A Study on Automatic Keyphrase Extraction and Its Refinement for Scientific Articles

Keyphrase extraction is a fundamental, but very important task in NLP that map documents to a set of representative words/phrases. However, state-of-the-art results on benchmark datasets are still immature stage. As an effort to alleviate the gaps between human annotated keyphrases and automatically extracted ones, in this paper, we introduce our on-going work about how to extract meaningful keyphrases of scientific research articles. Moreover, we investigate several avenues of refining the extracted ones using pre-trained word embeddings and its variations. For the experiments, we use two different datasets (i.e., WWW and KDD) in computer science domain.