Beyond Prompting: An Efficient Embedding Framework for Open-Domain Question Answering

2025-03-03

AI Summary

To address the high computational overhead, inference instability, and insufficient retrieval coverage in open-domain question answering (ODQA) caused by multi-turn prompt dependency in the retrieve-then-read paradigm, this paper proposes EmbQA—a unified embedding-level framework. Methodologically, EmbQA introduces three key innovations: (1) a lightweight contrastive learning–driven query embedding fine-tuning mechanism that operates without labeled supervision; (2) exploratory embedding expansion to enrich semantic coverage; and (3) an automatic filtering strategy based on answer confidence entropy. Extensive experiments across three open-source large language models, three mainstream retrievers, and four standard ODQA benchmarks demonstrate that EmbQA consistently achieves superior accuracy and inference efficiency, outperforming recent state-of-the-art baselines across all settings.

Technology Category

• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Intelligent Query Processing
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Large language models have recently pushed open domain question answering (ODQA) to new frontiers. However, prevailing retriever-reader pipelines often depend on multiple rounds of prompt level instructions, leading to high computational overhead, instability, and suboptimal retrieval coverage. In this paper, we propose EmbQA, an embedding-level framework that alleviates these shortcomings by enhancing both the retriever and the reader. Specifically, we refine query representations via lightweight linear layers under an unsupervised contrastive learning objective, thereby reordering retrieved passages to highlight those most likely to contain correct answers. Additionally, we introduce an exploratory embedding that broadens the model's latent semantic space to diversify candidate generation and employs an entropy-based selection mechanism to choose the most confident answer automatically. Extensive experiments across three open-source LLMs, three retrieval methods, and four ODQA benchmarks demonstrate that EmbQA substantially outperforms recent baselines in both accuracy and efficiency.

Problem

• Reduces computational overhead in open-domain question answering
• Improves retrieval coverage and stability in ODQA pipelines
• Enhances answer accuracy and efficiency using embedding-level framework

Innovation

• Enhances retriever and reader via embedding-level framework
• Refines query representations using unsupervised contrastive learning
• Introduces exploratory embedding with entropy-based answer selection