Natural Language Contents Evaluation System for Detecting Fake News using Deep Learning

This Recently, a lot of information is spreading rapidly on SNS. Inaccurate communication of news media includes fears about unreliable sources and fake news that lacks confirmation of facts. Fake news is spread through SNS, causing social confusion and further economic loss. The purpose of the news is accurate information transmission. In this regard, it is very important to judge the discrepancies in the contents of the text and the distorted reports. We try to solve the problem of judging whether the sentence to be verified is correct after collecting the facts. This paper defines the problem of extracting the related sentences from the input sentence in Fact Data Corpus which is assumed to be fact and judging whether the extracted sentence and the input sentence are true or false. In the various NLP tasks, we create a Korean-specific pre-Training model using state-of-The-Art BERT. Using this model, fine-Tuning is performed to match the data set detected by Korean fake news. The AUROC score of 83.8% is derived from the test set generated using the fine-Tuned model.