Robust meta learning for risk-aware recommender systems

推荐系统是许多在线服务的核心，但它们极易受到先令攻击、隐私泄露和意外更改等风险的影响。该项目旨在开发新的基于贝叶斯的、隐私保护的、自适应的模糊Meta学习方法和Meta推荐系统，这些方法和系统对这些风险、不确定和动态的环境具有鲁棒性。预期的结果应该显著提高推荐系统的可靠性，特别有利于在线个性化服务系统，例如，电子政务、电子商务和电子学习。这些成果还将通过一个新的强大的Meta学习模式来推进机器学习知识，用于一般数据分析和应用。

Recommender systems are the core of many online services but they are highly vulnerable to risks like shilling attacks, privacy leaks, and unexpected change. This project aims to develop new adversarial Bayesian-based, privacy-preserved and self-adaptive fuzzy meta learning methods and meta recommender systems that are robust to these risky, uncertain and dynamic environments. The anticipated outcomes should significantly improve the reliability of recommender systems with particular benefits for online personalised service systems, e.g., e-government, e-business and e-Learning. The outcomes will also advance machine learning knowledge with a new robust meta learning schema for general data analytics and applications.