Abstractive summarization aims to generate a shorter version of the documentcovering all the salient points in a compact and coherent fashion. On the otherhand, query-based summarization highlights those points that are relevant inthe context of a given query. The encode-attend-decode paradigm has achievednotable success in machine translation, extractive summarization, dialogsystems, etc. But it suffers from the drawback of generation of repeatedphrases. In this work we propose a model for the query-based summarization taskbased on the encode-attend-decode paradigm with two key additions (i) a queryattention model (in addition to document attention model) which learns to focuson different portions of the query at different time steps (instead of using astatic representation for the query) and (ii) a new diversity based attentionmodel which aims to alleviate the problem of repeating phrases in the summary.In order to enable the testing of this model we introduce a new query-basedsummarization dataset building on debatepedia. Our experiments show that withthese two additions the proposed model clearly outperforms vanillaencode-attend-decode models with a gain of 28\% (absolute) in ROUGE-L scores.
展开▼
