Deep Learning for RegEx

Recently I decided to try my hand at the Extraction of product attribute values competition hosted on CrowdAnalytix, a website that allows companies to outsource data science problems to people with the skills to solve them. I usually work with image or video data, so this was a refreshing exercise working with text data. The challenge was to extract the Manufacturer Part Number (MPN) from provided product titles and descriptions that were of varying length – a standard RegEx problem. After a cursory look at the data, I saw that there were 54,000 training examples so I decided to give Deep Learning a chance. Here I describe my solution that landed me a 4th place position on the public leaderboard.