Conversational LLMs for school & career guidance: language, trust, and social justice in educational AI systems

About the Project

This PhD position is part of the ENACT excellence axis on Natural Language Processing and

Multimodal LLMs. The project investigates conversational guidance systems powered by large

language models (LLMs) used by high-school students for educational and vocational

orientation. It sits at the crossroads of computational linguistics, AI research, and the sociology

of education.

The research addresses a central societal challenge: Can LLM-based chatbots support students

in school and career decisions without reproducing, and ideally helping to reduce, social,

territorial, and gender inequalities ?

Research Objectives

The PhD is structured around three scientific axes:

1. Linguistic modelling ; Build and annotate a French guidance dialogue corpus; develop

NLP/LLM models to characterise discourse registers (informational, motivational,

reflective), detect engagement markers, and model interactive dynamics.

2. Trust & acceptability ; Design a multidimensional framework combining linguistic,

behavioural, and psychosocial indicators to model the conditions under which students

build or lose trust in conversational agents.

3. Algorithmic fairness & bias mitigation ; Define fairness metrics tailored to

educational guidance, run contrastive analyses between LLM-generated

recommendations and human counsellors (PsyEN), and test mitigation strategies (DPO,

output filtering, sensitive-attribute constraints).

Expected Outcomes

• New methods and tools for detecting and reducing algorithmic biases in educational

LLMs

• Design principles for transparent, explainable, and well-calibrated guidance chatbots.

• Operational equity dashboards for rectorates, schools, and guidance professionals.

• High-impact publications in NLP, AI, and educational technology venues.

Laboratoire LORIA

Le Loria, Laboratoire lorrain de Recherche en Informatique et ses Applications est une Unité Mixte de Recherche (UMR 7503), commune à plusieurs établissements : le CNRS, l’Université de Lorraine, CentraleSupélec et Inria.

Depuis sa création en 1997, le Loria a pour mission la recherche fondamentale et appliquée en sciences informatiques.

Le laboratoire fait partie du pôle scientifique AM2I (Automatique, Mathématiques, Informatique et leurs interactions) de l’Université de Lorraine.

Nos travaux scientifiques sont menés au sein de 28 équipes structurées en 5 départements, dont 14 sont communes avec Inria, représentant un total de plus de 500 personnes. Le Loria est un des plus grands laboratoires de la région Grand Est.

Required:

Master's degree (or equivalent) in Computer Science, Artificial Intelligence, Computational Linguistics, or a related field

Strong background in Natural Language Processing and/or machine learning

Programming proficiency in Python (transformers, PyTorch/HuggingFace ecosystem)

Ability to work in both French and English (corpus data is in French; publications in English)

Desirable:

Experience with LLM fine-tuning, RLHF, or Direct Preference Optimization (DPO)

Familiarity with algorithmic fairness methods and explainability techniques (XAI)

Interest in educational technology, ethics of AI, or social sciences

Previous experience with corpus annotation or conversational data