Learning to rank for determining relevant document in Indonesian-English cross language information retrieval using BM25

Abstract

One important task in cross-language information retrieval (CLIR) is to determine the relevance of a document from a number of documents based on user query. In this paper we applied pointwise learning to rank in SVM (Support Vector Machine) to determine the relevance of a document and used BM25 (Best Match 25) ranking function for selecting words as features. We did the experiment in Indonesian-English CLIR The results show an average ability of SVM to identify relevant documents is 88.51%, while the average accuracy of SVM to identify non relevant documents is 88%.