a2c [1911.03324v1] Transforming Wikipedia into Augmented Data for Query-Focused Summarization

Enable JavaScript to see more content

arXiv2code // top new 14d 1m 2m 3m // Enable JavaScript to see more content

[share]	Go to arXiv [ Microsoft Research,Harbin Institute of Technology ]Download as Jupyter Notebook	2019-11-13
[1911.03324v1] Transforming Wikipedia into Augmented Data for Query-Focused Summarization The results on DUC benchmarks show that our model with data augmentation outperforms the state-of-the-art Abstract The manual construction of a query-focused summarization corpus is costly and timeconsuming. The limited size of existing datasets renders training data-driven summarization models challenging. In this paper, we use Wikipedia to automatically collect a large query-focused summarization dataset (named as WIKIREF) of more than 280,000 examples, which can serve as a means of data augmentation. Moreover, we develop a query-focused summarization model based on BERT to extract summaries from the documents. Experimental results on three DUC benchmarks show that the model pre-trained on WIKIREF has already achieved reasonable performance. After fine-tuning on the specific datasets, the model with data augmentation outperforms the state of the art on the benchmarks. ‹ Figure 1: An example of the automatic query-focused summarization example construction. Given a statement in Wikipedia article “Marina Beach”, we take the body text of citation as the document, use the article title along with section titles to form a query (i.e., “Marina Beach, Incidents”), and the statement is the summary. (Introduction)Figure 4: ROUGE-2 score on the DUC 2007 evaluation set with various training data. The horizontal lines indicate training on DUC 2005, on DUC 2006, and on both. The green line indicates training on only a proportion of DUC 2005 and DUC 2006 examples with the WIKIREF data augmentation. The x-axis indicates the number of used training examples, along with data augmentation. (Settings)›

Related: TFIDF [1801.07704] Query Focused Abstractive Summarization: Incorporating Query Relevance, Multi-Document Coverage, and Summary Length Constraints into seq2seq Models[1604.00125] AttSum: Joint Learning of Focusing and Summarization with Neural Attention[1908.08345] Text Summarization with Pretrained Encoders[1603.07252] Neural Summarization by Extracting Sentences and Words[1907.07323] STRASS: A Light and Effective Method for Extractive Summarization Based on Sentence Embeddings[1906.01749] Multi-News: a Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model[1909.12231] Learning to Create Sentence Semantic Relation Graphs for Multi-Document Summarization[1910.00523] BillSum: A Corpus for Automatic Summarization of US Legislation[1801.10198] Generating Wikipedia by Summarizing Long Sequences[1909.04028] Countering the Effects of Lead Bias in News Summarization via Multi-Stage Training and Auxiliary Losses Related: Semantic Math [1910.04331] Agent with Warm Start and Active Termination for Plane Localization in 3D Ultrasound[1706.02434] Automatic tracking of vessel-like structures from a single starting point[1701.08644] Security Game with Non-additive Utilities and Multiple Attacker Resources[1802.03345] A Two-Stage Method for Text Line Detection in Historical Documents1903.06150v1[1903.06150] Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-grained Image Recognition[1907.05185] Edge Heuristic GAN for Non-uniform Blind Deblurring[1908.06648] Graph-Based Object Classification for Neuromorphic Vision Sensing[1803.01400] Concatenated Power Mean Word Embeddings as Universal Cross-Lingual Sentence Representations[1909.00915] Counterfactual Depth from a Single RGB Image