Fig. 1. The statistics of article counts across different categories in the open-access arXiv datasetThere are 58 834 articles in nuclear theory (nucl-th), 26 586 articles in nuclear experiment (nucl-ex), 185 401 articles in high-energy phenomenology (hep-ph), and 171 695 articles in high-energy theory (hep-th).

Fig. 2. The performance comparison of Recall@k among three different retrieval methods (color online) The blue dots represent the traditional keyword-based retrieval method, the orange squares denote the semantic retrieval results using the user's original query directly, and the green pentagrams illustrate the improved semantic retrieval results achieved through query expansion questions generated by DeepSeek-r1.

Fig. 3. The performance comparison of Precision@k among three different retrieval methods (color online)The blue dots represent the traditional keyword-based retrieval method, the orange squares denote the semantic retrieval results using the user's original query directly, and the green pentagrams illustrate the improved semantic retrieval results achieved through query expansion questions generated by DeepSeek-r1.