Using Word Embeddings To Enhance Keyword Identification For Scientific Publications

Automatic keyword identification is a desirable but difficult task. It requires considerations of not only the extraction of important words or phrases from a text, but also the generation of abstractive ones that do not appear in the text. In this paper, we propose an approach that uses word embedding vectors as an external knowledge base for both keyword extraction and generation. Our evaluation shows that our approach outperforms many baseline algorithms, and is comparable to the state-of-the-art algorithm on our chosen dataset. In addition, we also introduce a new approach for evaluating the task of keyword extraction, that overcomes a common problem of overly strict matching criteria. We show that using word embedding vectors is a simpler, yet effective, method for both keyword extraction and generation.