Optimizing Long-term Social Welfare in Recommender Systems: A Constrained Matching Approach
Authors: Martin Mladenov, Elliot Creager, Omer Ben-Porat, Kevin Swersky, Richard Zemel, Craig Boutilier
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4. Experiments We evaluate our LP-rounding method (Sec. 3.1), dubbed LP-RS, for additive utility models (we consider both cumulative and discounted reward). ... We compare the policy πLP based on LP-RS to a myopic baseline policy4 πMy w.r.t. social welfare, max regret and provider diversity/viability, assessing both on several domains using a RS ecosystem simulator that captures provider viability dynamics. We outline the simulator, describe our datasets and trained embeddings, then discuss our findings. |
| Researcher Affiliation | Collaboration | 1Google Research 2University of Toronto 3Vector Institute 4Technion. Correspondence to: Martin Mladenov <mladenov@google.com>. |
| Pseudocode | No | The paper describes algorithms using prose and mathematical equations but does not include a distinct pseudocode block or a clearly labeled algorithm section. |
| Open Source Code | Yes | We use the RECSIM framework for simulating RS environments (Ie et al., 2019a), and the main experiments can be found in the RECSIM codebase.5 5https://github.com/google-research/recsim/blob/master/README.md#Papers |
| Open Datasets | Yes | Movie ratings We train an embedding on the Movielens dataset (Harper & Konstan, 2015)... |
| Dataset Splits | No | The paper describes the datasets used and the simulation setup, but it does not provide specific details on how the data was split into training, validation, and test sets (e.g., percentages, sample counts, or cross-validation scheme). |
| Hardware Specification | No | The paper discusses the simulation environment and datasets used but does not provide specific details about the hardware (e.g., GPU/CPU models, memory) on which the experiments were run. |
| Software Dependencies | No | The paper mentions the RECSIM framework and implicitly uses Python, but it does not specify exact version numbers for any software components, libraries, or solvers required to replicate the experiments. |
| Experiment Setup | Yes | At each epoch, the RS observes a single query per user, and must serve a slate of s providers to each user. ...all viability thresholds have the same value ν. ...User-provider reward is given by f(a, b) = ||a b||2 (with max value 0). ...setting αt = γt 1 for some γ (0, 1]. |