Matrix-like visualization based on topic modeling for discovering connections between disjoint disciplines

Interdisciplinary research is challenging because of the knowledge overspecialization problem, which makes it diffi- cult for researchers to discover connections between disjoint disciplines. Closed Literature-based discovery shows the potentials of solving this problem by using information retrieval and natural language processing techniques. However, it still faces some drawbacks, such as large amounts of manual works with prior knowledge, difficulty in understanding the discoveries, and limi- tation of the extension of the domain. In this paper, we propose a matrix-like visualization approach based on topic modeling for discovering connections between disjoint disciplines. With our approach, we expect interdisciplinary connections to be efficiently discovered by detecting the topics we call mixed topics, which contain literature from disjoint disciplines. For achieving this purpose, we visualize the document-phrase matrix generated by topic modeling and develop an original biclustering algorithm to extract mixed topics in the matrix. Experiment results show that our approach can help users without prior knowledge to detect connections between disciplines.