Pseudo Relevance Feedback by linking WordNet for Expanding Queries in Information Retrieval Process

Abstract

precision without compromising recall growth. The proposed framework is aimed to test on Telugu item sets. Telugu is one of the Indian Languages belongs to Dravidian Family. Indian languages are rich in morphology (3). Words in Telugu language have more variants, this identical feature is also a major problem to apply query expansion techniques. The Term selection plays a key role in retrieving relevant results for user query. In this method WordNet is used to extract candidate words for query expansion. We perform query expansion by generating lexical paraphrases of queries. These paraphrases replace content terms in the queries with their synsets. The resources used for selecting such term required Telugu POS Tagger or Morphological Analyzer, Stemmer and WordNet. This paper is organized into five chapters with introduction as Chapter 1. Chapter 2 is all about Query Expansion and the way how we use Query Expansion. Complete framework has been discussed in Chapter 3. Stop words are removed from each of these questions, and are passed to a shallow stemming program. The shallow stemmer is a rule based stemmer which stems the topic words to return the stem (4). Use of only global models for query expansion may not affect the results. The integrated approach for query expansion gives observable growth in performance. In this paper our approach uses PRF (Local Model) with WordNet (Global Model). This provides a new platform for query expansion, which reduces the user interaction and preserves the meaning of the query. Off course using WordNet alone may not give effective results in terms of precision as the web is drastically increasing with new terminology day by day. In addition to this approach, to overcome the limitation of WordNet we also propose an online WordNet that would give updated lexical relations to the query terms. This paper aimed to find the difficulties in expanding the Telugu query and propose a solution to improve the Search results in Indian Information Retrieval (IIR). Once the framework is built the same model can be extended for other Indian Languages. Throughout the Paper examples are given in Telugu Language with phonetic representation.