Modeling Human Engagement State to Lower Cognitive Burden and Increase User Interaction Responsiveness
- Author(s): Ho, Bo-Jhang
- Advisor(s): Srivastava, Mani B
- et al.
Mobile phones and wearable devices are becoming increasingly ubiquitous, and the relation between mobile sensing devices and human beings is getting more and more intimate. Computation is no longer merely a machine's job, where hardware executes a sequence of operations; human beings are often involved in the process, as in medication intervention, preference configuration, and crisis alert. Product manufacturers and service providers rely on user engagement to make revenue, and users can benefit from these products and maximize their utility only if users are willing to engage with them. We argue that next-generation sensing computation systems should be user-engagement aware, i.e., the systems should treat "user engagement" as part of system resources to make decisions and prioritize tasks.
Although different sensing modalities and optimization techniques have been proposed, user engagement cannot be gauged directly because it is a hidden construct. Users' engagement depends on multiple variables including environmental conditions, physical constraints, psychological status, and self-interests. Unfortunately, the limitations in existing sensing techniques exacerbate the difficulty of engagement measurement. For instance, sensor data noise increases the uncertainty of inferences, battery size constrains sensor coverage both temporally and spatially, and form factor directly impacts users' willingness to carry these devices. Such issues multiply the complexity of modeling user engagement.
In this dissertation, we adapt the performance-based observation approach to describe user engagement, bridge the gap between engagement measurement with sensing techniques, and seek opportunities to further increase user engagement. We first showcase how wearable sensing systems can increase user engagement in the workout domain by performing opportunistic sensing. This dissertation then discusses how to use sensory data to model user engagement by reinforcement learning algorithms. Finally, we point out concerns specifically in sensing systems that can negatively impact user engagement.