Abstract:
© 2017 IEEE. A growing interest in the natural language processing methods applied to healthcare applications has been observed in the recent years. In particular, new drug pharmacological properties can be derived patient observations shared in social media forums. Developing approaches designed to automatically retrieve this information is of no low interest for personalized medicine and wide-scale drug tests. The full potential of the effective exploitation of both textual data and published biological data for drug research often goes untapped mostly because of the lack of tools and focused methodologies to curate and integrate the data and transform it into new, experimentally testable hypotheses. Deep learning architectures have shown promising results for a wide range of tasks. In this work, we propose to address a challenging problem by applying modern deep neural networks for disease named entity recognition. An essential step for this task is recognition of disease mentions and medical concept nor-malization, which is highly difficult with simple string matching approaches. We cast the task as an end-to-end problem, solved using two architectures based on recurrent neural networks and pre-trained word embeddings. We show that it is possible to assess the practicability of using social media data to extract representative medical concepts for pharmacovigilance or drug repurposing.
