Retrieving web search results using Max–Max soft clustering for Hindi query

Information retrieval (IR) is the process of finding relevant information from the millions of unstructured documents on the web. Despite of all the success in IR, it faces many problems such as lexical ambiguity, compound word formation and language morphology etc. To address the ambiguity problem, in this paper the authors proposed a graph based soft clustering method which improves the performance of IR system. Initially text snippet words are taken for constructing a co-occurrence graph corresponding to the Hindi query given by a user. Then other words (relevant to the query terms) present in the text corpus are added on the basis of the dice coefficient. For each interpretation of the user query, we retrieve results in the form of a web cluster. Sometimes more than one interpretation of the query are closely related, therefore many results returned from IR corresponding to these interpretations are common. This type of issue can be better dealt by using soft clustering method, so we use Max–Max soft clustering approach. We use various similarity measures like word overlap, degree overlap, token overlap and average similarity respectively for ranking the results within each cluster. This is the first attempt to fuzzy IR for a query in Hindi language, experimental evaluations shows promising results.