NLP Solutions for infodemic management
summary
InfoCoV builds a multilayer pipeline for COVID-19 infodemic monitoring on Croatian social media: collected datasets include ~200k tweets and large news corpora (many portals publish 45–60 COVID articles), and many resources are public. Models fine-tune trilingual BERT and an ELECTRA/BERTić variant with additional COVID unlabeled tuning, and training uses 10,000 annotated tweets (only 475 positive examples, severe class imbalance). Results show ~50% negative sentiment, a largest cluster (~30%) about measures/vaccines, and a retweet‑prediction prototype combining transformer features with tabular models and Optuna hyperparameter tuning.
AI-written summary · about
chapters (7)
- 0:00Start
- 4:00Information Monitoring
- 7:45Project Results
- 19:30Mask Language Modeling
- 21:38Overfitting
- 22:21Common Approaches How To Approach Overfitting
- 24:02The Retweet Classification Project
related talks
57mHow Machine Learning and AI Can Support the Fight Against COVID-19
37mUsing the power of Azure and AI to help fight the COVID pandemic
29mBioNLP: Problems of Current Solutions. Solving Current Problems
34mApplying Machine Learning Techniques to Detect and Understand Fake News
32mAnalyzing social media for suicide risk using natural language processing
37mAI/ML Strategies & Techniques for a Post Pandemic World
27mBeyond a Machine Learning Model
19mBig data in the service of reliable news