Generative Adversarial User Model for Reinforcement Learning Based Recommendation System
Authors: Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, Le Song
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct three sets of experiments to evaluate our generative adversarial user model (called GAN user model) and the resulting RL recommendation policy. Our experiments are designed to investigate the following questions: (1) Can GAN user model lead to better user behavior prediction? (2) Can GAN user model lead to higher user reward and click rate? and (3) Can GAN user model help reduce the sample complexity of reinforcement learning? Dataset and Feature Description. We use 6 real-world datasets: (1) Movie Lens... (2) Last.fm... (3) Yelp... (4) Taobao... (5) Rec Sys15 Yoo Choose... (6) Ant Financial News dataset... The results in Table 1 show that GAN model performs significantly better than baselines. |
| Researcher Affiliation | Collaboration | 1School of Mathematics, 2School of Industrial and Systems Engineering, 3School of Computational Science and Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA, 4Ant Financial, Hangzhou, China. |
| Pseudocode | Yes | Thus, we can obtain an optimal action in O (k|I|) computations by applying these functions in a cascading manner. See Algorithm 1 and Figure 2(c) for a summary. The overall cascading Q-learning algorithm is summarized in Algorithm 2 in Appendix B, where we employ the cascading Q functions to search the optimal action efficiently. |
| Open Source Code | No | The paper does not contain any explicit statement or link indicating that the source code for the methodology is open-source or publicly available. |
| Open Datasets | Yes | We use 6 real-world datasets: (1) Movie Lens contains a large number of movie ratings, from which we randomly sample 1,000 active users... (2) Last.fm contains listening records from 359,347 users... (3) Yelp contains users reviews to various businesses... (4) Taobao contains the clicking and buying records of users in 22 days... (5) Rec Sys15 Yoo Choose contains click-streams... (6) Ant Financial News dataset contains clicks records from 50,000 users for one month... |
| Dataset Splits | Yes | Users are randomly divided into train(50%), validation(12.5%) and test(37.5%) sets. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions various models and algorithms but does not specify any software dependencies with version numbers, such as programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | No | While the paper describes some experimental settings, such as evaluating policies repeatedly for 50 times and setting η = 1, it does not provide specific hyperparameter values like learning rates, batch sizes, number of epochs, or optimizer configurations, which are crucial for reproducibility. |