Search Results (2)

Despite the remarkable progress of large language models (LLMs) in understanding and generating unstructured text, their application in structured data domains and their multi-role capabilities remain underexplored. In particular, utilizing LLMs to perform complex reasoning tasks on knowledge graphs (KGs) is still an emerging area with limited research. To address this gap, we propose KG-EGV, a versatile framework leveraging LLMs to perform KG-based tasks. KG-EGV consists of four core steps: sentence segmentation, graph retrieval, EGV, and backward updating, each designed to segment sentences, retrieve relevant KG components, and derive logical conclusions. EGV, a novel integrated framework for LLM inference, enables comprehensive reasoning beyond retrieval by synthesizing diverse evidence, which is often unattainable via retrieval alone due to noise or hallucinations. The framework incorporates six key stages: generation expansion, expansion evaluation, document re-ranking, re-ranking evaluation, answer generation, and answer verification. Within this framework, LLMs take on various roles, such as generator, re-ranker, evaluator, and verifier, collaboratively enhancing answer precision and logical coherence. By combining the strengths of retrieval-based and generation-based evidence, KG-EGV achieves greater flexibility and accuracy in evidence gathering and answer formulation. Extensive experiments on widely used benchmarks, including FactKG, MetaQA, NQ, WebQ, and TriviaQA, demonstrate that KG-EGV achieves state-of-the-art performance in answer accuracy and evidence quality, showcasing its potential to advance QA research and applications. Full article

Open-domain question-answering systems need models capable of referencing multiple passages simultaneously to generate accurate answers. The Rational Fusion-in-Decoder (RFiD) model focuses on differentiating between causal relationships and spurious features by utilizing the encoders of the Fusion-in-Decoder model. However, RFiD reliance on partial token information limits its ability to determine whether the corresponding passage is a rationale for the question, potentially leading to inappropriate answers. To address this issue, we propose a Quantum-Inspired Fusion-in-Decoder (QFiD) model. Our approach introduces a Quantum Fusion Module (QFM) that maps single-dimensional into multi-dimensional hidden states, enabling the model to capture more comprehensive token information. Then, the classical mixture method from quantum information theory is used to fuse all information. Based on the fused information, the model can accurately predict the relationship between the question and passage. Experimental results on two prominent ODQA datasets, Natural Questions and TriviaQA, demonstrate that QFiD outperforms the strong baselines in automatic evaluations. Full article