Bring Privacy To The Table: Interactive Negotiation for Privacy Settings of Shared Sensing Devices

Document Title

Bring Privacy To The Table: Interactive Negotiation for Privacy Settings of Shared Sensing Devices

Document Information

• Research Area: Privacy protection technologies, IoT user privacy negotiation
• Keywords: Usable privacy, IoT, negotiation agents, privacy-enhancing technologies, privacy profiles, preference elicitation

Research Background and Problem

• Issues and Challenges:
  1. With the proliferation of IoT devices, the large-scale collection of personal data poses significant privacy risks.
  2. In shared environments, conflicts between different users' (e.g., homeowners and visitors) privacy preferences remain unresolved, and traditional control mechanisms struggle to balance multiple interests.
• Research Importance:
  1. Privacy issues exposed by IoT devices may undermine users' trust in technology, necessitating new approaches to mitigate privacy conflicts.
  2. Addressing privacy negotiation in shared environments can enhance user satisfaction and promote the adoption of IoT technologies.
• Research Motivation and Related Work:
  1. Existing studies have proposed privacy notification and control mechanisms, but most fail to address negotiation issues arising from multi-user privacy conflicts.
  2. Previous privacy negotiation research has largely focused on bilateral negotiations, with limited exploration of system designs that balance multi-party interests.

Solution

• Methods and Solutions:
  1. Proposing the system ThingPoll, designed to provide negotiation tools for privacy settings of IoT devices in shared spaces.
  2. ThingPoll models user privacy preferences and incorporates negotiation guidance to help parties reach consensus.
• Innovations:
  1. Observing verbal negotiation behaviors of 12 participants to design a system that ensures efficient, fair, and user-friendly negotiations.
  2. Utilizing Bayesian network-based privacy preference modeling to predict and update user preferences.
  3. Introducing dynamic interaction mechanisms that improve negotiation efficiency through queries and suggestions, reducing lengthy dialogues.
• Implementation Steps:
  1. Generating user privacy configuration models based on privacy preference survey data from 198 Amazon Mechanical Turk users.
  2. Designing an interactive process to guide user preferences, including preset preference allocation, feature preference selection, privacy inquiries, and configuration suggestions.
  3. Employing dynamic optimization algorithms to balance the information gain from querying user preferences with minimizing user burden.

Research Outcomes

• Specific Results:
  1. The system helped users reach consensus in 97.5% of scenarios within an average of 3.27 minutes.
  2. Participants' overall satisfaction with ThingPoll reached 83.3%, significantly outperforming other baseline methods (e.g., homeowner configuration or voting mechanisms).
• Advantages:
  1. Reduces social awkwardness associated with face-to-face negotiations and improves negotiation efficiency.
  2. Facilitates mutual understanding of privacy needs through contextualized suggestions.
  3. Offers higher fairness and transparency compared to traditional methods.
• Experiment or Evaluation Results:
  1. System evaluations indicate that ThingPoll optimizes cooperative outcomes even in high-conflict scenarios.
  2. NASA Task Load Index results show overall low user burden, though some users reported experiencing mental strain during tasks.
• Limitations and Future Directions:
  1. Designed specifically for household scenarios, without considering complex social relationship networks.
  2. Does not fully address the issue of dynamically learning user preferences over extended periods.
  3. Future directions include introducing anonymous negotiation, asynchronous interaction, and dynamic adaptation to broader social contexts.