Research on the Correctness Determination Method of Question-Answering Based on Multiple LLMs

The advent of Large Language Models (LLMs) has revolutionized interactive experiences in artificial intelligence, with broad applications across various sectors including education, healthcare, and public welfare. However, challenges such as data bias, overgeneralization, and an absence of emotional cognition often lead to inaccurate or irrelevant responses in practical AI question-answering scenarios. This study addresses these issues by proposing a correctness determination method for question-answering based on multiple LLMs. We aim to enhance the accuracy of AI assistants in complex or ambiguous question handling by leveraging multiple pre-trained models and employing a decision model to select the most appropriate responses. Our approach integrates crossquestioning, historical feedback, and a novel election algorithm based on historical results weighting. The study demonstrates that this method significantly improves the accuracy of questionanswering results in smart university campus assistants, increasing the overall accuracy rate from $\mathbf{6 0. 3 3 \%}$ to $\mathbf{8 0. 1 6 \%}$. The proposed solution is not only cost-effective but also reduces reliance on labor, making it accessible for individuals and small businesses. This abstract summarizes our research background, purpose, methods, and the significant improvement in AI question-answering accuracy achieved through our multi-LLM approach.