Aggregation and Disaggregation of Information: A Holistic View

Ph.D. Candidate Yuyue Chen of Decision Sciences and MIS Department will be defending her Dissertation titled, “Aggregation and Disaggregation of Information: A Holistic View” on 05/25/2021. The time and location of her defense is 10:00 AM – 11:30 AM noon, via Zoom.

Many thanks to Yuyue’s dissertation committee:

• Committee Chair – Hande Y Benson – Professor of Decision Sciences and MIS – Drexel University • Committee Co-Chair: Chuanren Liu – Assistant Professor of Business Analytics & Statistics – University of Tennessee • Committee Member Jinwook Lee – Assistant Professor of Decision Sciences and MIS - Drexel University • Committee Member: Muge Capan – Associate Clinical Professor of Decision Sciences and MIS – Drexel University • Committee Member: Hugo Woerdeman – Professor of Mathematics – Drexel University

Abstract: This dissertation develops optimization solutions to tackle two challenges in data mining: intelligence aggregation and signal disaggregation. For each challenge, our solution consists of an optimization model and an efficient and robust algorithm to solve it.

We first present a model and a solution algorithm that can decompose aggregated information into smaller constituents for multiple systems. For instance, our approach simultaneously disaggregates integrated energy signals from houses into specific measurements of appliances/activities. The results provide managerial insights for energy services without costly installing additional hardware. Unlike some traditional approaches in the literature that require large training datasets, our disaggregation algorithm uses contextual features as inputs combined with a transfer learning approach to improve model accuracy and interpretation.

Next, we propose a new optimization approach to aggregating information from multiple information sources. Our model is solved with an alternated quadratic programming optimization and has been evaluated to accurately predict a stock’s future trading actions by capturing inter-correlations and quantifying the reliability scores of analysts’ recommendations. We further extend our aggregation model with additional time-related relevance features. The extension allows us to improve the prediction accuracy of stock future performance as well as to better understand the impact of time-dimension on aggregation tasks.

Although disaggregation is the inverse process of aggregation, we investigate both topics by incorporating various types of inter-correlations based on graph-theory in complex systems. The optimization models in this dissertation can be further applied to other research fields such as marketing, airline business, and operations management.