Enhancement of Search Results Using Dynamic Document Seed Reranking Algorithm

Abstract

We proposed an algorithm to improve the precision of top retrieved documents by re-ordering the retrieved documents in the initial retrieval. To re-order the documents, we first automatically extract key terms and key phrases from top N retrieved documents and generate a document index for each document. Using the standard similarity metrics, a document similarity matrix is generated for these documents. The document similarity matrix is considered as an adjacency matrix, where the nodes are documents and the distances are their similarity measures. The objective of this algorithm is, to rerank the documents so as to minimize the similarity mean absolute distance between them. Moreover, the user can choose a document of interest as the seed document and initiate the reranking algorithm by which documents are reranked based on is similarity distance from the seed document. From the experimental results, it is demonstrated that the algorithm reduces the mean absolute difference. Further it is proved that the proposed reranking algorithm minimizes the mean absolute distance between the top N results obtained from search engines and helps users to rerank documents based on any seed document as a query.