Any large organization often receives hundreds, probably thousands, of emails everyday from their customers and redirecting them to respective divisions within the company can require significant time and manpower. An automatic classification of these emails into several categories in order to redirect them to repective divisions can help save the time as well money resources.
I developed tools to clean the data with significant noise and a pipeline to generate synthetic text representative of the underrepresented classes. I also set-up the backend Postgres database for data storage.
The modelling process included architectures like Self-Attention Networks, Recurrent Neural Networks, Long-Short Term Memory, and Gated Recurrent Unit Networks.
The implementation code is propriety property of Aganitha Cognitive Solutions.