Generating Diversified Comments via Reader-Aware Topic Modeling and Saliency Detection

Automatic comment generation is a special and challenging task to verify the model ability on news content comprehension language generation. Comments not only convey salient interesting information in articles, but also imply various different reader characteristics which we treat as essential clues for diversity. However, most of approaches focus saliency extraction, while reader-aware factors implied by comments are neglected. To address this issue, propose unified topic modeling detection framework enhance quality generated comments. For modeling, design variational generative clustering algorithm latent semantic learning mining from detection, introduce Bernoulli distribution estimating select information. The obtained representations well selected incorporated into decoder generate diversified informative Experimental results three datasets show that our outperforms existing baseline methods terms both automatic metrics human evaluation. potential ethical issues discussed detail.