Running summarizing tools on a new article

Abstract: Large language models have achieved great success in various natural language processing tasks, yet they still face several serious challenges, such as errors and hallucinations, and high inference cost. The contribution of this work is an extension framework called Refined Query Retrieval Augmented Generation (RQ-RAG) to enhance query handling through rewriting, decomposition, and disambiguation. The algorithm works with a specifically chosen dataset of some 42,810 samples, and search results from DuckDuckGo are used to provide answers that seemingly make sense in context. We are interested in exploiting LLaMA-4 Maverick, a state-of-the-art model with 17B parameters, as their base. We demonstrate on RQ-RAG dataset how the query refinement technique impacts upon facts' accuracy. On the single-hop (ARC-C, OBQA, POPQA) and multi-hop (HotpotQA, 2Wiki, MuSiQue) experimental results demonstrate that our method outperforms the baseline models and previous state-of-the-art retrievalaugmented approaches in terms of Accuracy and Match Score as well as F1 Score. The results show that query refinement can increase the reliability, robustness, and efficiency of LLM enabled systems.