This paper addresses the gap in deep learning models for ad-hoc retrieval tasks, which have shown limited success compared to other NLP tasks. The authors argue that the ad-hoc retrieval task primarily involves relevance matching, distinct from semantic matching in NLP tasks like paraphrase identification and question answering. They propose a Deep Relevance Matching Model (DRMM) that explicitly handles three key factors of relevance matching: exact matching signals, query term importance, and diverse matching requirements. DRMM employs a joint deep architecture at the query term level, using matching histogram mapping, a feed forward matching network, and a term gating network. Experimental results on two benchmark datasets demonstrate that DRMM outperforms traditional retrieval models and state-of-the-art deep matching models. The paper also discusses the limitations of existing deep matching models and highlights the importance of handling the specific characteristics of ad-hoc retrieval tasks.This paper addresses the gap in deep learning models for ad-hoc retrieval tasks, which have shown limited success compared to other NLP tasks. The authors argue that the ad-hoc retrieval task primarily involves relevance matching, distinct from semantic matching in NLP tasks like paraphrase identification and question answering. They propose a Deep Relevance Matching Model (DRMM) that explicitly handles three key factors of relevance matching: exact matching signals, query term importance, and diverse matching requirements. DRMM employs a joint deep architecture at the query term level, using matching histogram mapping, a feed forward matching network, and a term gating network. Experimental results on two benchmark datasets demonstrate that DRMM outperforms traditional retrieval models and state-of-the-art deep matching models. The paper also discusses the limitations of existing deep matching models and highlights the importance of handling the specific characteristics of ad-hoc retrieval tasks.