Detecting what type of knowledge constitutes a discipline, tracking how the knowledge changes, and understanding why the changes are triggered are the key issues in analyzing scientific development from a macro perspective, which is usually analyzed by the topic of evolution. However, traditional methods assume that the disciplinary structure is flat with only one-layer topics, rather than a tree-like structure with hierarchical topics, which leads to the inability of existing methods to effectively examine the details of the evolution, such as the interactions between different research directions. In this paper, we take artificial intelligence (AI) as a case in which we study its hierarchical structural evolution, more precisely inspecting disciplinary development, by analyzing 65,887 AI-related research papers published during a 10-year period from 2009 to 2018. From a hierarchical topic model that can construct a topic-tree with multi-layer organizations, we design a visual analysis model for the topic-tree to systematically and visually investigate how knowledge transfers along the topic-tree and how the topic-tree changes over time. Moreover, some assistant indicators are employed to help in the exploration of the complicated structural evolution. Then, we discover the latent relationship between the sub-structures within a topic as well as the triggering reason for the knowledge migration. Based on these results, we conclude that different topics have different development patterns and that the recent artificial intelligence revolution stems from the interactions among the different topics.
- Artificial intelligence
- Evolutionary patterns
- Hierarchical knowledge structure
- Nonnegative matrix factorization
- Topic evolution
- Visual analysis approach