is one important factor that affects the relevance between two words. For example, in places where tourists can view Tokyo Tower and cherry blossoms at the same time, it is considered that these words are related. However, in other places, these words would not be considered related. In this research, we propose a method to extract the relationship between words in an area based on this hypothesis, using geotagged texts obtained from Twitter. To extract the relevant words of a word posted at each place, our approach uses propagation of words co-occurring with co-occurring words in texts. We apply Biased LexRank, adapted from PageRank, to a graph constructed from co-occurrence relationships in each area. We also determined the places where there are two characteristic objects based on the relationship of the words. We visualize and discuss such places based on the relationship, such as places where you can appreciate "Tokyo Tower" and "cherry blossom" together.