A Probabilistic Afaan Oromo Information Retrieval System

Abstract

Developing information retrieval (IR) system for Afaan Oromo language fills the gap between the information stored and the people who use this language as their mother tongue or as a second language to search relevant documents as per their information need. As the language is the most widely spoken language in Ethiopia, it is important to find out ways to access the information available in this language. Therefore, in this study a probabilistic Afaan Oromo IR system is developed. There are two main operations in the designed IR system: indexing and searching. Documents are indexed using inverted index file and searching operation is done based on probabilistic concept in which Binary Independence Model (BIM) is used. The retrieval performance of the designed IR system is calculated using the three most common IR effectiveness measures; precision, recall, and F-measure. In probabilistic based IR process there are at least two cyclic document retrieval processes: initial retrieval and after query terms are reweighted. The two iterative retrieval processes are done by three different users; each of the users used on average three queries out of the ten prepared for testing. At initial retrieval, the system scored 86% and 41% precision recall and F-measure 53% respectively. After relevance feedback, the performance became 73% precision, 64% recall and 68% F-measure respectively. As the system retrieval performance evaluation shows, the recall at initial retrieval has improved by 23% and 15% F-measure is well after relevance feedback; which indicates that the system become more effective in terms of retrieving more relevant documents than the first attempt considering the user?�?s relevance feedback. There is a possibility of improvement for the designed system by extending it to another query reformulation technique such as query expansion, in which such polysemy and synonymy problems can be solved.