A Divide-and-Conquer Approach to the Summarization of Long Documents

13 Apr 2020  ·  Alexios Gidiotis, Grigorios Tsoumakas ·

We present a novel divide-and-conquer method for the neural summarization of long documents. Our method exploits the discourse structure of the document and uses sentence similarity to split the problem into an ensemble of smaller summarization problems... In particular, we break a long document and its summary into multiple source-target pairs, which are used for training a model that learns to summarize each part of the document separately. These partial summaries are then combined in order to produce a final complete summary. With this approach we can decompose the problem of long document summarization into smaller and simpler problems, reducing computational complexity and creating more training examples, which at the same time contain less noise in the target summaries compared to the standard approach. We demonstrate that this approach paired with different summarization models, including sequence-to-sequence RNNs and Transformers, can lead to improved summarization performance. Our best models achieve results that are on par with the state-of-the-art in two two publicly available datasets of academic articles. read more

PDF Abstract

Results from the Paper


Ranked #3 on Text Summarization on Pubmed (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Text Summarization arXiv DANCER PEGASUS ROUGE-1 45.01 # 5
ROUGE-2 17.6 # 6
ROUGE-L 40.56 # 3
Text Summarization arXiv DANCER RUM ROUGE-1 42.7 # 10
ROUGE-2 16.54 # 9
ROUGE-L 38.44 # 6
Text Summarization arXiv DANCER LSTM ROUGE-1 41.87 # 13
ROUGE-2 15.92 # 10
ROUGE-L 37.61 # 7
Text Summarization Pubmed DANCER LSTM ROUGE-1 44.09 # 10
ROUGE-2 17.69 # 8
ROUGE-L 40.27 # 5
Text Summarization Pubmed DANCER RUM ROUGE-1 43.98 # 11
ROUGE-2 17.65 # 9
ROUGE-L 40.25 # 6
Text Summarization Pubmed DANCER PEGASUS ROUGE-1 46.34 # 3
ROUGE-2 19.97 # 6
ROUGE-L 42.42 # 1

Methods