Extraction of Medication-Effect Relations in Twitter Data with Neural Embedding and Recurrent Neural Network.

Area of Science:

• Pharmacovigilance
• Computational Linguistics
• Social Media Analytics

Background:

• Social media platforms like Twitter are increasingly utilized as a data source for pharmacovigilance.
• Existing research often identifies adverse drug events but struggles to establish the relationship between specific medications and reported effects.

Purpose of the Study:

• To develop and evaluate a classification model for extracting the relationship between medications and adverse effect expressions from Twitter data.
• To address the gap in understanding patient-generated medication-related information from social media.

Main Methods:

• The study framed the relation extraction task as a classification problem.
• Twitter text data was represented using neural embeddings.
• A recurrent neural network classifier was employed to predict the medication-effect relationship.
• Performance was evaluated against four baseline word embedding methods.

Main Results:

• The recurrent neural network model demonstrated classification performance in identifying medication-effect relationships.
• The study utilized a corpus of 9516 annotated tweets for evaluation.

Conclusions:

• Neural embeddings and recurrent neural networks offer a viable approach for relation extraction in pharmacovigilance using social media data.
• This method enhances the ability to mine patient-reported adverse drug events from platforms like Twitter.