Behind-The-Scenes: Recommendations

Sammanfattning

Recommender agents providing personalised recommendations for customers on online webstores or video streaming platforms have gained a lot of attention recently with the advancement of Artificial Intelligence improving their accuracy and efficiency. However, due to the black box nature of machine learning, sometimes it is difficult for consumers to understand why a certain set of items are recommended to them. Several key players, like Amazon or Netflix, have already started implementing explanations to increase the transparency of the agent and, thus, help individuals see the reason why the algorithm chose those recommendations for them. Previous research have studied how this transparency influences consumer behaviour, such as trust, perceived control or willingness to accept the recommendation. Multiple other scholars looked into how product type moderates these relations. Experience products, such as movies and mobile apps, have hard-to-evaluate attributes that are subjective and, therefore, their quality can only be determined after purchase or use. Search goods, on the other hand, have attributes that can be objectively compared and evaluated even before purchase. In the current body of literature there is a disagreement whether recommender agents are more useful for experience or search products. That is why this thesis aims to investigate how the relationship between the transparency of a recommender agent and users’ perceived uncertainty regarding how the agent works is moderated by different product types. Grounded in Uncertainty Reduction Theory and Search and Experience Theory, two hypotheses were developed, then an experimental survey was designed with four scenarios mixing non-transparent and transparent agents with the different product types. The target sample was individuals belonging to Generation Z, as they are expected to become the biggest online spenders by 2030, so their perception of recommender agents can be highly relevant for current investments in improving transparency. The results showed that the transparency of the recommender agents significantly influences individuals’ perceived uncertainty. However, product type is not interacting together with transparency in this relation, as the search product scenarios received higher perceived uncertainty scores than the experience product scenarios. These findings confirm the applicability of Uncertainty Reduction Theory to recommender agents and present a novel contribution to Search and Experience Theory, as in perceived risk or brand unfamiliarity might outweigh product type when it comes to uncertainty. The practical implications of these observations are critical for developers of recommender agents, online retailers and streaming services as to how they should calibrate the transparency of recommender agents to decrease customers’ perceived uncertainty and, thus, avoid impeding their purchase intentions.