École Nationale Supérieure de
Techniques Avancées

Abstract

In this internship, conducted at BNP Paribas CIB Analytics and Consulting, I focused on two main areas: a research study on topic modeling and internal consulting on verbatim analysis. The first part of the project explored methods for topic extraction, comparing traditional techniques such as Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF) with advanced methods utilizing Large Language Models (LLMs). We developed a pipeline that integrates LLMs in a zero-shot or few-shot setup for topic modeling tasks, emphasizing few-shot learning, modularity, and prompt engineering to enhance performance. The second part of the internship involved working with internal clients, analyzing feedback from departments such as Human Resources, Marketing and Customer Advocacy, and Global Banking. A dedicated tool was developed to automate the extraction and summarization of client quotes using LLMs. This tool provides insights, summarizes key trends, and evaluates the evolution of feedback over time, thus offering a powerful solution for internal consulting and analysis. The overall results demonstrate the effectiveness of LLM based methods in both research and practical applications.