Papers of the day   All papers

Evaluating the Factual Consistency of Abstractive Text Summarization

Comments

Richard Socher: Summarization is one of the most important & least solved tasks in #NLProc Problem with all #DeepLearning models: they are not optimized for factual correctness We introduce a new task, dataset and model. Work by @iam_wkr @BMarcusMcCann @CaimingXiong Paper https://arxiv.org/abs/1910.12840 https://t.co/pumDBw4RbO

3 replies, 704 likes


Sebastian Gehrmann: New paper by salesforce on learning a model-based fact checker for Abstractive Summarization. Definitely a much needed evaluation approach, let's hope that these kinds of metrics will become a new standard. Link: https://arxiv.org/abs/1910.12840 #NLProc

0 replies, 46 likes


Wojciech Kryściński: New work in which we approach the problem of evaluating the factual consistency of abstractive summarization models is out! 📫📑 Work w/ @BMarcusMcCann @CaimingXiong @RichardSocher Paper: https://arxiv.org/abs/1910.12840 Key points in thread (1/6): https://t.co/4peXBya1OR

1 replies, 25 likes


Richard Socher: Keeping the facts straight! Work on Factual Consistency of Abstractive Text Summarization by @iam_wkr @BMarcusMcCann @CaimingXiong and me. Paper: https://arxiv.org/abs/1910.12840 This is really important work for an information society. Nice summary also here: https://blog.deeplearning.ai/blog/the-batch-amazons-surveillance-network-ai-that-gets-the-facts-right-deepfakes-get-regulated-predicting-volcanic-eruptions https://t.co/nuG2FSHgTN

0 replies, 9 likes


MJ: Summarization is such an important NLP task. Imagine a TLDR for everything you read! Talk about time savings and impact.

0 replies, 6 likes


Allen Schmaltz: The key point: "Such high levels of factual inconsistency render automatically generated [abstractive] summaries virtually useless in practice." http://arxiv.org/abs/1910.12840

1 replies, 3 likes


Bryan McCann: More new work with @iam_wkr! This time focusing on evaluation of factual consistency in abstractive text summarization.

0 replies, 3 likes


cathal horan: This is a good example of how to use BERT for #MachineLearning #DeepLearning #NLP classification. The authors use BERT to train a classifier to identify when text summation is factually inconsistent from the source material. https://arxiv.org/pdf/1910.12840.pdf #analytics #datascience

0 replies, 1 likes


Content

Found on Oct 29 2019 at https://arxiv.org/pdf/1910.12840.pdf

PDF content of a computer science paper: Evaluating the Factual Consistency of Abstractive Text Summarization