A Study on User Perception and Experience Differences in Recommendation Results by Domain Expertise: The Case of Fashion Domains

Taehyung Noh, Haein Yeo, Myungjin Kim, Kyungsik Han

Abstract: To improve user satisfaction to the recommender systems (RS), it is essential to identify how users would perceive the system and what recommendation results they would prefer. Domain experts also use RS that relate to their work and decision-making, but existing studies have primarily focused on user experience from the general public and somewhat neglected the degree of user perception, understanding, and preference to the RS according to domain knowledge and interest and recommendation algorithm types. In this paper, we present My Own Style (MOS), a dashboard tool designed to analyze a given input fashion image and recommend outfit based on three recommendation algorithms that have different degrees of similarity and diversity. Based on the results from a large-scale user study with 166 participants, our results showed that the participants who have high fashion knowledge and interest (i.e., domain experts) well understood the results of RS and preferred the recommendation algorithm that provides similar outfit, while those who have low fashion knowledge and interest did not well understand the recommendation results as much as the expert group and preferred the algorithm that suggests diverse outfit. Our work is meaningful in providing empirical evidence on how to develop, select, and utilize recommendation algorithms according to domain expertise.