A Unified Multi-task Adversarial Learning Framework for Pharmacovigilance Mining

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5234–5245 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 5234 A Unified Multi-task Adversarial Learning Framework for Pharmacovigilance Mining Shweta Yadav, Asif Ekbal, Sriparna Saha, Pushpak Bhattacharyya Indian Institute of Technology Patna Patna, India {shweta.pcs14,asif,sriparna,pb}@iitp.ac.in Abstract The mining of adverse drug reaction (ADR) has a crucial role in the pharmacovigilance. The traditional ways of identifying ADR are reliable but time-consuming, non-scalable and offer a very limited amount of ADR relevant information. With the unprecedented growth of information sources in the forms of so- cial media texts (Twitter, Blogs, Reviews etc.), biomedical literature, and Electronic Medical Records (EMR), it has become crucial to ex- tract the most pertinent ADR related informa- tion from these free-form texts. In this paper, we propose a neural network inspired multi- task learning framework that can simultane- ously extract ADRs from various sources. We adopt a novel adversarial learning-based ap- proach to learn features across multiple ADR information sources. Unlike the other exist- ing techniques, our approach is capable to ex- tracting fine-grained information (such as ‘In- dications’, ‘Symptoms’, ‘Finding’, ‘Disease’, ‘Drug’) which provide important cues in phar- macovigilance. We evaluate our proposed approach on three publicly available real- world benchmark pharmacovigilance datasets, a Twitter dataset from PSB 2016 Social Me- dia Shared Task, CADEC corpus and Medline ADR corpus. Experiments show that our uni- fied framework achieves state-of-the-art per- formance on individual tasks associated with the different benchmark datasets. This estab- lishes the fact that our proposed approach is generic, which enables it to achieve high per- formance on the diverse datasets. The source code is available here1. 1 Introduction Early detection and monitoring of adverse drug reactions (ADRs) can minimize the deleteri- ous impact on patients and health-care systems (Hakkarainen et al., 2012; Sultana et al., 2013). 1https://bit.ly/2EMln36 For prevention, the drug safety organizations known as pharmacovigilance agencies conduct post-market surveillance to identify the drug’s side effects post-release. However, the majority of the existing ADE surveillance systems utilizes passive spontaneous reporting system databases, such as the Federal Drug Administration’s Ad- verse Event Reporting System (FAERS) (Li et al., 2014). These systems are often under-reported, bi- ased and delayed. To overcome the limitation of a passive reporting system, active methods to ADR monitoring continuously explores frequently up- dated ADR data sources (Behrman et al., 2011). The quantity and near-instantaneous nature of social media provide potential opportunities for real-time monitoring of Adverse Drug Reaction (ADR). The fact that this data is up-to-date and is generated by patients overcomes the weaknesses of traditional ADR surveillance techniques (Lea- man et al., 2010). Thus, social media could com- plement traditional information sources for more effective pharmacovigilance studies, as well as po- tentially serve as an early warning system for un- known ADR, which may be important for a clin- ical decision. Additionally, the high statistically significant correlation (p < 0.001, ρ = 0.75) be- tween FAERS and ADRs (extracted through Twit- …

Content truncated to 500 words