Adaptive Recommendations in Complex Mass Customization Systems

Configurators and comparators are the two major types of software used by companies to let their customers and salesforce (henceforth collectively called users) select customized products. Configurators provide an interactive environment for users to gradually customize product characteristics whereas comparators present the commonalities and differences between products to assist users in choosing between competing alternatives. Although such tools abound in industry and over the web, their development and usage are still very challenging. Customizable products with a high number of characteristics and constrained by numerous technical and business rules lead to customization process becoming complex, tedious, error-prone and cost-ineffective. As a result, users may select inappropriate or suboptimal products, or just abandon using the tool. The aforementioned problems call for new techniques to ease customization processes and guide users towards products that best suit their individual needs. This project addresses this demand by developing novel navigation-based recommendation algorithms. First we propose a modeling language for recommendation rules for complex products. We also propose to use model-driven software architecture for rapidly prototyping mass customization systems from recommendation rules. We plan on investigating the use of natural language processing techniques for reverse engineering knowledge from online consumer reviews about products. Our goal is to use that extracted knowledge to design novel interactions within product search utilities that enable users to make better decisions more easily. This thesis requires both basic research in language and algorithm design and applied research in evaluating them. The foreseen solution is likely to be deployed in the industry within 2-3 years and provide significant competitive advantage to its adopters.