Supply-Side Equilibria in Recommender Systems
Authors: Meena Jagadeesan, Nikhil Garg, Jacob Steinhardt
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide an empirical analysis of supply-side equilibria using the Movie Lens-100K dataset and recommendations based on nonnegative matrix factorization (NMF). In particular, we compute the single-genre equilibrium direction for different cost functions as well as an upper bound on β (i.e., the threshold where specialization starts to occur) for different values of the dimension D. These experiments provide qualitative insights that offer additional intuition for our theoretical results. |
| Researcher Affiliation | Academia | Meena Jagadeesan UC Berkeley mjagadeesan@berkeley.edu Nikhil Garg Cornell Tech ngarg@cornell.edu Jacob Steinhardt UC Berkeley jsteinhardt@berkeley.edu |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | See Appendix A and https://github.com/mjagadeesan/supply-side-equilibria for details. |
| Open Datasets | Yes | We provide an empirical analysis of supply-side equilibria using the Movie Lens-100K dataset... The Movie Lens 100K dataset consists of 943 users, 1682 movies, and 100,000 ratings [Harper and Konstan, 2015]. |
| Dataset Splits | No | The paper uses the Movie Lens-100K dataset but does not specify explicit training, validation, or test dataset splits (e.g., percentages or sample counts) for the empirical analysis. It mentions obtaining D-dimensional user embeddings by running NMF, but not data partitioning for model training/evaluation in the typical sense. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions using the 'scikit-surprise library' and the 'cvxpy library' but does not provide specific version numbers for these software dependencies, which is required for reproducibility. |
| Experiment Setup | No | The paper mentions parameters like 'D factors' for NMF and the use of 'projected gradient descent' or 'cvxpy library' for optimization. However, it does not provide specific experimental setup details such as concrete hyperparameter values (e.g., learning rates, number of iterations for gradient descent, convergence criteria) or system-level training settings. |