Machine Learning for Misinformation Containment: A Candid Assessment of the State of the Art

The invited talk delivered as part of the 23rd New Frontiers in Computing conference provides insights into the current state-of-the-art solutions for fake news detection and analyzes why they are not helping enough.

Misinformation containment was proven to be NP-hard more than a decade ago. It is undoubtedly a complex problem to solve and appropriately attracted plenty of attention from the research community.

A wide variety of machine learning algorithms such as support vector machines and logistic regression, ensemble techniques like random forest and Adaboost, deep learning frameworks such as LSTM and GAN, language models like BOW / TF-IDF and BERT, and many more have been tried out in the attempts to solve the problem.

In terms of feature engineering as well, no stone has been left unturned. Manual feature extraction, graph embeddings, and other approaches to representational learning have all been tried. Not just supervised and unsupervised learning, but various other types of learning such as few-shot learning, meta-learning, transfer learning, self-supervised learning, semi-supervised learning, reinforcement learning, and active learning have been explored extensively for the problem.

Despite the voluminous research literature purporting to solve the problem using machine learning methods, misinformation containment is largely unsolved and is growing by the day. It is therefore pertinent to understand this huge disconnect between what is claimed in the literature and the actual reality.

Speaker Dr. Vishnu S. Pendyala presented this talk virtually during the 23rd NFIC Conference sponsored by the Silicon Valley Chapters of IEEE Computer Society and North America Taiwanese Engineering & Science Association (NATEA) on August 13, 2022.

For access to past video webinars or to join our Dlist to hear about future programs, please visit https://r6.ieee.org/scv-cs