RIT Paris is looking for a motivated candidate with solid coding skills for a 6-month research internship. This internship is an opportunity to work with a diverse and highly interdisciplinary team of scientists, exploring methods for building counterfactual, robust, and multiobjective techniques for learning to rank. The ideal candidate is passionate about forming research frameworks to solve challenging, real-world problems.

Rakuten Group

Rakuten, founded in 1997, is a Global Innovation Company based in Japan. With over 70 diverse businesses spanning e-commerce, digital content, fintech, and communications, and 32,000 employees, we serve 1.6 billion members worldwide. Our mission is to empower people and society through innovation and entrepreneurship.

Rakuten Tech in Europe

Rakuten Tech in Europe, a part of the Rakuten Group's Global Innovation Hub, serves as the regional hub for the European-based members of the Technology Division. We provide and optimize global platforms to support businesses within the Rakuten Ecosystem, tailoring them to specific use cases in Europe and beyond.

With over 130 members across seven countries and 12 offices, our presence spans France (Paris), Spain (Barcelona), the UK (Belfast and London), Estonia (Tallinn), and Germany (Berlin). Our diverse team is formed of more than 20 nationalities and collaborates with all members of the Technology Divisions on a regular basis.

Team's presentation

Rakuten Institute of Technology (RIT) is the Research and Innovation Department of Rakuten, with teams spread across the globe. RIT is a unique environment for scientific research and innovations in the domain of Human-Computer Interactions, Computer Vision, Natural Language Processing, and Machine Learning.

Context: Learning to rank (LTR) in search and recommendation systems is essential as it determines the final order of items that users will see. At this stage, it's important to monitor various metrics and meet the objectives of different stakeholders. Consequently, the final ranking system must be diverse, reflect user preferences, and maximize engagement with the platform. Additionally, ranking algorithms are prone to various sources of bias that contaminate the training data, e.g., presentation/exposure or popularity. To reduce the impact of these biases, Robust and Counterfactual learning aims to correct the learning objective using a causal model of the bias, e.g., Inverse Propensity Scoring using a Position-based Click Model to model user behavior [1]. However, the impact of these Counterfactual models still needs to be explored in a multiobjective setting.

Methods: As specified above, the project will first focus on Counterfactual learning in a multiobjective setting [4, 5]. An essential consideration is that multiobjective can appear at the source, i.e., multiple biases [2, 3], or at the target (stakeholder's metrics). The two main objectives are the following:

• Benchmark existing methods: This will require a thorough state-of-the-art review and defining the relevant metrics for evaluating robust/counterfactual and multiobjective LTR (computation time/cost, quality of the solution...) and implementing appropriate methods for the study.
• Designing new efficient methods: To account for the, devise innovative approaches to tackle based on Predict-then-optimize (a.k.a., contextual optimization) or amortized learning [6]. Finally, we will communicate the findings at an information retrieval/machine learning conference.

Position's description

As an intern, you will be at the forefront of our research, playing a key role in exploring the potential of robust/counterfactual and multiobjective optimization in item recommendation and search. Your role is not just about implementing models and evaluating results but about shaping the future of our research. We are committed to supporting you in this role, and your responsibilities will include implementing deep learning models for the recommendation/search task and evaluating the recommendation capabilities rankers under the robust/counterfactual and multiobjective settings.

Profile

- Currently enrolled in a computer Science Master 2 (or equivalent), or pursuing a PhD specializing in Machine Learning, Computer Science, statistics, or a related field.

- Applicants should have a solid mathematical background and knowledge of statistical analysis and machine learning.

- Proficiency in Python.

- Experience with Git for version control and collaboration.

- Proven interest in recommender systems and information retrieval.

- Good mathematical background and problem-solving skills.

- Full proficiency in English and the ability to work independently and collaboratively as part of a team.

A plus

- Proven experience implementing ML papers.

- Experience in PyTorch for developing machine learning models.

- Experience with deep learning for recommendation systems.

References 

Gupta et al. Safe Deployment for Counterfactual Learning to Rank with Exposure-Based Risk Minimization. SIGIR 2023. https://dl.acm.org/doi/pdf/10.1145/3539618.3591760 Buchholz et al. Counterfactual Ranking Evaluation with Flexible Click Models. SIGIR 2024. https://assets.amazon.science/5a/dc/8c93b2894c37aa1ae6c2419e3cc8/counterfactual-ranking-evaluation-with-flexible-click-models.pdf Huang et al. Going Beyond Popularity and Positivity Bias: Correcting for Multifactorial Bias in Recommender Systems. SIGIR 2024. https://arxiv.org/abs/2404.18640 Chen et al. Controllable Multiobjective Re-ranking with Policy Hypernetworks. KDD 2023. https://arxiv.org/pdf/2306.05118.pdf Wu et al. A Multiobjective Optimization Framework for Multi-stakeholder Fairness-aware Recommendation. ACM Transactions on Information Systems 2023. https://arxiv.org/pdf/2105.02951.pdf A. Hoyos-Idrobo. Learning to Re-rank with Constrained Meta-Optimal Transport. SIGIR 2023. https://arxiv.org/abs/2305.00319 Sadana et al. A Survey of Contextual Optimization Methods for Decision-Making under Uncertainty. European Journal of Operational Research 2024. https://arxiv.org/pdf/2306.10374

Benefits

• Compensation between 1100 and 1320 euros
• Restaurant vouchers
• Being part of multicultural teams with more than 20 nationalities
• Office in Sentier neighbourhood

As an employer, Rakuten Tech in Europe is committed to developing an inclusive working environment. Access to employment is open to all, regardless of gender, age, disability, ethnicity, religion, sexual orientation, or social status.

Languages:

English (Overall - 3 - Advanced)