Abstract

Information Retrieval in a Telugu language is upcoming area of research. Telugu is one of the recognized Indian languages. We present a novel approach in reformulating item terms at the time of crawling and indexing. The idea is not new, but use of synset and other lexical resources in Indian languages context has limitations due to unavailability of language resources. We prepared a synset for 1,43,001 root words out of 4,83,670 unique words from training corpus of 3500 documents during indexing. Index time document expansion gave improved recall ratio, when compared to base line approach i.e. simple information retrieval without term expansion at both the ends. We studied the effect of query terms expansion at search time using synset and compared with simple information retrieval process without expansion, recall is greatly affected and improved. We further extended this work by expanding terms in two sides and plotted results, which resemble recall growth. Surprisingly all expansions are showing improvement in recall and little fall in precision. We argue that expansion of terms at any level may cause inverse effect on precision. Necessary care is required while expanding documents or queries with help of language resources like Synset, WordNet and other resources.