A deep relevance matching model (DRMM) is proposed for ad-hoc retrieval tasks. The model addresses three key factors in 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 to effectively handle these factors. The model is interaction-focused, with a hierarchical deep architecture over local interaction matrices. Experimental results on two benchmark collections show that DRMM significantly outperforms traditional retrieval models and state-of-the-art deep matching models. The model's effectiveness is demonstrated through comparisons with various baselines, including traditional retrieval models and deep matching models. The DRMM addresses the differences between semantic matching and relevance matching, leading to better performance in ad-hoc retrieval tasks. The model's components, including matching histogram mapping, feed forward matching network, and term gating network, are designed to handle the unique requirements of relevance matching in ad-hoc retrieval. The results show that DRMM achieves significant improvements in retrieval performance, particularly in terms of mean average precision (MAP), normalized discounted cumulative gain (nDCG@20), and precision at rank 20 (P@20). The model's performance is further validated through analysis of different model components and term embedding dimensions, demonstrating its effectiveness in handling long queries and diverse matching requirements in ad-hoc retrieval.A deep relevance matching model (DRMM) is proposed for ad-hoc retrieval tasks. The model addresses three key factors in 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 to effectively handle these factors. The model is interaction-focused, with a hierarchical deep architecture over local interaction matrices. Experimental results on two benchmark collections show that DRMM significantly outperforms traditional retrieval models and state-of-the-art deep matching models. The model's effectiveness is demonstrated through comparisons with various baselines, including traditional retrieval models and deep matching models. The DRMM addresses the differences between semantic matching and relevance matching, leading to better performance in ad-hoc retrieval tasks. The model's components, including matching histogram mapping, feed forward matching network, and term gating network, are designed to handle the unique requirements of relevance matching in ad-hoc retrieval. The results show that DRMM achieves significant improvements in retrieval performance, particularly in terms of mean average precision (MAP), normalized discounted cumulative gain (nDCG@20), and precision at rank 20 (P@20). The model's performance is further validated through analysis of different model components and term embedding dimensions, demonstrating its effectiveness in handling long queries and diverse matching requirements in ad-hoc retrieval.