Group-Specific Individualized Modeling and Recommender Systems for Large-Scale Complex Data

该研究项目旨在开发新的统计理论、方法和计算算法，以解决数据呈现独特特征的实际问题，如数据量大、动态变化速度快、来自不同个体的信息高度异质。传统的一个模型适用于所有人的范例可能没有足够的能力来检测不同亚群的重要预测因素。这项研究旨在开发适用于电子健康记录数据的替代方法，并对分配有效的个性化治疗以实现更有效的医疗服务具有价值。预计该项目将促进与来自不同领域的其他科学家的跨学科合作，这项工作也将在营销、商业和金融服务方面得到应用。正在开发的软件将被传播，以方便大规模复杂数据的应用，并将及时向行业提供，以最大限度地发挥对社会的影响。通过参与研究来培养研究生是该项目的一部分。该项目旨在开发一种新的协同过滤方法，利用来自用户和项目的聚类信息来提供更高效的推荐系统。研究还着眼于个性化变量选择的发展，同时提高了个性化变量系数的估计效率和预测能力。此外，还将发展一种混合效应估计方程方法，以减少对信息缺失数据的估计偏差。另一个研究目标是开发适用于大规模复杂数据的高效计算算法和工具。研究计划的每个组成部分都包含一系列主题，从方法学和计算发展到在现实世界问题中的应用。此外，该项目将有助于解决统计科学中的基本问题，并将激发大批科学家对推荐系统、随机效应建模、高维模型选择、分组和分组、纵向/相关数据、信息性缺失数据和点心抽样等领域的兴趣。先进的优化技术、算法和计算技术的发展对于其他类型的复杂数据问题也将是有价值的。

This research project aims to develop new statistical theory, methods, and computing algorithms to solve practical problems where the data present unique features such as large volume, large velocity of dynamic changes, and highly heterogeneous information from different individuals. The traditional one-model-fits-all paradigm may not have sufficient power to detect important predictors for heterogeneous subgroups. This research aims to develop alternative methods applicable to electronic health record data and valuable for assigning effective personalized treatments for more effective medical care. It is anticipated that the project will stimulate interdisciplinary collaborations with other scientists from disparate fields and that the work will also have applications in marketing, business, and financial services. The software under development will be disseminated to facilitate applications for large-scale complex data, and will be made available to industry in a timely manner to maximize the impact on society. Training of graduate students through involvement in the research is a part of this project. This project aims to develop a new collaborative filtering method utilizing cluster information from users and items to provide more efficient recommender systems. The research also targets the development of personalized variable selection, while improving the estimation efficiency of the personalized variable coefficients and the prediction power. In addition, a mixed-effects estimating equation approach will be developed to reduce the estimation bias for informative missing data. Another research goal is to develop efficient computational algorithms and tools applicable for large-scale complex data. Each component of the research plan contains a range of topics, from methodological and computational development to applications in real world problems. In addition, the project will help to tackle fundamental questions in statistical science and will stimulate interest from large groups of scientists in the fields of recommender systems, random effects modeling, high-dimensional model selection, subgrouping and clustering, longitudinal/correlated data, informative missing data, and refreshment sampling. The development of advanced optimization techniques, algorithms, and computational technology will be valuable for other types of complex data problems as well.