Keyphrase extraction aims to find representative phrases for a document. Keyphrases are expected to cover main themes of a document. Meanwhile, keyphrases do not necessarily occur frequently in the document, which is known as the vocabulary gap between the words in a document and its keyphrases. In this paper, we propose Topical Word Trigger Model (TWTM) for keyphrase extraction. TWTM assumes the content and keyphrases of a document are talking about the same themes but written in different languages. Under the assumption, keyphrase extraction is modeled as a translation process from document content to keyphrases. Moreover, in order to better cover document themes, TWTM sets trigger probabilities to be topic-specific, and hence the trigger process can be influenced by the document themes. On one hand, TWTM uses latent topics to model document themes and takes the coverage of document themes into consideration; on the other hand, TWTM uses topic-specific word trigger to bridge the vocabulary gap between the words in document and keyphrases. Experiment results on real world dataset reveal that TWTM outperforms existing state-of-the-art methods under various evaluation metrics.
展开▼
