Bonjour à tous,

Vous êtes tous et toutes cordialement invité.es à assister à la soutenance de thèse de Oussama Ben Sghaier.

Title: Empowering Code Review Automation: A Data-Centric, Multi-Task-Driven, and Human-Aware Approach

Date: jeudi 26 juin 9h30

Location:  AA1140 (Pavillon André-Aisenstadt)

Jury

Abstract

The integration of artificial intelligence, particularly large language models (LLMs), has transformed software engineering by enabling new levels of automation for tasks like code review—crucial yet time-consuming and complex. While traditional tools relied on rule-based static analyzers, LLMs now make it possible to generate review comments similar to those written by humans. However, fully automating this task remains challenging due to rigid analyzers, isolated modeling of interdependent code review subtasks, low-quality of the code review training data, and the used evaluation metrics that neglect human aspects.

This thesis tackles these issues through four contributions: (1) a multi-step learning framework for identifying review issues; (2) a unified architecture jointly modeling comment generation, code refinement, and quality estimation (3) a data curation pipeline using LLMs to improve comment quality; and (4) a human-centered evaluation framework that accounts for developer well-being. Together, these advances pave the way for intelligent, reliable, and human-aligned code review assistants.