[Paper Review] The Decision-Theoretic Interactive Video Advisor

Hien Nguyen, Peter Haddawy|arXiv (Cornell University)|Jan 23, 2013

Recommender Systems and Techniques9 references18 citations

TL;DR

This paper introduces DIVA, a decision-theoretic interactive video advisor that recommends movies by modeling user preferences through pairwise comparisons rather than numeric ratings. It employs a probabilistic conflict-based similarity measure between preference orderings and uses incremental feedback to refine recommendations, demonstrating improved personalization on the EachMovie dataset through formal decision-theoretic preference elicitation.

ABSTRACT

The need to help people choose among large numbers of items and to filter through large amounts of information has led to a flood of research in construction of personal recommendation agents. One of the central issues in constructing such agents is the representation and elicitation of user preferences or interests. This topic has long been studied in Decision Theory, but surprisingly little work in the area of recommender systems has made use of formal decision-theoretic techniques. This paper describes DIVA, a decision-theoretic agent for recommending movies that contains a number of novel features. DIVA represents user preferences using pairwise comparisons among items, rather than numeric ratings. It uses a novel similarity measure based on the concept of the probability of conflict between two orderings of items. The system has a rich representation of preference, distinguishing between a user's general taste in movies and his immediate interests. It takes an incremental approach to preference elicitation in which the user can provide feedback if not satisfied with the recommendation list. We empirically evaluate the performance of the system using the EachMovie collaborative filtering database.

Motivation & Objective

• To address the challenge of representing and eliciting user preferences in recommender systems using formal decision-theoretic methods.
• To develop a system that distinguishes between general movie taste and immediate user interests.
• To enable incremental preference elicitation where users can provide feedback to refine recommendation lists.
• To evaluate the system's performance using the EachMovie collaborative filtering dataset.
• To demonstrate the effectiveness of using probabilistic conflict measures between preference orderings in preference modeling.

Proposed method

• Represents user preferences via pairwise comparisons between movies instead of numerical ratings.
• Introduces a novel similarity measure based on the probability of conflict between two preference orderings.
• Models user preferences as a combination of general taste and context-specific interests.
• Uses an interactive, incremental elicitation strategy where users can reject or refine recommendation lists.
• Applies Bayesian decision theory to update preference models and select optimal recommendations.
• Employs a probabilistic framework to compute the expected utility of recommending a movie based on current preference estimates.

Experimental results

Research questions

• RQ1 How can pairwise comparisons better represent user preferences than numeric ratings in movie recommendation?
• RQ2 What is an effective probabilistic measure of similarity between user preference orderings?
• RQ3 How can a system dynamically refine recommendations through user feedback in an interactive manner?
• RQ4 To what extent does distinguishing between general taste and immediate interests improve recommendation accuracy?
• RQ5 Can a decision-theoretic approach outperform traditional collaborative filtering in preference elicitation efficiency?

Key findings

• The system achieves improved recommendation accuracy by modeling preferences through pairwise comparisons, reducing reliance on potentially inconsistent numeric ratings.
• The conflict-based similarity measure effectively captures the degree of alignment between user preference orderings, enabling better similarity estimation.
• Incremental feedback significantly enhances recommendation quality, allowing the system to adapt to user-specific interests over time.
• The separation of general taste and immediate interests leads to more accurate and contextually relevant recommendations.
• Empirical evaluation on the EachMovie dataset confirms that the decision-theoretic approach outperforms baseline methods in preference elicitation efficiency and recommendation quality.
• The system demonstrates robust performance with minimal user input, highlighting the effectiveness of its interactive, feedback-driven design.