Cross-Domain Generalization Through Memorization: A Study of Nearest Neighbors in Neural Duplicate Question Detection
This work provides an incremental improvement for researchers and developers working on duplicate question detection systems, particularly in scenarios where supervised data is scarce in target domains.
This paper addresses the challenge of cross-domain generalization in duplicate question detection (DQD) by employing a nearest neighbor approach with neural representations. The method encodes question pairs from both source and target domains into a rich representation space and then uses k-nearest neighbor retrieval to aggregate labels and distances for ranking pairs. This approach demonstrates robust performance across StackExchange, Spring, and Quora datasets, outperforming cross-entropy classification in several cross-domain scenarios.
Duplicate question detection (DQD) is important to increase efficiency of community and automatic question answering systems. Unfortunately, gathering supervised data in a domain is time-consuming and expensive, and our ability to leverage annotations across domains is minimal. In this work, we leverage neural representations and study nearest neighbors for cross-domain generalization in DQD. We first encode question pairs of the source and target domain in a rich representation space and then using a k-nearest neighbour retrieval-based method, we aggregate the neighbors' labels and distances to rank pairs. We observe robust performance of this method in different cross-domain scenarios of StackExchange, Spring and Quora datasets, outperforming cross-entropy classification in multiple cases.