Drug-Drug Interaction Extraction

Joseph Killian Jr.

Faculty: Bruno Andriamanalimanana

Information on adverse drug reactions, and specifically, drug-drug interactions, is growing rapidly, and this research needs to be sorted through to pull out those relationships. Work has been done on this problem using mainly neural networks, and methods of deep learning, with results slowly improving year on year, but still not good enough to be used in practice.

This paper aims to see if a simpler statistical approach could attain similar or better results. The prediction algorithm used attained results of 28% accuracy, and this is similar to others who have approached this problem from this angle. The detection algorithm results of 75% accuracy are much more promising, but do not extract as much information.

The results showed that the neural approach is still superior, but this methodology has the benefit of being more transparent.

View Presentation (opens in new tab)