Improving zero-shot retrieval using dense external expansion

Pseudo-relevance feedback (PRF) is a classical technique to improve search engine retrieval effectiveness, by closing the vocabulary gap between users’ query formulations and relevant documents. While PRF typically applied on same target corpus as final retrieval, in past, external expansion techniques have sometimes been obtain high-quality pseudo-relevant set using . However, such approaches only studied for sparse (BoW) methods, its effectiveness recent dense methods remains under-investigated. Indeed, ANCE ColBERT, which conduct similarity based encoded contextualised document embeddings, are of increasing importance. Moreover, pseudo-relevance mechanisms proposed further enhance effectiveness. In particular, this work, we examine application zero-shot i.e. evaluation corpora without training. Zero-shot experiments with six datasets, including two TREC datasets four BEIR when applying MSMARCO passage collection corpus, indicate that obtaining documents ColBERT can significantly NDCG@10 (by upto 28%) 12%). addition, brings 30% improvements 29% retrieval. • Dense improves performance. High quality be retrieved expansion. Experimental results show significant ColBERT-PRF ANCE-PRF