Abstract: [Purpose/Significance] Academic papers are the important strategic resources for the development of scientific and technological innovation. They are also the primary data that reflect the research trends of one subject, which provide the valuable methodological and innovative basis for the follow-up researchers. Recently, the knowledge organization of academic papers still lack of the fine-grained knowledge, which hinders the upgrading of scientific and technological information services to computerization and precision. [Method/Process] Firstly, this paper provides a framework of analyzing the semantic of article content: the "research topics" and "key technologies" are extracted from papers by using a semi-automatic model. Secondly, a multi-level clustering method for phrases are designed. The synonymous phrases are merged by clustering in the horizontal direction, and the hierarchical relations are built by clustering in the vertical direction. Finally, the experiments are carried out by using the massive abstracts from the core journals in the discipline of geographic information science. Based on the bibliometric analysis, we analyzed the top N of "research topics" and "key technologies", and their development trajectories over time. [Results/Conclusions] The proposed method can provide technologies and datasets for the intelligent service of the scientific and technological information.

Key words: artificial intelligence, semantic annotation, neural network, phrase clustering, bibliometric analysis