SAKE: A Graph-Based Keyphrase Extraction Method Using Self-attention

Keyphrase extraction is a text analysis technique that automatically extracts the most used and most important words and expressions from a text. It helps summarize the content of texts and recognize the main topics discussed. The majority of the existing techniques are mainly domain-specific, which require application domain knowledge and employ higher-order statistical methods. Supervised keyphrase extraction requires a large amount of labeled training data and has poor generalization ability outside the training data domain. Unsupervised systems have poor accuracy, and often do not generalize well. This paper proposes an unsupervised graph-based keyphrase extraction model that incorporates the words' self-attention score. Specifically, the proposed approach identifies the importance of each source word based on a word graph built by the self-attention layer in the Transformer and further introduces a new mechanism to capture the relationships between words in different sentences. The experimental results show that the proposed approach achieves remarkable improvements over the state-of-the-art models.