Ai+ Training

Natural language processing (NLP) has seen significant advances in the recent past, and is now being used for various tasks that once required a lot of human involvement. In this seminar, we will share our journey by applying NLP techniques to a robotic process automation problem.

In particular, we share our challenges automating the extraction of the Table of Contents from PDF research reports, a task which becomes challenging due to the heterogeneity of report formats, and many arbitrary constraints designed for humans to handle.

Using a combination of approaches from transfer learning with a Single Shot Detector model to rules, we highlight situations where such a hybrid approach might prove useful.

We will also cover how to break down and scope a complex set of data science requirements, and common pitfalls handling a dynamic set of requirements that were originally tailored for human operators.