Publications

Analytics for Real-Time Editorial Support: Lessons from The Telegraph.

Improving Group Recommendations with Sparse Autoencoders by Balancing Accuracy, Fairness, and Efficiency.

Simplicity Wins: Linear Beats Deep in Next-Basket Recommendation.

90% Slimmer Production Embeddings.

Jointly Learning Implicit and Explicit feedback.

Scalable Recommendation in Industrial Scale.

Towards Accurate Uncertainty and Test-set Retrieval Performance Estimation.

Open Dataset for Harmful and Sensitive Query Alignment.

Proposition of a multitask learning approach for the triple odd-one-out problem in cognitive sciences.

Improve recommendation of cold start items by training transformers on interactions.

Detecting and measuring popularity rates among loyal and curious audiences for online items.

Enhance recommendations by aligning them more closely with local preferences and region-specific tastes.

Enhanced matrix completion methods with new constraints, improving prediction accuracy and efficiency through theoretical analysis and practical experiments.

Proposing a cognitive modeling approach that predicts selections from item triplets while providing interpretable context and item representations.

Proposition and analysis of a methodology for assessing alignment of large language models with cognitive processes.

Proposing a method to enhance the fairness (by dealing with bias) of item exposure in recommendation lists.

Proposition of a method for improving inductive matrix factorization, achieving better accuracy, especially in noisy or incomplete data scenarios.

Development of a method for improving matrix-factorization-based recommendations by adjusting uncertainty in the feedback process by using side information.

Proposition of a technique to enhance recommendations from matrix factorization by incorporating uncertainty adjustments in the feedback mechanism through the use of graph neural networks.

Proposing a method to bridge offline-online evaluation in time-dependent and popularity contexts.