Semantic Based Information Retrieval System Using Modified Inverse Document Frequency

Today, Information Retrieval (IR) provides users with documents that will satisfy their information need. Word sense ambiguity is a cause of poor performance in IR system. IR performance will be increased if ambiguous words can be correctly disambiguated. Word Sense disambiguation (WSD) is the task to assign the correct meaning to such ambiguous words based on the surrounding context. Various senses provided by WSD process have been used as semantics for indexing the documents to aid the information retrieval system. K-Nearest Neighbour (KNN) is used for effective text classification in WSD process and the Vector Space Model (VSM) is used for IR process. The cosine similarity method is used in both KNN and VSM to calculate the similarity in which term frequency and inverse document frequency (TF-IDF) scheme is used to calculate the weight of each word. There is a challenge that the original TF-IDF scheme eliminates the related senses although there is a related sense. This paper thus proposes the modified TF-IDF method, so called TF-MIDF, to solve the no-relevant problem by modifying the IDF equation to improve the accuracy of IR performance. By comparing the performance between the original IDF scheme and the MIDF scheme, the average precision results of the original TF-IDF method is 71% and the average precision results of the TF-MIDF is 80%. Therefore, the proposed methodology is more precise than the original method while retrieving the relevant documents of the required information