Adapting the Neural Encoder-Decoder Framework from Single to Multi-Document Summarization

Abstract

Generating a text abstract from a set of documents remains a challenging task. The neural encoder-decoder framework has recently been exploited to summarize single documents, but its success can in part be attributed to the availability of large parallel data automatically acquired from the Web. In contrast, parallel data for multi-document summarization are scarce and costly to obtain. There is a pressing need to adapt an encoder-decoder model trained on single-document summarization data to work with multiple-document input. In this paper, we present an initial investigation into a novel adaptation method. It exploits the maximal marginal relevance method to select representative sentences from multi-document input, and leverages an abstractive encoder-decoder model to fuse disparate sentences to an abstractive summary. The adaptation method is robust and itself requires no training data. Our system compares favorably to state-of-the-art extractive and abstractive approaches judged by automatic metrics and human assessors.

Multi-News: A Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model

“…We use DUC 2004, as results on this dataset are reported in Lebanoff et al (2018), although this dataset is not the focus of this work. For results on DUC 2004, models were trained on the CNNDM dataset, as in Lebanoff et al (2018). PG-BRNN and CopyTransformer models, which were pretrained by OpenNMT on CNNDM, were applied to DUC without additional training, analogous to PG-Original.…”

Abstractive Multi-Document Summarization Based on Semantic Link Network

“…Attempts to realize abstractive MDS under the framework have been made, e.g., generating English Wikipedia through multi-document summarization [49]. Several works apply pre-trained abstractive summarization model of single document summarization to multi-document summarization task [50][51] [52] to overcome the lack of enough train examples for MDS tasks. Unsupervised neural abstractive MDS is to leverage large nonannotated corpus [53].…”

Teaching Natural Language Processing through Big Data Text Summarization with Problem-Based Learning

“…Table 8 lists the various approaches developed by the different event teams and their corresponding ROUGE scores. In addition to the techniques mentioned in Section 4, we noticed that some teams applied other methods during implementation, including a rule-based classifier (RBF) or a multilayer perceptron (MLP) classifier for relevance judgment, and the pointer-generator with maximal marginal relevance (PG-MMR) developed by Lebanoff, Song, & Liu (2018) for summarization.…”