Search and Recommendation research - NAVER LABS Europe


Human-centric, multimodal strategies and modules built on accountable algorithms for transparency and trust

Search and Recommendation illustrating image

The ubiquity of sensors and the apparition of new user interfaces has profoundly changed the nature of search and recommendation engines and how we interact with them.

Our research is carried out in collaboration with the NAVER search group, responsible for the world's 5th biggest search engine.
This provides research opportunities that go far beyond the traditional information retrieval and recommendation frameworks.

We have become more reluctant to express our needs or preferences in an explicit manner as previously done through primarily text-based interfaces. The way we give feedback to search and recommendation (S&R) systems is noisier, more biased and more heterogeneous. Users, service providers and regulators have greater requirements whereby they want more accountable, transparent and ethical technologies that respect the privacy-utility trade-off, whilst also offering some kind of guarantee on the fairness of the results they provide.

This change brings new exciting research opportunities that our research group enjoys initiating and experimenting with in the real world.

We firmly believe that search and recommendation processes must be considered as multi-stakeholder, multi-round games (continuous learning loops), where the engine and the user interplay collaboratively with the different types of feedback they each provide. Facilitating communication between the system and the user naturally leads to the design of multimodal strategies where text in any language, images, sound and speech can be consistently handled together. With the same communication objective in mind, and in particular to establish mutual confidence, we focus on developing unbiased ranking algorithms that can explain the advice they provide in a transparent and understandable form, while preserving fairness and diversity. The technology we use to pursue our research relies on Deep-IR/Deep-matching architectures, context-aware sequential recommendation models, causal and counter-factual learning-to-rank algorithms, conversational S&R and reinforcement learning.

Recent Publications

Blog Colbert Image
By dissecting the matching process of the recent ColBERT model, we make a step towards unveiling the ranking properties of BERT-based ranking models and show that ColBERT (implicitly) learns a notion of term importance that correlates with IDF. Blog article by Thibault Formal and Stéphane Clinchant
web image search blog image
A new approach to image search uses images returned by traditional search methods as nodes in a graph neural network through which similarity signals are propagated, achieving improved ranking in cross-modal retrieval. Blog article by Thibault Formal, Stéphane Clinchant and Jean-Michel Renders
fair_ranking_blog image
A new greedy, brute-force solution improves fairness in ranking and encompasses realistic scenarios with multiple, unknown protected groups. Blog article by Thibaut Thonet and Jean-Michel Renders

Search and Recommendation team:

Thibault Formal
PhD candidate
Fabien Guillot
Till Kletti
PhD candidate
Carlos Lassance
Jean-Michel Renders
Group lead
Thibaut Thonet