Automatic spelling error detection and correction for Tigrigna information retrieval: a hybrid approach

Abstract

This paper proposes a hybrid approach to design and implement query spelling error detection and correction (SEDC) for Tigrigna information retrieval (IR). Our approach, which is the main contribution to this work, is fast and robust to achieve better performance and also helps the users to easily insert their corrected queries to retrieve relevant information from the IR. This is achieved by combining the normalized measure of bigram overlap using the Jaccard coefficient (J.C) technique, a dynamic programming algorithm for edit distance, and probability of occurrence, which were used to make suggestions for the misspelt words. Our approach was evaluated on the SEDC subtasks separately. It achieved an F-measure of 98.85% on the spelling error detection subtask and an accuracy of 95.36% on the spelling error correction subtask. Thus, a comparison was conducted between our approach and the existing Tigrigna spell checker. It is found that our approach outperformed the existing spell checker and shows a 5.36% improvement in accuracy. This is by far the most promising result with regard to correcting the misspelt users’ queries and improving the overall performance of the IR.