| 本文已被：浏览 4683次 下载 10442次 |
码上扫一扫！ |
|
|
| 因子分解机模型研究综述 |
|
赵衎衎1,2, 张良富1,2, 张静1,2, 李翠平1,2, 陈红1,2
|
|
1.中国人民大学 信息学院, 北京 100872;2.数据工程与知识工程教育部重点实验室(中国人民大学), 北京 100872
|
|
| 摘要: |
| 传统矩阵分解方法因其算法的高可扩展性和较好的性能等特点，在预测、推荐等领域有着广泛的应用.然而大数据环境下，更多上下文因素的获取变得可能，传统矩阵分解方法缺乏对上下文信息的有效利用.在此背景下，因子分解机模型提出并流行.为了更好地把握因子分解机模型的发展脉络，促进因子分解机模型与应用相结合，针对因子分解机模型及其算法进行了综述.首先，对因子分解机模型的提出进行了溯源，介绍了从传统矩阵分解到因子分解机模型的演化过程；其次，从模型准确率和效率两方面对因子分解机模型存在的基本问题和近年来的研究进展进行了总结，然后综述了适用于因子分解机模型求解的4种代表性优化算法；最后分析了因子分解机模型目前仍存在的问题，提出了可能的解决思路，并对未来的研究方向进行了展望. |
| 关键词: 因子分解机 高阶交互 特征选择 概率模型 凸优化 分布式框架 优化方法 |
| DOI：10.13328/j.cnki.jos.005698 |
| 分类号: |
| 基金项目:国家自然科学基金（61772537，61772536，61702522，61532021） |
|
| Survey on Factorization Machines Model |
|
ZHAO Kan-Kan1,2, ZHANG Liang-Fu1,2, ZHANG Jing1,2, LI Cui-Ping1,2, CHEN Hong1,2
|
|
1.School of Information, Renmin University of China, Beijing 100872, China;2.Key Laboratory of Data Engineering and Knowledge Engineering of Ministry of Education(Renmin University of China), Beijing 100872, China
|
| Abstract: |
| The traditional matrix factorization method has a wide range of applications in prediction and recommendation tasks because of its high scalability and good performance. In the big data era, more and more contextual features can be obtained easily, while the traditional matrix factorization approach lacks effective use of context information. In this context, Factorization Machines (FM) is proposed and popular. To better grasp the development process of FM model and adapt FM approach to the real application, this paper reviews existing FM models and their optimization algorithms. First, it introduces the evolution process from traditional Matrix Factorization (MF) to FM model. Second, the paper summarizes the existing researches on FM method from the perspective of model accuracy and efficiency; Third, the paper presents the studies of four representative optimization algorithms, which are suitable for various FM models. Finally, the paper analyzes the challenges in the current FM model, proposes possible solutions for these problems, and discusses the future work. |
| Key words: factorization machine high-order interaction feature selection probability model convex optimization distributed framework optimization algorithm |
