What is it about?
Reinforcement Learning is a powerful technique for decision making. Moreover, while properly optimizing objective, unexpected strategies can be found. We observe cheating strategies in popular recommendation system dataset MovieLens 1M. Resulted recommendations are precise, but unnecessary in the production (the same popular item small subset of possible items is recommended for all users). The new noise injection strategy was proposed for DDPG algorithm. After applying such strategy agent managed to find both precise and diverse recommendations.
Featured Image
Photo by John Schnobrich on Unsplash
Why is it important?
There are many examples of using Deep Deterministic Policy Gradient (DDPG) for recommendations. Nevertheless, for exploration they use different techniques compared to Ornstein-Uhlenbeck noise from non-recsys environments. This particular article successfully incorporate OU noise injection strategy for improving relevance-diversity tradeoff.
Perspectives
Article is related to both multiobjective evaluation and reinforcement learning for recommendations.
Alexey Grishanov
Moscow Institute of Physics and Technology
Read the Original
This page is a summary of: Multiobjective Evaluation of Reinforcement Learning Based Recommender Systems, September 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3523227.3551485.
You can read the full text:
Contributors
The following have contributed to this page
