Similarity Based Hierarchical Clustering with an Application to Text Collections

Julien Ah-Pine (1) , Xinyu Wang (1)

1 ERIC - Entrepôts, Représentation et Ingénierie des Connaissances

Lance-Williams formula is a framework that unifies seven schemes of agglomerative hierarchical clustering. In this paper, we establish a new expression of this formula using cosine similarities instead of distances. We state conditions under which the new formula is equivalent to the original one. The interest of our approach is twofold. Firstly, we can naturally extend agglomerative hierarchical clustering techniques to kernel functions. Secondly, reasoning in terms of similarities allows us to design thresholding strategies on proximity values. Thereby, we propose to sparsify the similarity matrix in the goal of making these clustering techniques more efficient. We apply our approach to text clustering tasks. Our results show that sparsifying the inner product matrix considerably decreases memory usage and shortens running time while assuring the clustering quality.

Mots clés: Agglomerative hierarchical clustering; Lance-Williams formula; Scalable hierarchical clustering; Kernel machines; Text clustering

Domaines: Intelligence artificielle [cs.AI]; Informatique et langage [cs.CL]

Julien Ah-Pine, Xinyu Wang. Similarity Based Hierarchical Clustering with an Application to Text Collections. Intelligent Data Analysis, Oct 2016, Stockholm, Sweden. pp.320 - 331, ⟨10.1007/978-3-319-46349-0_28⟩. ⟨hal-01437124⟩