Utilizing covariate information in recommender systems

Entertainment and e-commerce websites produce abundant user-item interaction clicking data. This can be used for making recommendations to the users through recommender systems. Good quality recommendations are of value to both providers and users of items, helping users to find the items they are interested in.

The so-called cold-start problem concerns new items and users with no previous clicking information. Hence recommendation methods depending solely on the user/item interaction data need tools to handle this. Incorporating covariate information is hence important. This an area with groundbreaking, high-impact potential.

Methods that utilize covariate information also have potential to improve recommendations for existing items or users, for which information through clicks is scarce. Recommender systems can be model-free or model-based, with model-based meaning a statistical model that represents the stochastic process generating the data.

For our model-based BMCD method, it is natural to think of the covariates entering the model at different levels of the model. User covariates may help clustering the users, while the item covariates may improve ranking items within each cluster.

Requirements

  • Applicants must hold a MSc degree or equivalent in Statistics, Data Science, Machine Learning, Mathematics, Computer Science or a related quantitative subject with proven competence in statistics, as well as excellent computing skills.

Supervisors

Associate Professor Ida Scheel

Professor Arnoldo Frigessi

Call 1: Project start autumn 2021

This project is in call 1, starting autumn 2021. Read about how to apply

Published Sep. 9, 2020 9:45 AM - Last modified Nov. 17, 2020 3:14 PM