MSc Thesis Proposal Announcement of Kamonashish Saha:"Multimodal Fake News Detection"

 

Date: Friday October 14, 2022

Time:  1:00pm-2:30pm

Passcode: If interested in attending this event, contact the Graduate Secretary at csgradinfo@uwindsor.ca with sufficient notice before the event to obtain the passcode.

 

 

In recent times, with the ease of accessibility of different social platforms, fake news has been spreading like never before. Fake news has been impacting people’s views, opinions, sentiments, etc. about several issues by spreading false information on social media platforms. Hence, nowadays the detection of fake news has gained increased importance. Although, initially most of the fake news had been concentrated in textual content and so has been its research in detecting it; however, there is a need for multimodal detection as recently, fake news has been spreading in text, image, video, and audio combined. Therefore, in this thesis, we propose the multimodal fake news detection which incorporates both textual and image content. We explore with the use of an auto-regressive language model called XLNet (an extension of the Transformer-XL) which is one of the latest models to emerge from the Natural Language Processing (NLP) field. Transformer models such Bidirectional Encoder Representations from Transformers (BERT) have been the highlights in NLP recently, however, XLNet could overcome some of the limitations of BERT and achieve more effective results. We plan to fuse this state-of-the art NLP model with a Convolutional Neural Network (CNN) model for multimodal fake news detection. Furthermore, we also explore on the Fakeddit dataset which is one of the latest and biggest datasets for multimodal fake news detection.

 

Keywords: Transformers, Natural Language Processing, XLNet, Convolutional Neural Network, Multimodal

Internal Reader: Dr. Pooya Moradian Zadeh       

External Reader: Dr. Ning Zhang             

Advisor: Dr. Ziad Kobti