A K-means approach based on concept hierarchical tree for search results clustering

Search results clustering aims to facilitate users' information retrieval process and query refinement by online grouping similar documents returned from the search engine. It has stringent requirements on performance and meaningful cluster labels. Thus, most existing clustering algorithms such as K-means and agglomerative hierarchical clustering cannot be directly applied to the task of online search results clustering. In this paper, we propose a K-means approach based on concept hierarchical tree to cluster search results. This algorithm not only overcomes weaknesses of the classic K-means method: the results produced depend on the initial seeds and the parameter k is often unknown, but also satisfies the requirements of online search results clustering. Our method utilizes the semantic relation among documents by mapping terms to concepts in the concept hierarchical tree, which can be constructed by WordNet. We have developed a meta-search and clustering system based on our approach, followed by using an impersonal and repeatable evaluation solution. Experimental results indicate that our proposed algorithm is effective and suitable in performing the task of clustering search results.