Automatic detection of research interest using topic modeling

Abstract

We demonstrated the possibility of inferring the research interest of an MIT faculty member given the title of only a few research papers written by him or her, using a topic model learned from a corpus of research paper text not necessarily related to any faculty members of MIT, and a list of topic keywords such as that of the Library of Congress. The topic model was generated using a variant of Latent Dirichlet Allocation coupled with a pointwise mutual information analysis between topic keywords and latent topics.