ABSTRACT:
Research on recommendation agents (RAs) originally focused on interactive RAs, which rely on explicit methods, i.e., eliciting user-provided inputs to learn about consumers’ needs and preferences. Recently, due to the availability of large amounts of data about individuals, the focus shifted toward non-interactive RAs that use implicit methods rather than explicit ones to understand users’ needs. This paper examined the differences between interactive and non-interactive RA types in terms of how they influence the impacts of two important antecedents of RA adoption, namely recommendation quality and trust on users’ cognitive and affective attitudes and behavioral intention. To that end, we developed a set of hypotheses and tested them empirically using a meta-analytic structural equation modeling approach. Our findings provide strong support for the influence of interactivity on RA users’ attitudes and cognitions. While we found that recommendation quality exerts a strong influence on consumers’ cognitive attitudes toward interactive RAs, this influence is statistically non-significant in the context of non-interactive RAs, in which recommendation quality mainly drives consumers’ affective attitudes toward the agent. Furthermore, while we found that cognitive attitudes exert a stronger influence than affective ones on consumers’ adoption of non-interactive RAs, our results indicate that the reverse is true with interactive RAs. Given the recent rise in the popularity of non-interactive RA tools, our results carry important implications for researchers and practitioners. Specifically, this study contributes to the extensive literature on consumers’ use of RAs by providing a better understanding of the differences between interactive and non-interactive RAs. For practitioners, the findings provide guidance for designers and providers of RAs on developing and improving RAs that are more likely to be adopted by consumers.
Key words and phrases: Recommendation agents, electronic commerce, system adoption, online trust, recommendation quality, meta-analysis, TSSEM, heuristic-systematic model, affective response model, interactivity, recommenders