Afaan Oromo Text Retrieval System Using Probabilistc Approach

Abstract

Modern information retrieval (IR) system should be able to take natural language queries and retrieve documents written in natural languages. Since obtaining relevant information from a collection of informational resources in Afaan Oromo is very important for Afaan Oromo speakers, developing a system that help users of Afaan Oromo is mandatory. That is why this study is envisioned to make possible retrieval of Afaan Oromo text documents by applying techniques of modern information retrieval system. In the developed Afaan Oromo prototype, Probabilistic approach was used as an information retrieval models and precision and recall measurement were used as the performance measurement or evaluation technique. In addition to this, F-measure is also used to show the accuracy of the system by avoiding the precision recall trade-off. The similarity measurement used was also BM25Similairty. Apache Solr was also used as an environmental programming language to achieve the evaluation goal. Afaan Oromo text retrieval is evaluated using 158 documents and 13 arbitrarily selected queries that can determine the effectiveness of retrieval using the precision-recall. The average result obtained by our evaluation before the addition of synonymy was 72.91% precision and 86.8% recall respectively. After the addition of synonymy and polysemy the value was changed to 71.39% average precision and 90.5% average recall. The F-measure for the evaluation before synonymy addition was 79.25% and after addition changed to 79.82%. The addition of synonymy improves the system performance by 0.57%. The study therefore, experimentally prove that the addition of the thesaurus system can improve the system performance. Spellchecking, pagination, hit highlighting and autosuggestion are also possible in the developed prototype for Afaan Oromo.