Department of Computer Science & Engineering

Trade-offs have been used in learning feature dependent preferences in recommender systems literature. We use trade-offs not only to model user preferences but also to characterize products in the domain. In our previous work, we proposed representation for trade-offs and a similarity measure between a pair of trade-offs and used them to develop a CCBR-RS. We extend our previous work to include diversity based on the compromise the products makes with the query in the current interaction cycle. We then propose a novel view of the process of conversation. The feedback in each interaction cycle was used for aggregating evidence for each product in the domain. The evidence could be both positive and negative. The relation among products both in terms of its Multi-Attribute Utility Theory based similarity and its similarity based on trade-offs is exploited to propagate positive and negative evidence among the products. We also extend the idea of evidence propagation to include higher order evidence propagation by posing the evidence propagation as a random surfer model.