GAN-based Collaborative Filtering Datasets

本研究利用生成对抗网络（GAN）技术，从真实数据集如Movielens、Netflix和MyAnimeList中生成参数化的协同过滤数据集。这些合成数据集可以根据需要调整用户数、物品数、样本数和数据变异性。研究中使用深度矩阵分解模型（DeepMF）提取用户和物品的密集嵌入表示，通过GAN生成合成数据，并通过K-Means聚类将密集样本转换为离散的稀疏样本。这些数据集主要用于机器学习和推荐系统领域的研究，旨在通过模拟不同参数下的数据集，测试和优化推荐算法的性能。

This study utilizes Generative Adversarial Networks (GANs) to produce parameterized collaborative filtering datasets sourced from real-world datasets including Movielens, Netflix, and MyAnimeList. These synthetic datasets support configurable parameters, such as the counts of users, items, and samples, as well as data variability, to accommodate various research requirements. In this research, the Deep Matrix Factorization (DeepMF) model is adopted to extract dense embedding representations for users and items; GANs are then used to generate synthetic data, followed by K-Means clustering that converts dense samples into discrete sparse samples. These datasets are primarily designed for research in the domains of machine learning and recommender systems, with the goal of testing and optimizing the performance of recommendation algorithms by simulating datasets under different parameter configurations.