ITRANS Encoded Marathi Literature Document Relevance Ranking for Natural Language Flexible Queries

Abstract

Information Retrieval (IR) is an ever evolving concept of computer and communication applications. The emergence of mobile technology and its widespread service domains have given new dimensions to this as a research topic. Number of key players of search industry, are coming up with problems and their solutions with respect to SMS based Information systems and retrieval facilities. An extensive literature survey reveals that SMS based IR models for Indic languages are not available at large scale. In addition to this an universal transliteration encoding formats for Indic languages named ITRANS, has attracted us to resolve the problem of a suitable knowledgebase. With all these exploration we have formulated our research problem by combining both these features Viz. ITRANS encoded document corpus and SMS based Information Retrieval on this corpus. With reference of a few initiations of this emerging topic, we present an extended IR model for Marathi documents in ITRANS format. A collection of approximately hundred documents including Marathi songs, poems, abhangas and powada is incorporated as the document corpus for our experimental work. Several queries are formulated in ITRANS format and are used for the testing. The system developed by us ranks the documents based on the similarity scores applying Cosine Similarity measures. The documents in their ranked orders are checked by an expert for their relevance with respective queries. The feedback submitted by the expert is deployed to compute the relevance measure in terms of precision and recall. Finally the results are analyzed to correlate the similarity degree to the relevance of documents.