Language-Net: The Large Scale Paraphrase Dataset

The Corpus

• The Language-Net is a collection of sentence level paraphrases from Twitter by linking tweets through shared URLs. This corpus is the largest up to date with 51,524 human annotated sentence pairs: 42200 for training and 9324 for testing. It can grow 30,000 new sentential paraphrases per month with ∼70% precision. Now we have 1-year data available: 2,869,657 candidate pairs!
  
  The following paper introduces the corpus in detail:
  A Continuously Growing Dataset of Sentential Paraphrases
  Wuwei Lan, Siyu Qiu, Hua He and Wei Xu. EMNLP 2017.
  pdf BibTeX slides poster

Example Pairs

Sentence 1	Label	Sentence 2
Samsung halts production of its Galaxy Note 7 as battery problems linger.	True	#Samsung temporarily suspended production of its Galaxy #Note7 devices following reports
CO2 levels mark ‘new era’ in the world’s changing climate.	True	CO2 levels haven’t been this high for 3 to 5 million years.
The 7 biggest changes Obamacare made , and those that may disappear.	False	What a repeal of Obamacare would look like , in plain English.
Fraugster , a startup that uses AI to detect payment fraud , raises $5M.	False	AI is on the rise and in this case being applied to something worthwhile payment fraud.

Baseline Results

Publication	Model	F1
Das et al.'09	Logistic Regression: n-gram overlap features	0.683
Xu et al.'14	LEX-WMF: logistic regression + weighted matrix factorization	0.693
He et al.'16	PWIM: pairwise word interaction model	0.749
Lan et al.'18	Subword-PWIM: subword embedding based PWIM with multi-task LM	0.768

Download

Please fill in the following form to request access to the TwitterPPDB corpus and 1-year candidate pairs. It is released for non-commercial use under the CC BY-NC-SA 3.0 license. Use of the data must abide by the Twitter Terms of Service and Developer Policy. For any comments or questions, please email Wuwei Lan.

Related Resource

PIT-2015: sentence level paraphrases from Twitter based on the same trending topic. Please check this website for more info.