How we trained our NLP model to respect the underrepresented (class)

Jueves 12

18:30h - 19:20h

Track 2

How we trained our NLP model to respect the underrepresented (class)

For a company in the service industry it is common to have thousands of daily requests for support. Answering these requests on time is crucial, especially for cases like harassment or violent behaviours. The first step to solving a ticket is to route it correctly to a specific group of agents who can solve it, and this step was being done manually in our company. Here we present a solution that uses natural language processing to automate the process of routing a ticket to the right agent by assigning them tags based on the text written by the customer in their complaint ticket. The aim of the talk is to depict our journey and share the different techniques employed to deal with problems such as little amount of tagged data, a highly imbalanced dataset and multi-objective optimization, as well as how the process of building and implementing a machine learning model took place within our company.

Big Data / Data Science
Inteligencia Artificial

Marta Timon

Data Scientist at Cabify


Marta Timon’s background includes a degree in Physics and a master in Computer Science. She has worked in a wide range of topics including optics, biomedical microtechnology, robotics and AI.

Ares Aguilar

Data Scientist at Returnly


Ares Aguilar has studied Mathematics and Computer Science at UAM (Madrid). He has worked in varied business areas (telcom, mobility, fintech) and he has always found a way to apply NLP to them.