{Mirella Lapata Laura Perez-Beltrachini}

Abstract

The ability to convey relevant and diverse information is critical in multi-documentsummarization and yet remains elusive for neural seq-to-seq models whose outputs are of-ten redundant and fail to correctly cover important details. In this work, we propose anattention mechanism which encourages greater focus onrelevanceanddiversity. Attentionweights are computed based on (proportional) probabilities given by Determinantal PointProcesses (DPPs) defined on the set of content units to be summarized. DPPs have beensuccessfully used in extractive summarisation, here we use them to select relevant anddiverse content for neural abstractive summarisation. We integrate DPP-based attentionwith various seq-to-seq architectures ranging from CNNs to LSTMs, and Transformers.Experimental evaluation shows that our attention mechanism consistently improves sum-marization and delivers performance comparable with the state-of-the-art on the MultiNewsdataset.