Multi-Document Summarization with Determinantal Point Process Attention

The ability to convey relevant and diverse information is critical in multi-document summarization yet remains elusive for neural seq-to-seq models whose outputs are often redundant fail correctly cover important details. In this work, we propose an attention mechanism which encourages greater focus on relevance diversity. Attention weights computed based (proportional) probabilities given by Determinantal Point Processes (DPPs) defined the set of content units be summarized. DPPs have been successfully used extractive summarisation, here use them select abstractive summarisation. We integrate DPP-based with various architectures ranging from CNNs LSTMs, Transformers. Experimental evaluation shows that our consistently improves delivers performance comparable state-of-the-art MultiNews dataset