On the Role of Long-tail Knowledge in Retrieval Augmented Large Language Models
Dongyang Li, Junbing Yan, Taolin Zhang, Chengyu Wang, Xiaofeng He, Longtao Huang, Hui Xue, Jun Huang
TL;DR
This paper argues that long-tail knowledge is essential for effective retrieval-augmented generation and introduces Generative Expected Calibration Error (GECE) to detect long-tail instances using semantics, word frequency, and gradient signals. By selectively retrieving documents only for long-tail queries, the proposed method achieves substantial inference speedups (around 4x) and maintains or improves performance across QA benchmarks like NQ and TriviaQA as well as MMLU. The GECE-guided pipeline demonstrates that targeted augmentation can reduce noise from common queries while patching knowledge gaps, offering a practical, scalable improvement to RAG systems. Overall, the work presents a principled, calibration-based approach to integrate long-tail knowledge into LLM retrieval, with demonstrated efficiency and accuracy gains and clear directions for further validation and extension.
Abstract
Retrieval augmented generation (RAG) exhibits outstanding performance in promoting the knowledge capabilities of large language models (LLMs) with retrieved documents related to user queries. However, RAG only focuses on improving the response quality of LLMs via enhancing queries indiscriminately with retrieved information, paying little attention to what type of knowledge LLMs really need to answer original queries more accurately. In this paper, we suggest that long-tail knowledge is crucial for RAG as LLMs have already remembered common world knowledge during large-scale pre-training. Based on our observation, we propose a simple but effective long-tail knowledge detection method for LLMs. Specifically, the novel Generative Expected Calibration Error (GECE) metric is derived to measure the ``long-tailness'' of knowledge based on both statistics and semantics. Hence, we retrieve relevant documents and infuse them into the model for patching knowledge loopholes only when the input query relates to long-tail knowledge. Experiments show that, compared to existing RAG pipelines, our method achieves over 4x speedup in average inference time and consistent performance improvement in downstream tasks.