Unravelling the Factors Affecting Consumer's Usage of Context-Aware Recommender Systems from a Human-Computer Interaction (HCI) Perspective


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date7 Oct 2022


In the age of ubiquitous computing, technological advancements such as big data have enabled digital devices to gain an ever-increasing amount of contextual information (e.g., location and weather) for recommender systems, producing more personalized and relevant recommendations by adapting them to users’ contextual situations. Those systems that utilize context to provide users with personalized recommendations are known as context-aware recommendation systems (CARSs), which have become prevalent in our daily life. Compared to traditional recommender systems, CARSs incorporate contextual information into the recommendation process, deciphering the meaning behind the collected user context. At the delivery stage, CARSs are capable of performing actions proactively instead of waiting for users’ commands. As such, CARSs users are likely to encounter user experience challenges such as privacy concerns, loss of control, and perceived intrusiveness, hindering their usage decisions in the pre-adoption stage or post-adoption stage. In view of this, how to promote users’ usage decisions under user experience challenges remains unclear from a human-computer interaction (HCI) perspective.

Despite extensive prior research on CARS being conceptual and system-centered, scholars have begun to focus on an HCI and a user-centric perspective. After performing a thorough analysis of the relevant literature, this thesis identifies various research gaps and tries to explore factors influencing users’ usage decisions under the pressure of identified user experience challenges. In the pre-adoption stage, users are faced with privacy issues. The thesis firstly explored how the privacy features in CARSs influence users’ intention to adopt a CARS through their perceived privacy control. In the post-adoption stage, users’ continuance intention is vital for long-term usage. However, how system features influence users’ perceptions and in turn boost or impede their continuance intentions at the CARS also gets little attention. Hence, secondly, to fill the abovementioned research gap, the thesis conducts two sub-studies (i.e., Study 2 and Study 3) at the post-adoption stage to explore long-term continuance intentions and short-term discontinuance intentions. The former relates to whether users’ intention to use the system is strengthening or weakening over time, whereas the latter refers to whether users intend to discontinue using the system at some point.

Study 1 explores the influence of privacy features on users’ perceived privacy control, which in turn affects users’ adoption decisions in the pre-adoption stage. The study specifically identified three granular privacy features as privacy collection features, privacy usage features, and privacy protection features. Additionally, instead of regarding privacy control as a unidimensional construct, this study decomposed privacy control in CARSs into boundary permeability control, boundary ownership control, and boundary temporary control based on communication privacy management theory. The study collected data from a scenario-based online experiment and questionnaires. ANOVA results identified a cascading effect on the role of privacy features on privacy control. Specifically, privacy collection features significantly increase three types of privacy control, privacy usage features significantly influence two types of privacy control (i.e., boundary ownership control and boundary temporary control), and privacy protection features have a significant effect on only one type of privacy control (i.e., boundary temporary control). Additionally, the results showed that boundary permeability control influences consumers’ adoption decisions via reducing privacy risk probability, whereas boundary ownership control influences consumers’ adoption decisions via the reduction of both privacy risk probability and privacy risk severity. In addition, no significant effect was found for boundary temporary control.

Study 2 proposed a novel angle to illustrate how to promote users’ continued usage of CARSs: cooperative learning, which refers to a process whereby consumers cooperate with the system to accumulate knowledge about themselves. Extending the application scope from human-to-human relationships, this study applied social interdependence theory to human-system interactions. Furthermore, a research model was developed to explore how user-oriented affordances induce positive interdependence, which in turn promotes their long-term continuance intentions. Results from an online survey confirmed the role of cooperative learning in CARSs.

Study 3 examined antecedents and consequences of users’ psychological reactance in CARSs. Grounded in psychological reactance theory, this study theoretically articulated and empirically examined how context-aware features (i.e., contextual triggered actions, context reconfiguration, and contextual tagging) influence users’ perceptions of “threat to freedom” and psychological reactance, resulting in their short-term discontinuance intention. This study collected data from an online survey, and the results revealed how to reduce users’ psychological reactance by configuring context-aware features.

The findings from the three studies inform academics regarding: (1) This study summarizes prominent user experience challenges for various usage decisions in CARSs, emphasizing that different usage decisions should focus on distinct user experience challenges; (2) While previous research on CARSs from an HCI standpoint has grown, it has not delved into user perceptions. This study adds to previous research on CARS usage decisions from a “user-centered” HCI perspective; (3) Since previous studies have mostly focused on conceptual designs or theoretical explanations, this study expands the scope of empirical research on CARS. The perceptions of real-world CARS users were investigated in three separate studies using questionnaires and experiments; (4) Prior research regarding the role of system features on users’ perceptions is fragmented and coarse-grained. The impact of granular system features on users’ perceptions and usage decisions is explored further in this research. Empirical findings broaden the understandings of how fine-grained system features elicit users’ privacy control, social interdependence, and psychological reactance, respectively; (5) This thesis extends the application scope of social interdependence theory and psychological reactance theory.

    Research areas

  • Context-aware, recommendation systems, cooperative learning, user-system interaction, human-machine relationships, psychological reactance, privacy control