Encoder-decoder models have achieved high performance in their application to keyphrase generation. However, keyphrases for a source text generated by these models are similar to each other because each keyphrase is independently generated. To improve the diversity, we propose a model that iteratively generates each keyphrase while considering the formerly generated keyphrase. The experimentally obtained results indicate that our model generates more diverse keyphrases with a performance that is superior or comparable to conventional models.
展开▼
