SgSum:Transforming Multi-document Summarization into Sub-graph Selection

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4063–4074 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 4063 SgSum:Transforming Multi-document Summarization into Sub-graph Selection Moye Chen∗, Wei Li∗, Jiachen Liu, Xinyan Xiao, Hua Wu, Haifeng Wang Baidu Inc., Beijing, China {chenmoye,liwei85,liujiachen,xiaoxinyan wu_hua,wanghaifeng}@baidu.com Abstract Most of existing extractive multi-document summarization (MDS) methods score each sentence individually and extract salient sen- tences one by one to compose a summary, which have two main drawbacks: (1) neglect- ing both the intra and cross-document relations between sentences; (2) neglecting the coher- ence and conciseness of the whole summary. In this paper, we propose a novel MDS frame- work (SgSum) to formulate the MDS task as a sub-graph selection problem, in which source documents are regarded as a relation graph of sentences (e.g., similarity graph or discourse graph) and the candidate summaries are its sub- graphs. Instead of selecting salient sentences, SgSum selects a salient sub-graph from the relation graph as the summary. Comparing with traditional methods, our method has two main advantages: (1) the relations between sentences are captured by modeling both the graph structure of the whole document set and the candidate sub-graphs; (2) directly out- puts an integrate summary in the form of sub- graph which is more informative and coher- ent. Extensive experiments on MultiNews and DUC datasets show that our proposed method brings substantial improvements over several strong baselines. Human evaluation results also demonstrate that our model can produce significantly more coherent and informative summaries compared with traditional MDS methods. Moreover, the proposed architec- ture has strong transfer ability from single to multi-document input, which can reduce the resource bottleneck in MDS tasks.1 1 Introduction Currently, most extractive models treat summariza- tion as a sequence labeling task. They score and select sentences one by one (Zhong et al., 2020). ∗Equal contribution. 1Our code and results are available at: https: //github.com/PaddlePaddle/Research/tree/ master/NLP/EMNLP2021-SgSum These models (called sentence-level extractors) do not consider summary as a whole but a combina- tion of independent sentences. This may cause incoherent and redundant problem, and result in a poor summary even if the summary consists of high score sentences. Some works (Wan et al., 2015; Zhong et al., 2020) treat summary as a whole unit and try to solve the weakness of sentence- level extractors by using a summary-level extrac- tor. However, these models neglect the intra and cross-document relations between sentences which also have benefits for extracting salient sentences, detecting redundancy and generating overall coher- ent summaries. Relations become more necessary when input source documents are much longer and more complex such as multi-document input. In this paper, we propose a novel MDS frame- work called SgSum which formulates the MDS task as a sub-graph selection problem. In our frame- work, source documents are regarded as a relation graph of sentences (e.g., similarity graph or dis- course graph) and the candidate summaries are its sub-graphs. In this view, how to generate a good summary becomes how to select a proper sub- …

Content truncated to 500 words