Comparing Neighbors Together Makes it Easy: Jointly Comparing Multiple Candidates for Efficient and Effective Retrieval
Abstract: A common retrieve-and-rerank paradigm involves retrieving a broad set of relevant candidates using a fast bi-encoder, followed by applying expensive but accurate cross-encoders to a limited candidate set. However, relying on this small subset is often prone to error propagation from the bi-encoders, restricting the overall performance. To address these issues, we propose the Comparing Multiple Candidates (CMC) framework, which compares a query and multiple candidate embeddings jointly through shallow self-attention layers. While providing contextualized representations, CMC is scalable enough to handle multiple comparisons simultaneously, where comparing 2K candidates takes only twice as long as comparing 100. Practitioners can use CMC as a lightweight and effective reranker to improve top-1 accuracy. Moreover, negligible extra latency through parallelism enables CMC reranking to \textit{virtually enhance} a neural retriever. Experimental results demonstrate that CMC, virtually enhancing retriever, significantly improves recall@k (+6.7, +3.5\%-p for R@16, R@64) compared to the first retrieval stage on the ZeSHEL dataset. Also, we conduct experiments for direct reranking on entity, passage, and dialogue ranking. The results indicate that CMC is not only faster (11x) than \ce\ but also often more effective, with improved prediction performance in Wikipedia entity linking (+0.7\%-p) and DSTC7 dialogue ranking (+3.3\%-p).
Paper Type: Long
Research Area: Information Retrieval and Text Mining
Research Area Keywords: Entity linking/disambiguation, Passage retrieval, Dense retrieval, Retrieval and Re-ranking
Contribution Types: Approaches to low-resource settings, Approaches low compute settings-efficiency
Languages Studied: Engish
Submission Number: 657
Loading