Clustering in tweets using a fuzzy neighborhood model

Clustering of keywords in tweets is studied. A series of tweets is handled as a sequence of words and an inner product space is introduced to a set of keywords on the basis of positive definite kernels using a fuzzy neighborhood defined on that sequence. Methods of agglomerative hierarchical clustering as well as c-means clustering are applied. Pairwise constraints are moreover introduced to improve interpretability of clusters. Real tweets are analyzed with discussion of the resulting clusters.