Extracting Information from Unstructured Data

Overview

That event was held on DataFest Tbilisi 2018. There were 2 sessions: Talk and Workshop.

On Talk session Me and Ioseb were talking about the project that we had in MaxinAI. It was about extracting Merchant names from unstructured data (bank transactions). We talked about our approaches and models used. On workshop session I was showing implementation details on small dataset. Explaining approaches, models and their performance in details. The best model was CNN based Deep Learning Model. The model was built from scratch using an idea from one blog that was actually provided with the project proposal. Two interesting details were there: text encoding for CNN and Residual connections in CNN based custom architecture. You will see architecture details in source code and presentation file also.

Resources

Talk materials

Workshop materials

Updated: