Learning to Suggest Breaks: Sustainable Optimization of Long-Term User Engagement

Authors: Eden Saig, Nir Rosenfeld

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we provide an empirical evaluation of our approach on semi-synthetic data.
Researcher Affiliation Academia 1Department of Computer Science, Technion Israel Institute of Technology, Haifa, Israel. Correspondence to: Eden Saig <edens@cs.technion.ac.il>, Nir Rosenfeld <nirr@cs.technion.ac.il>.
Pseudocode Yes Algorithm 1 Sample from SLV(p; u) and Algorithm 2 Adaptive policy optimization using sparse rating signals are provided in the appendix.
Open Source Code Yes Code is available at: https: //github.com/edensaig/suggest-breaks.
Open Datasets Yes The Movie Lens 1M dataset (Harper & Konstan, 2015) includes 1,000,209 ratings provided by 6,040 users and for 3,706 items... The dataset is publicly available at: https://grouplens.org/datasets/movielens/1m/.
Dataset Splits No The remaining 70% data points were used for training and testing. For these, we first randomly sampled 1,000 users to form the test set. Then, the remaining users were partitioned into the main train set S, which included 70% ( 3,528 for ML1M, 28,652 for Goodreads) of these users, and the experimental treatment sets D(j), each including 10% ( 504 for ML1M, 4,093 for Goodreads) users for N = 3. The paper defines training and test sets and experimental treatment sets but does not explicitly use the term "validation set" with its own specific split.
Hardware Specification Yes Hardware: All experiments were run on a single laptop, with 16GB of RAM, M1 Pro processor, and with no GPU support.
Software Dependencies No The paper mentions specific software packages like "SURPRISE package", "SCIKIT-LEARN", and "SCIPY.OPTIMIZE" but does not provide version numbers for these dependencies.
Experiment Setup Yes Softmax temperature was set to 0.5. We set α = 0.065, and chose γ = 0.02, δ = 0.001 (which together determine scale) so that typical values for engagement rate 1 T |Su| are on the order of 10 for the chosen T = 100.