The Application Of Machine Learning To Field Insights
In this short communication we describe an actual (successful!) application of NLP/Machine Learning to support Medical Affairs and MSL teams when analysing field insights.
Insights captured from HCP/KOL discussions are entered into our X-Fly platform by MSLs and other users. A machine system, that’s previously been trained by being exposed to thousands of insights, analyses the new insight and generates machine-generated tags (MGTs) to label the insight. This happens instantly.
The user can then review the MGTs and help teach the machine further – tell the machine to ignore certain tags from now on, merge similar tags, and add new “words” to look out for (words which the machine may not have thought important so far).
These machine-generated tags can then be used to power analytics (e.g. show tags generated by the machine for all insights from the last month), and also to support additional cool functions (e.g. instantly find other insights that match your insight, based on MGT correlation).
Every night, the machine takes into account all user-submitted learnings and new insights from the past day and reruns the entire tag generation process on all insights from scratch – so it incrementally gets better and better every single day.
We can work with many AI tools, including Stanford, IBM Watson, etc., and on this occasion we have used a Python open source toolkit and the RAKE algorithm. The advantages of this particular combination are that it’s low cost, it can be customised, your data never leaves your environment, you own all the machine learnings, and we’ve built it in a way that’s pretty fast – analyse several thousand insights in around 3-5 seconds!
There is always much hype around Machine Learning and AI being able to solve all your problems (NB they can’t!). We love to focus on real, practical applications that you can implement tomorrow.
Omabuwa Tetsola
Services & Innovations
+44 (0)20 3951 0246
