In Conclusion Not Repetition: Comprehensive Abstractive Summarization With Diversified Attention Based On Determinantal Point Processes

TL;DR

This paper proposes a diversified attention mechanism using Determinantal Point Processes in Seq2Seq models to enhance the abstraction and comprehensiveness of machine-generated summaries.

Contribution

It introduces the DivCNN Seq2Seq model with DPP-based attention to improve summary diversity and abstraction without altering the end-to-end architecture.

Findings

Achieves higher ROUGE scores than baseline models.

Produces more comprehensive and diverse summaries.

Maintains end-to-end training compatibility.

Abstract

Various Seq2Seq learning models designed for machine translation were applied for abstractive summarization task recently. Despite these models provide high ROUGE scores, they are limited to generate comprehensive summaries with a high level of abstraction due to its degenerated attention distribution. We introduce Diverse Convolutional Seq2Seq Model(DivCNN Seq2Seq) using Determinantal Point Processes methods(Micro DPPs and Macro DPPs) to produce attention distribution considering both quality and diversity. Without breaking the end to end architecture, DivCNN Seq2Seq achieves a higher level of comprehensiveness compared to vanilla models and strong baselines. All the reproducible codes and datasets are available online.