health-kg-misinfo-llm

A repository for a research project on health-related misinformation, LLM, poison attack on social media data

1. Introduction

2. Literature Review

https://github.com/penghui-yang/awesome-data-poisoning-and-backdoor-attacks

Paperpile -- last updated Jan 30, 2025.

BERT-LSTM hybrid model for Social Media Networks Digital Literacy

3. Methods

3.1 Data Source

3.1.1 COVIDLies -- Annotated 6000+ tweets

• Hossain, T.; Logan, R.L., IV; Ugarte, A.; Matsubara, Y.; Young, S.; Singh, S. COVIDLies: Detecting COVID-19 Misinformation on Social Media. In Proceedings of the Workshop on NLP for COVID-19 (Part 2) at EMNLP 2020; September 4 2020. [Paper] [Data]
• Rizvi, S. Misinformation Retrieval. Master Thesis. University of Waterloo, ON, Canada 2021. [Paper] -- This study used COVIDLies

3.1.2 VaccineLies -- Annotated tweets on covid & hpv [Paper] [Dataset]

3.2 Tools

• Chroma -- vector database solution for knowledge graph used in Alber et al.:

3.3 Pre-trained Twitter Bert Models

• twhin-bert-base
• twhin-bert-large
• TwHIN-BERT-Misinformation-Classifier

3.4 Knowledge Graphs

3.4.1 Kaggle Knowledge Graph Tutorial

• https://www.kaggle.com/code/yuantian0/knowledge-graph-nlp-tutorial-bert-spacy-nltk/edit

3.4.2 Precision Medicine Knowledge Graph (PrimeKG)

PrimeKG is an Off-the-shelf Medical KG maintained by Harvard University.

• Chandak, P.; Huang, K.; Zitnik, M. Building a Knowledge Graph to Enable Precision Medicine. Sci. Data 2023, 10, 67, doi:10.1038/s41597-023-01960-3. [Paper] [Github] [Dataset]
• • Chandak, P.; Huang, K.; Zitnik, M. Building a Knowledge Graph to Enable Precision Medicine. Sci. Data 2023, 10, 67, doi:10.1038/s41597-023-01960-3. [Paper] [Github] [Dataset] -- Used PrimeKG for data poisoning attack

3.5 Attacking Bert: Weight Poisoning Attacks on Pre-Trained Models

• Weight Poisoning Attacks on Pre-trained Models -- An interesting read with Colab code

4. Results

5. Discussion

Reference

Paperpile

• Chandak, P.; Huang, K.; Zitnik, M. Building a Knowledge Graph to Enable Precision Medicine. Sci. Data 2023, 10, 67, doi:10.1038/s41597-023-01960-3. [Paper] [Github] [Dataset]

• Yang, J.; Xu, H.; Mirzoyan, S.; Chen, T.; Liu, Z.; Liu, Z.; Ju, W.; Liu, L.; Xiao, Z.; Zhang, M.; et al. Poisoning Medical Knowledge Using Large Language Models. Nat. Mach. Intell. 2024, 6, 1156–1168, doi:10.1038/s42256-024-00899-3.

• Yang, P. Awesome-Data-Poisoning-and-Backdoor-Attacks: A Curated List of Papers & Resources Linked to Data Poisoning, Backdoor Attacks and Defenses against Them (No Longer Maintained); Github;

• Alber, D.A.; Yang, Z.; Alyakin, A.; Yang, E.; Rai, S.; Valliani, A.A.; Zhang, J.; Rosenbaum, G.R.; Amend-Thomas, A.K.; Kurland, D.B.; et al. Medical Large Language Models Are Vulnerable to Data-Poisoning Attacks. Nat. Med. 2025, 1–9, doi:10.1038/s41591-024-03445-1.

• Mozaffari-Kermani, M.; Sur-Kolay, S.; Raghunathan, A.; Jha, N.K. Systematic Poisoning Attacks on and Defenses for Machine Learning in Healthcare. IEEE J. Biomed. Health Inform. 2015, 19, 1893–1905, doi:10.1109/JBHI.2014.2344095.

• Kurita, K.; Michel, P.; Neubig, G. Weight Poisoning Attacks on Pre-Trained Models. arXiv [cs.LG] 2020.

Out-of-date Stuff