Search Result Clustering through Density Analysis Based K-Medoids Method

After obtaining search results through web search engine, classifying into clusters enables us to quickly browse them. Currently, famous search engines like Google, Bing and Baidu always return a long list of web pages which can be more than a hundred million that are ranked by their relevancies to the search key words. Users are forced to examine the results to look for their required information. This consumes a lot of time when the results come into so huge a number that consisting various kinds. Traditional clustering techniques are inadequate for readable descriptions. In this research, we first build a local semantic thesaurus (L.S.T) to transform natural language into two dimensional numerical points. Second, we analyze and gather different attributes of the search results so as to cluster them through on density analysis based K-Medoids method. Without defining categories in advance, K-Medoids method generates clusters with less susceptibility to noise. Experimental results verify our method's feasibility and effectiveness.