Wrapper Based Feature Selection in Semantic Medical Information Retrieval

Information Retrieval (IR) has emerged as an empirical system to evaluate document collections and broadly encompasses finding unstructured documents that satisfies the query given to a large collection of documents. Document frequency measures are popular ways to extract feature from unstructured document but suffer from a huge dimensionality feature space. Adding semantics to the extracted features grossly increases the feature space. In this paper, a wrapper based feature selection technique for semantic IR is proposed. Semantics are the base for Information Retrieval's content description and query processing techniques. Semantic Similarity is about computing similarity between conceptually similar but lexically dissimilar terms. This work explored IR semantic features extraction based on word co-occurrence from web pages. Feature reduction is achieved through use of wrapper based feature selection technique comprising Latent Semantic Analysis (LSA) followed by Shuffled frog algorithm. The proposed technique showed improved Precision and Recall when evaluated using Decision stump, BF tree, and Random tree.