Scholarlink

<p>Nowadays, detecting Adverse Drug Reactions (ADRs) from texts is crucial for pharmacovigilance. The publicly available resources for this task include social media and other relevant sources. Moreover, social media data can reflect the reactions of drug users and update quickly. For this reason, the objective is to leverage artificial intelligence techniques to extract pertinent information from this platform, to identify ADRs that could avert fatalities and hospitalizations. This study investigated the automated detection of ADRs in patients by employing supervised machine learning classifiers, including LinearSVC, Complement Naive Bayes, XGBoost, and AdaBoost. These classifiers were applied to a combination of drug-related comments extracted from tweets and drug reviews in the PsyTAR (Psychiatric Treatment Adverse Reactions) dataset. Experimental results derived from the fusion dataset indicated that the evaluation of individual supervised machine learning models achieved an F1-score of 85.38%, an Accuracy of 92.10%, and an ROC–AUC of 96.98%, thereby demonstrating their effectiveness in ADRs detection. This finding could promote active (real-time) ADR surveillance and advance pharmacovigilance research. Specifically, ADRs detection plays a crucial role in understanding the safety and benefit profiles of medicines for physicians and mental healthcare professionals.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated detection of adverse drug reactions from twitter and PsyTAR using supervised machine learning models

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Brahami Menaouer,
  • Abdeldjouad Fatima Zohra,
  • Hadj Benaïchouche Meroua

摘要

Nowadays, detecting Adverse Drug Reactions (ADRs) from texts is crucial for pharmacovigilance. The publicly available resources for this task include social media and other relevant sources. Moreover, social media data can reflect the reactions of drug users and update quickly. For this reason, the objective is to leverage artificial intelligence techniques to extract pertinent information from this platform, to identify ADRs that could avert fatalities and hospitalizations. This study investigated the automated detection of ADRs in patients by employing supervised machine learning classifiers, including LinearSVC, Complement Naive Bayes, XGBoost, and AdaBoost. These classifiers were applied to a combination of drug-related comments extracted from tweets and drug reviews in the PsyTAR (Psychiatric Treatment Adverse Reactions) dataset. Experimental results derived from the fusion dataset indicated that the evaluation of individual supervised machine learning models achieved an F1-score of 85.38%, an Accuracy of 92.10%, and an ROC–AUC of 96.98%, thereby demonstrating their effectiveness in ADRs detection. This finding could promote active (real-time) ADR surveillance and advance pharmacovigilance research. Specifically, ADRs detection plays a crucial role in understanding the safety and benefit profiles of medicines for physicians and mental healthcare professionals.