A Transformer-Based Model for Evaluation of Information Relevance in Online Social-Media: A Case Study of Covid-19 Media Posts.

Online social media has become a major source of information gathering for a huge section of society. As the amount of information flows in online social media is enormous but on the other hand, the fact-checking sources are limited. This shortfall of fact-checking gives birth to the problem of misinformation and disinformation in the case of the truthfulness of facts on online social media which can have serious effects on the wellbeing of society. This problem of misconception becomes more rapid and critical when some events like the recent outbreak of Covid-19 happen when there is no or very little information is available anywhere. In this scenario, the identification of the content available online which is mostly propagated from person to person and not by any governing authority is very needed at the hour. To solve this problem, the information available online should be verified properly before being conceived by any individual. We propose a scheme to classify the online social media posts (Tweets) with the help of the BERT (Bidirectional Encoder Representations from Transformers)-based model. Also, we compared the performance of the proposed approach with the other machine learning techniques and other State of the art techniques available. The proposed model not only classifies the tweets as relevant or irrelevant, but also creates a set of topics by which one can identify a text as relevant or irrelevant to his/her need just by just matching the keywords of the topic. To accomplish this task, after the classification of the tweets, we apply a possible topic modelling approach based on latent semantic analysis and latent Dirichlet allocation methods to identify which of the topics are mostly propagated as false information.