Projection-based Relevance Model for Table Retrieval

Abstract

We propose a novel relevance model for table-retrieval using table columns ("projections") as pseudo-relevance feedback. To demonstrate the merits of our proposed approach, we re-rank tables that were initially retrieved from a Wikipedia corpus by state-of-the-art table retrieval methods. Overall, our proposed projection-based relevance model (PTRM) results in significant performance gains.