Abstract: 【Purpose/significance】In order to address the challenges of knowledge acquisition posed by the explosive growth of re⁃ search literature, this study proposes an Automatic Summarization method based on functional sentence recognition. This provides new ideas and validates methods for the automatic generation of research literature reviews.【Method/process】Firstly, models such as LSTM, TextCNN, and BERT are employed for functional sentence recognition training. The LSTM model, achieving an optimal F1 value of 0.90, is selected. This model is combined with the feature rule matching method to extract functional sentences of different cat⁃ egories. Secondly, candidate functional sentences are processed through scoring filtration, topic clustering, and optimization using a generative large language model. Finally, the processed content is filled into a constructed general template to generate the required Automatic Summarization of research literature.【Result/conclusion】The functional sentence recognition model and feature rule match⁃ ing method accurately extract functional sentences of different categories from research literature. The advantages of feature rules and generative large language models are transformed into improved recognition accuracy, expanded content richness, and optimized read⁃ ability. Ultimately, an Automatic Summarization on the topic of "citation analysis" in the field is generated.【Innovation/limitation】The method did not automatically identify conceptual sentences and insufficient sentences in research literature through model training. When evaluating the importance of candidate functional sentences, more precise and standardized scoring rules for semantic relation⁃ ships were not formulated.