Date of Award
Department of Computer Science
Daniel N. Rockmore
A network models relationships. For a network that either encodes or supports internal information sharing activities, a better understanding of the network may enable data-driven applications (e.g., social network based recommendation), and boost both descriptive and predictive modeling of information flow in itself. In a multi-faceted manner, we propose in this thesis to contribute to several challenges that arise in the development of personalized applications in the general area of information and networks: 1) articulation of new patterns (and associated metrics) for individual user behavior and network structure; 2) exploitation of new forms of feature vector representations derived from large datasets integrating users and network structure; 3) modeling the space of information flow with network science models and in particular, the prediction of direction, outlier, and outcome for information flow; 4) improving the transparency of a network-based recommender system to enable exploration of the underlying information space. The proposed methodologies combine machine learning models, network analysis and statistical analysis, which can successfully address open problems in the field. They are validated on a range of real data and show practical significance in providing widely applicable models and displaying increased accuracy over useful baselines.
An, Chuankai, "Data-driven Personalized Applications in Networks" (2020). Dartmouth College Ph.D Dissertations. 58.