Multi-task learning for abstractive text summarization with key information guide network

摘要: Neural networks based on the attentional encoder-decoder model have good capability in abstractive text summarization. However, these models are hard to be controlled process of generation, which leads a lack key information. And some information, such as time, place, and people, is indispensable for humans understand main content. In this paper, we propose information guide network summarization multi-task learning framework. The core idea automatically extract that people need most an end-to-end way use it generation process, so get more human-compliant summary. our model, document encoded into two parts: results normal encoder encoding, includes sentences keywords. A framework introduced sophisticated model. To fuse novel multi-view attention obtain dynamic representations source addition, incorporated module summary generation. We evaluate CNN/Daily Mail dataset experimental show significant improvements.