DMRetriever: A Family of Models for Improved Text Retrieval in Disaster Management

2025-10-16

AI Summary

General-purpose text retrieval models exhibit unstable performance in disaster management due to their inability to adapt to diverse, domain-specific search intents. To address this, we propose DMRetriever—the first dedicated dense retrieval model family for disaster management—built upon a three-stage training framework: (1) unsupervised contrastive pretraining, (2) difficulty-aware progressive instruction tuning, and (3) bidirectional attention adaptation, augmented by a high-quality data refinement pipeline. DMRetriever significantly enhances generalization across six representative disaster-related search intents. On multiple benchmarks, it consistently outperforms state-of-the-art baselines: its 596M-parameter variant surpasses a model 13.3× larger, while its lightweight 33M-parameter variant achieves superior performance using only 7.6% of the parameters. This work is the first to systematically tackle the joint optimization of intent diversity and model efficiency in disaster management retrieval.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Search and Optimization: Distributed Search
• Computer Vision: Image and Video Retrieval

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Effective and efficient access to relevant information is essential for disaster management. However, no retrieval model is specialized for disaster management, and existing general-domain models fail to handle the varied search intents inherent to disaster management scenarios, resulting in inconsistent and unreliable performance. To this end, we introduce DMRetriever, the first series of dense retrieval models (33M to 7.6B) tailored for this domain. It is trained through a novel three-stage framework of bidirectional attention adaptation, unsupervised contrastive pre-training, and difficulty-aware progressive instruction fine-tuning, using high-quality data generated through an advanced data refinement pipeline. Comprehensive experiments demonstrate that DMRetriever achieves state-of-the-art (SOTA) performance across all six search intents at every model scale. Moreover, DMRetriever is highly parameter-efficient, with 596M model outperforming baselines over 13.3 X larger and 33M model exceeding baselines with only 7.6% of their parameters. All codes, data, and checkpoints are available at https://github.com/KaiYin97/DMRETRIEVER

Problem

• Specialized text retrieval models for disaster management are lacking
• General-domain models fail to handle varied disaster search intents
• Existing approaches show inconsistent performance in disaster scenarios

Innovation

• Dense retrieval models specialized for disaster management
• Three-stage training with bidirectional attention adaptation
• Difficulty-aware progressive instruction fine-tuning for efficiency