UltraRE: Enhancing RecEraser for Recommendation Unlearning via Error Decomposition
Authors: Yuyuan Li, Chaochao Chen, Yizhao Zhang, Weiming Liu, Lingjuan Lyu, Xiaolin Zheng, Dan Meng, Jun Wang
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three real-world datasets demonstrate the effectiveness of Ultra RE. |
| Researcher Affiliation | Collaboration | Yuyuan Li College of Computer Science Zhejiang University Hangzhou, China 11821022@zju.edu.cn Chaochao Chen College of Computer Science Zhejiang University Hangzhou, China zjuccc@zju.edu.cn Yizhao Zhang College of Computer Science Zhejiang University Hangzhou, China 22221337@zju.edu.cn Weiming Liu College of Computer Science Zhejiang University Hangzhou, China 21831010@zju.edu.cn Lingjuan Lyu Sony AI Japan lingjuan.lv@sony.com Xiaolin Zheng College of Computer Science Zhejiang University Hangzhou, China xlzheng@zju.edu.cn Dan Meng OPPO Research Institute Shenzhen, China mengdan90@163.com Jun Wang OPPO Research Institute Shenzhen, China junwang.lu@gmail.com |
| Pseudocode | No | The paper describes algorithms such as Optimal Balanced Clustering and Sinkhorn Algorithm using text and mathematical equations, but does not provide them in structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source codes are available at https://github.com/ZhangYizhao/UltraRE. |
| Open Datasets | Yes | We conduct experiments on the following three real-world datasets: i) Movie Lens 100k (ML-100K)2: The Movie Lens datasets are among the most extensively used in recommendation researchn [14, 16]. ML-100K contains 100 thousand ratings; ii) Movie Lens 1M (ML-1M): This is a stable version of the Movie Lens dataset, containing 1 million ratings; and iii) Amazon Digital Music (ADM)3: The Amazon dataset contains several sub-datasets according to the categories of Amazon products. ADM is the sub-dataset containing digital music reviews. 2https://grouplens.org/datasets/movielens/ 3http://jmcauley.ucsd.edu/data/amazon/ |
| Dataset Splits | Yes | Specifically, we use 80% of ratings for training, 10% as a validation set for tuning hyper-parameters, and the remainder for testing. |
| Hardware Specification | Yes | We run all experiments on the same Ubuntu 20.04 LTS System server with 48-core CPU, 256GB RAM and NVIDIA Ge Force RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions software components such as 'Adam optimizer' and 'Python optimal transport' but does not specify their version numbers or provide a comprehensive list of software dependencies with specific versions. |
| Experiment Setup | Yes | Following the original papers, we adopt normalized binary cross entropy loss and Bayesian personalized ranking loss for DMF and Light GCN respectively, and employ Adam optimizer to train the above models. We run all experiments for 10 trials and report the average results. Following [4, 7], we set the number of shards to 10 for all unlearning methods that involve division, i.e., SISA, Rec Eraser, and Ultra RE. |