Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
True Impact of Cascade Length in Contextual Cascading Bandits
Authors: Hyun-jun Choi, Joongkyu Lee, Min-hwan Oh
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, comprehensive experiments validate our theoretical results and show the effectiveness of our proposed method. In this section, we empirically evaluate the performance of our proposed algorithm, UCB-CLB, and compare it against three UCB-based baselines UCB-CCA [7], CLog UCB [21], and VA-CLog UCB [21] in the contextual cascading logistic bandit setting. We conduct simulated experiments with a real-world dataset: Movie Lens 100K dataset. |
| Researcher Affiliation | Academia | Hyun-jun Choi Seoul National University EMAIL Joongkyu Lee Seoul National University EMAIL Min-hwan Oh Seoul National University EMAIL |
| Pseudocode | Yes | Algorithm 1 UCB-CLB Algorithm 2 DO-SWAP |
| Open Source Code | Yes | 5. Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We use public data and provide anonymized code and instructions for reproducibility. |
| Open Datasets | Yes | This experiment is based on the Movie Lens 100K dataset2, and we defer the experimental details to Section 6. Available at https://grouplens.org/datasets/movielens/100k/. We use the Movie Lens 100K dataset, which contains 100,000 ratings (on a 1-5 scale) from 943 users for 1,682 movies. |
| Dataset Splits | No | The paper describes data transformation (binarization of ratings) and an online evaluation setting where users are sampled uniformly at random, but does not provide specific training/test/validation dataset splits or percentages for the Movie Lens 100K dataset. |
| Hardware Specification | Yes | Computational resources. All experiments were conducted on a server equipped with an Intel Xeon Gold 6526R CPU (16 cores, 2.8GHz, 37.5MB cache, 3UPI, 195W). |
| Software Dependencies | No | The paper discusses algorithmic approaches (e.g., OMD-based estimator, ridge regression) but does not provide specific version numbers for any software, libraries, or programming languages used in the implementation of these methods. |
| Experiment Setup | Yes | Input: penalty λ, radius βt, step size η. Initialize θ1 Θ, H1 = λId. Set the step size η = 1/2 log 2 + 2 and penalty λ = max{84/2dη, K}. To evaluate the effect of cascade length, we compare cumulative regret under two settings, K = 5 and K = 10, with all other parameters held constant. We apply truncated SVD with rank r = 5 to obtain M UΣV. ...For each user-movie pair (t, i), we construct the 25-dimensional context vector xt,i = vec(ut vi), where denotes the outer product. We choose the random unknown parameter θ R25. |