Investigating users' eye movement behavior in critiquing-based recommender systems

Recommender systems have increasingly become popular in various web environments (such as e-commerce and social media) for automatically generating items that match to individual users' personal interests. Among different types of recommender systems that have been developed so far, critiquing-based recommender systems have been widely recognized as an effective approach to obtaining users' feedback on the system's generated recommendations. Such systems have been demonstrated particularly helpful for serving new users. That is, by means of eliciting and refining their preferences through real-item feedback, the system is able to gradually improve its recommendation accuracy and aid users to make better decision. However, how to precisely acquire users' critiquing feedback is still a challenging issue. Most of existing systems rely on users to specify the feedback on their own, which unavoidably let users consume extra efforts. In our work, we have been engaged in analyzing users' eye-movement behavior when they evaluate recommendations, with the objective of identifying the correlation between eye movements and their critiquing feedback. The results can hence be constructive for developing an eye-based feedback elicitation method, so as to reduce users' self-critiquing efforts. Based on a collection of real users' eye-gaze data, we have tested this idea's feasibility. Moreover, we have compared different recommendation interfaces (the interface that displays a set of recommended products), and found the category layout performs better than the list structure in terms of stimulating users to view recommended products. As a result, multiple design guidelines are derived from our user experiment.