Checkworthiness and claim span identification from tweets

Abstract

In Natural Language Processing (NLP), identification of checkworthiness and detection of span from any news article is a crucial yet difficult topic. Automatic fake news identification is a practical NLP problem that is helpful to all online content producers, given the vast amount of Web material. The rapid expansion of social networking platforms has resulted in a tremendous increase in information accessibility while also hastening the dissemination of false information. Identification of checkworthiness from any news article is very much required task. Not only that which part of the article is containing the main claim that is detection of claim span is a very challenging and interesting task. India has no such automatic or AI system to identify fake news, so it is critical to develop these systems in Indian languages. The first challenge is to find check-worthy words among the massive amounts of data created daily from news and other sources on the internet. Only completing this step completes half of the fact-checking task. Though a lot of training data is available for English, no such dataset is available in the Indian vernacular languages. So we have created our dataset from tweeter which contains Bengali, English, Hindi & Codemix data. Using this data we have built our systems to identify the checkworthiness and to detect the claim span from tweet.