Beyond Semantics: Learning a Behavior Augmented Relevance Model with Self-supervised Learning

Relevance modeling aims to locate desirable items for corresponding queries, which is crucial for search engines to ensure user experience. Although most conventional approaches address this problem by assessing the semantic similarity between the query and item, pure semantic matching is not everything.