Combinatorial Reinforcement Learning with Preference Feedback

Authors: Joongkyu Lee, Min-Hwan Oh

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we empirically evaluate the performance of our algorithm, MNL-VQL, in two settings: a synthetic environment (Subsection 6.1) and a real-world dataset (Subsection 6.2). We compare our algorithm against two baselines: Myopic and LSVI-UCB (Jin et al., 2020).
Researcher Affiliation Academia 1Seoul National University, Seoul, Korea.
Pseudocode Yes Algorithm 1 MNL-VQL, MNL Preference Model with Variance-weighted Item-level Q-Learning
Open Source Code No The paper does not provide any explicit statement about releasing code or a link to a code repository.
Open Datasets Yes The Movie Lens dataset contains 25 million ratings on a 5-star scale for 62,000 movies (base items a) provided by 162,000 users (u).
Dataset Splits No The paper mentions using a subset of the Movie Lens dataset containing "1.1 × 10^3 users and a varying number of movies, N ∈ {50, 100, 200}". However, it does not specify how this data was split into training, validation, or test sets, nor does it mention any cross-validation setup.
Hardware Specification No The paper does not explicitly describe any specific hardware (e.g., GPU, CPU models, or cloud resources with specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the implementation of the algorithms.
Experiment Setup Yes We set the parameters as follows: K = 10000, H = 3, M = 4, |S| = 100 + (H-1)*4 = 400 (including the absorbing state), d = 26 (MNL feature dimension), dlin = 204 (Linear MDP feature dimension), N ∈ {50, 100, 200} (number of base items) and |A| = ∑(M-1) m=1 N choose m ∈ {20875, 166750, 1333500}. The proportion of junk items is set to 30%. For our experiments, we use a subset of the dataset containing 1.1 × 10^3 users and a varying number of movies, N ∈ {50, 100, 200}. To construct MNL features, we follow a similar experimental setup as in Li et al. (2019), employing low-rank matrix factorization. For linear MDP features, we apply the same approach as used in our synthetic data experiments.