Using statistical analysis to mine topic information as input for generative models improves the interpretability of academic topic selection. According to manual preliminary directions, literature is retrieved from CNKI, processed by topic analysis, feature dimensionality reduction, topic combination mining and multi-dimensional recommendation, and then titles are generated by the model.Compared with direct generation, this method produces more specific and in-depth topics, and the statistical process improves interpretability. However, limitations exist: step-by-step selection is only at the process level, not integrated into the model; user preferences are ignored; and results lack expert verification. In general, step-by-step topic selection combining statistics and generative technology greatly improves reliability and interpretability.