Graph-based Keyphrase Extraction Using Word and Document Em beddings*

With the increasing amount of text data in applications, the task of keyphrase extraction receives more attention that aims to extract concise and important information from a document. In this paper, we propose a novel graph-based keyphrase extraction method using word and document embedding vectors. Two graph construction schemes named GKE-w and GKE-p are designed in which candidate words and phrases are represented as nodes respectively. By calculating the similarity between a word/phrase and the document, each node is assigned an initial weight that reflects the preference to be a keyphrase. Then, we calculate the score of each candidate word/phrase using a semantic biased random walk strategy. Finally, the Top N scored candidate phrases are selected as the final keyphrases. Experiments on two widely used datasets show that the proposed keyphrase extraction algorithm outperforms the state-of-the-art keyphrase extraction methods in terms of precision, recall, and F1 measures.