2021/09/13 by Shivam Raval, Raval, Shivam, Hooman Sedghamiz +10
Computer Science · Medicine · Social Sciences · #Computation and Language (cs.CL) #Data-Driven Disease Surveillance #FOS: Computer and information sciences #Misinformation and Its Impacts #Sentiment Analysis and Opinion Mining #Topic Modeling #Vaccine Coverage and Hesitancy
paper · pdf · doi:10.48550/arxiv.2109.05815
openalex publication_date 2021/09/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Adverse Events (AE) are harmful events resulting from the use of medical\nproducts. Although social media may be crucial for early AE detection, the\nsheer scale of this data makes it logistically intractable to analyze using\nhuman agents, with NLP representing the only low-cost and scalable alternative.\nIn this paper, we frame AE Detection and Extraction as a sequence-to-sequence\nproblem using the T5 model architecture and achieve strong performance\nimprovements over competitive baselines on several English benchmarks (F1 =\n0.71, 12.7% relative improvement for AE Detection; Strict F1 = 0.713, 12.4%\nrelative improvement for AE Extraction). Motivated by the strong commonalities\nbetween AE-related tasks, the class imbalance in AE benchmarks and the\nlinguistic and structural variety typical of social media posts, we propose a\nnew strategy for multi-task training that accounts, at the same time, for task\nand dataset characteristics. Our multi-task approach increases model\nrobustness, leading to further performance gains. Finally, our framework shows\nsome language transfer capabilities, obtaining higher performance than\nMultilingual BERT in zero-shot learning on French data.\n