Knowledge-based Query Suggestions for Retrieval Improvement

Abstract

In recent past entity searches have been in the spotlight. Keyword based entity searches (e.g. Harry Potter) fetch diverse results because of multiple occurrences (ambiguous nature) of entity and show low precision. To address this problem this work presents an Expansion-Based-Query-Suggestion Scheme Entity-to-Entity (E-to-E). Our scheme uses attribute-oriented definitions as a knowledgebase to produce query suggestions. Proposed scheme can distinguish and find ambiguous entities like Harry Potter with high precision. Using the query-URL co-occurrences we evaluate the scheme performance while identifying whether the fetched URL is a good representation of the searched entity or not. We use evaluation matrix based on MAP@k, coverage, success rate, precision, recall and F-measure to prove the effectiveness of the approach. In our experiments, the proposed scheme is compared with three different baselines. Our scheme achieved MAP@9 = 0.7017, Coverage = 100%, Success rate = 89%, and F-measure = 0.72, which shows an edge over others. In general, our experiments show a clear significant advantage of the scheme in a web search for finding ambiguous entities