Learning from Titles to Recommend Keywords for Academic Papers.

With the increasing number of scientific papers, it is difficult for researchers to locate the most relevant and important keywords from the vast majority of papers and establish the research focus and preliminaries. Based on the commonly accepted assumption that the title of a document is always elaborated to reflect the content of a document and consequently keywords tend to be closely related to the title, a keyword ranking from paper titles involving both real-time and authoritativeness is presented in this paper. We suggest exploring paper titles as a weighted hypergraph and random walk is performed, which considers weights of both hyper-edges and hyper-vertices to model short documents social features as well as discriminative weights respectively, while measuring the centrality of words in the hyper-graph to obtain the recommended keywords. Experimental results demonstrate that the proposed approach is robust for extracting keywords from short texts.