Abstract: Accurate identification of the structure function modules from the ancient bibliographic synopses is important for searching ancient books, studying knowledge, identifying editions and examining the loss of ancient books. This paper proposes a fine-grained six-category synopses structure function classification strategy and construct a synopses structure function recognition model using the RoBERTa to meet the current practical needs of grasping the structure function connotation of ancient bibliographic synopses, from the perspective of intelligent synopses structure function recognition. The experiments on the synopses of ancient scientific and technical books in the Siku Quanshu Zongmu show that the proposed classification strategy and recognition model has certain domain adaptability and effectiveness. By combining pre-training models and automatic learning of semantic connotation features of ancient bibliographic synopses, we can effectively perform synopses structure function recognition. It effectively reduces the threshold of synopses utilization, improves the speed of synopses content presentation, and facilitates readers’ accurate search, rapid understanding and in-depth utilization of ancient books.