[Paper Review] INADVERT: An Interactive and Adaptive Counterdeception Platform for Attention Enhancement and Phishing Prevention.

Linan Huang, Quanyan Zhu|arXiv (Cornell University)|Jun 13, 2021

Spam and Phishing Detection21 references4 citations

TL;DR

INADVERT is an interactive, adaptive platform that uses real-time eye-tracking and reinforcement learning to enhance user attention and prevent phishing through dynamic visual aids. By analyzing gaze data and optimizing hyperparameters via Bayesian optimization, it improves users' phishing recognition accuracy while adapting to attention fluctuations.

ABSTRACT

Deceptive attacks exploiting the innate and the acquired vulnerabilities of human users have posed severe threats to information and infrastructure security. This work proposes INADVERT, a systematic solution that generates interactive visual aids in real-time to prevent users from inadvertence and counter visual-deception attacks. Based on the eye-tracking outcomes and proper data compression, the INADVERT platform automatically adapts the visual aids to the user's varying attention status captured by the gaze location and duration. We extract system-level metrics to evaluate the user's average attention level and characterize the magnitude and frequency of the user's mind-wandering behaviors. These metrics contribute to an adaptive enhancement of the user's attention through reinforcement learning. To determine the optimal hyper-parameters in the attention enhancement mechanism, we develop an algorithm based on Bayesian optimization to efficiently update the design of the INADVERT platform and maximize the accuracy of the users' phishing recognition.

Motivation & Objective

• Address the growing threat of deceptive attacks that exploit human cognitive vulnerabilities in information security.
• Develop a system that proactively detects and mitigates user mind-wandering and inattention during phishing scenarios.
• Enable real-time adaptation of visual aids based on user attention status derived from gaze tracking.
• Optimize the attention enhancement mechanism using Bayesian optimization to improve phishing detection accuracy.

Proposed method

• Utilize real-time eye-tracking data to monitor gaze location and duration as indicators of user attention.
• Apply data compression techniques to efficiently process and analyze gaze data for system responsiveness.
• Extract system-level metrics to quantify average attention levels and frequency/duration of mind-wandering.
• Employ reinforcement learning to dynamically adjust visual aids based on real-time attention feedback.
• Implement Bayesian optimization to automatically tune hyperparameters of the attention enhancement mechanism.
• Integrate visual aid generation into the platform to counter visual-deception attacks in real time.

Experimental results

Research questions

• RQ1 How can real-time gaze tracking be leveraged to detect and respond to user inattention during phishing simulations?
• RQ2 What system-level metrics best characterize user attention and mind-wandering behavior in interactive security tasks?
• RQ3 To what extent can reinforcement learning improve user attention and phishing recognition accuracy through adaptive visual feedback?
• RQ4 How does Bayesian optimization enhance the design and performance of the attention enhancement mechanism?

Key findings

• The platform successfully identifies user attention lapses using gaze-based metrics such as fixation duration and gaze location patterns.
• Reinforcement learning enables dynamic adaptation of visual aids, resulting in measurable improvements in user attention stability.
• Bayesian optimization efficiently identifies optimal hyperparameters for the attention enhancement mechanism, reducing manual tuning.
• The system demonstrates improved phishing recognition accuracy through real-time visual feedback tailored to individual attention states.