Improving online recommendation systems - a user-centric approach
Personalised recommendations are essential and highly familiar to regular Internet users. Big online platforms (e.g. Google, YouTube, and Facebook) are increasingly employing individual behavioural data to personalise content and keep users on their platforms. Hence, recommendation systems (RSs) are critical to the convenience, and usefulness users perceive when exploring new online content. Accordingly, practitioners and researchers across different disciplines (e.g. information systems, computer science or marketing) aim for better recommendation quality by improving RSs, proposing several improvement approaches.
However, much of the current endeavour to improve RSs focuses mainly on the technical rather than the user perspective, reflecting the fact that online platforms significantly invest in improving recommendation quality through passive data collection. Academic studies on RSs also prefer passive data collection, given its convenience and high availability. Nonetheless, despite the significance of technical improvement for better recommendation quality, such improvement does not guarantee user satisfaction because not all users are willing to give up their privacy and allow their data to be collected passively. Thus, the user perspective is also important.
From a user viewpoint, it is essential to consider the degree of users’ involvement with the product and their attitudes towards privacy and convenience. Product involvement affects how individual users value the usefulness of online recommendations as well as their interest in these recommendations. The degree of users’ orientation towards privacy or convenience may affect how they consider their privacy/convenience calculus when providing online data. Thus, this study proposes the recommendation system participation and outcomes (ReSPO) framework to conceptualise the relationships between these two inherent characteristics and user satisfaction and willingness to provide data.
This study employs a survey methodology to examine the relationships in the conceptual framework. The survey spanned three stages to ensure validity. The first stage (pilot tests) surveyed PhD students at the Victoria University of Wellington to determine the survey content and structure. The second stage employed a bigger sample of undergraduate students at the same university to examine the construct validity. The study then employed data from 505 users in Wellington, New Zealand, as a foundation for the main statistical analysis.
Accordingly, product involvement and individual orientation towards privacy and convenience significantly impact users’ willingness to provide online data. Moreover, privacy/convenience orientation is highly reliable in capturing individual orientation to the choice between online convenience and privacy risks when providing data. Notably, relative to product involvement, this orientation is more significant in influencing users’ willingness to provide data than product involvement. Further, the study confirms prior literature findings on the positive relationship between recommendation quality and users’ satisfaction with the recommendation process. In addition, four distinctive user categories in perceiving the value of online recommendation processes – Willing Critics, Private Connoisseurs, Acceptors and Private Receivers – emerge, building on the foundation of privacy/convenience orientation and product involvement. For each user category, as per their methods of content exploration, the analysis suggests specific recommendation approaches.
This study contributes to the theoretical and practical development of RSs. On the former, the study creates and validates a versatile new construct – Privacy/Convenience Orientation – to identify the degree to which users value their online convenience over their privacy or vice versa. Beyond the RSs context, this construct can be applied in any context requiring users to provide their online data. Regarding practical development, the findings can help practitioners better understand the potential of using actively collected data from some categories of users to improve the recommendation process. The study also reinforces the importance of using other mechanisms to complement RSs and help users explore online content (e.g. discussion forums and expert reviews).