Semi-Supervised Learning Classifier for Misinformation Related to Earthquakes Prediction on Social Media

Elroy, O. ORCID: & Yosipof, A. (2023). Semi-Supervised Learning Classifier for Misinformation Related to Earthquakes Prediction on Social Media. In: Artificial Neural Networks and Machine Learning – ICANN 2023. Eds. Iliadis, L., Papaleonidas, A., Angelov, P., & Jayne, C., pp. 256-267 Cham: Springer. 10.1007/978-3-031-44207-0_22.

Full text not available from this repository.


Social media is a fertile ground for the growth and distribution of misinformation. The belief in misinformation can have devastating consequences, and may lead to unnecessary loss of life. Properly identifying and countering misinformation on social media is therefore necessary for the fight against misinformation. In this research, we developed an Adjusted Semi-Supervised Learning for Social Media (ASSLSM) method to classify and analyze tweets regarding misinformation related to earthquakes prediction. The ASSLSM method adjusts the pseudo-labeling constraints based on assumptions related to metadata of the tweets and users, with the goal of providing better information to the underlying models. We collected a dataset of 82,129 tweets related to the subject of earthquakes prediction. Expert seismologists manually labeled 4,157 tweets. We evaluated and compared the performance of ASSLSM, supervised learning, and semi-supervised learning (SSL) methods on the dataset. We found that the ASSLSM methodology provides better and more consistent performance in comparison to supervised learning and SSL. Finally, we used an ASSLSM classifier to classify the full dataset and analyzed the classified dataset.

Item Type: Book Section
Uncontrolled Keywords: Semi-Supervised Learning, Misinformation, RoBERTa, NLP, Earthquakes, Social Media
Research Programs: Advancing Systems Analysis (ASA)
Advancing Systems Analysis (ASA) > Cooperation and Transformative Governance (CAT)
Young Scientists Summer Program (YSSP)
Depositing User: Luke Kirwan
Date Deposited: 25 Sep 2023 07:58
Last Modified: 26 Sep 2023 13:18

Actions (login required)

View Item View Item