A Decision-Guided Group Package Recommender Based on Multi-Criteria Optimization and Voting

Recommender systems are intended to help users make effective product and service choices, especially over the Internet. They are used in a variety of applications and have proven to be valuable for predicting the utility or relevance of a particular item and for providing personalized recommendations. State-of-the-art recommender systems focus on atomic (single) products or services and on individual users. This dissertation considers three ways of extending recommender systems: (1) to make composite (package) rather than atomic recommendations; (2) to use multiple rather than single criteria for recommendations; and, most importantly, (3) to support groups of diverse users or decision makers who might have different, even strongly conflicting, views on the weights of different criteria.;Complex group recommender systems with these features are important in such areas as public policy and budget recommendations, energy infrastructure investment, and health care plan selection by organizations. However, the problem of how to develop such systems has not been adequately addressed. Package recommendations present a unique challenge because they require the recommendation space to be very large, even infinite, and to be implicitly rather than explicitly defined. And group recommenders are considerably more complex than individual user recommenders. One reason for this complexity is the need to effectively aggregate users' preferences in a way that maximizes the group's satisfaction, fairness, and user-friendliness.;In this dissertation, I propose and develop a decision-guided group package recommender framework based on multi-criteria decision-optimization and voting. This framework operates on a very large, even infinite, recommendation space, which is implicitly defined by mathematical constraints. It is designed to provide a diverse set of optimal or near-optimal package recommendations to groups of users while taking into account the influence of individuals within the group, the dissimilarity of interest among the group's members, and the size and homogeneity of the group. The framework applies six alternative decision-making (voting) methods to refine its recommendations. Five of these come from social choice theories, namely the instant runoff voting (IRV), hybrid Condorcet-IRV, average, least misery, and average without misery methods.;In addition to them, I develop a new method, the structurally adjusted average. I also develop a technique for scaling up the group recommender system for very large, heterogeneous groups. In addition, I demonstrate how the proposed framework applies to a real problem through a case study of the Power Microgrid Operation and Investment Recommender (PMOIR). PMOIR supports (1) operational decisions on how to control each microgrid component on, say, a half-hourly basis, and (2) investment decisions on microgrid components (e.g., renewable sources of energy and power storage) over an investment time horizon. In order to implement PMOIR, I mathematically model different power components and formalize the overall optimization problem. I also implement the optimization model for PMOIR as a mixed-integer linear programming (MILP) model.;Finally, I validate the proposed framework with three experimental studies: (1) a study demonstrating that the proposed framework can produce a small set of recommendations that retain near-optimality, in terms of precision and recall, when compared with manual voting by human participants; (2) a study demonstrating the framework's ability to support very large, heterogeneous groups with only minor degradation in precision and recall; and (3) a study demonstrating the framework's feasibility, in terms of computational time, for applying PMOIR on microgrids involving 200 power components, over a five-year time horizon, with around 8 million binary variables.