Predictive Link Following Plug-In For Web Browsers
Lyle Pierson Stachecki, John Magee · 2017 · Proceedings of the 19th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS) · doi:10.1145/3132525.3134776
Summary
This demonstration paper presents the first public release of a target-aware pointing assistance plug-in for web browsers that uses Predictive Link Following to help people with motor impairments select links on web pages. Web browsing is challenging for users of mouse replacement interfaces (head trackers, eye-gaze systems, mouth joysticks) because small links and cluttered pages require high-precision movements that these users often cannot achieve. Clicks may miss intended links, nearby links may be followed unintentionally, and dwell-time selection interfaces suffer from the "Midas Touch" problem where pausing the pointer causes unintended clicks. The system, developed at Clark University and inspired by years of observations from Camera Mouse users, works by intercepting mouse click signals and instead following links based on accumulated prediction scores. Rather than requiring the cursor to be precisely positioned within a link's bounding box, the algorithm continuously calculates each link's score based on cursor proximity over time: P_t(a) = τ * P_{t-1}(a) + pointer(a), where τ is a time decay factor and the pointer function weights distance inversely. When a click occurs and a link's score exceeds a configurable threshold, that link is followed. The system provides visual feedback by highlighting the highest-scoring link in red and other high-scoring nearby links in orange. Configurable parameters (distance weighting β, click weight γ, threshold) allow customization for individual users' abilities. The key technical contribution is implementing this as a browser plug-in rather than a standalone application — because web browsers expose link locations to scripts and plug-ins, the system works on any web page, including web-based email and social media applications.
Key findings
Two prior experimental evaluations validated the approach. In the first study, seven participants used the Camera Mouse (head-controlled pointer) to browse custom web pages. Four of seven performed better with predictive link following, one was equal, and two performed worse. Participants reported being able to click links they would have missed with traditional dwell-time selection. Many early errors came from links being followed unintentionally during page loading, which was resolved by disabling clicking during page loads. In the second study, seven new participants used a Tobii EyeX eye-gaze system. All seven reported better performance with predictive link following compared to blink-clicking. The additional jitter inherent in eye-gaze control actually favored the predictive approach because scores accumulate over time rather than depending on an instantaneous click position — effectively smoothing out the noise in cursor position. The plug-in release makes the system freely available and compatible with any input device or mouse replacement interface, working on real web pages rather than just custom test pages.
Relevance
This tool addresses a practical barrier that affects many people with motor impairments: the inability to precisely click small or closely clustered links during web browsing. The approach is notable for several reasons. First, it reframes clicking from a precise command ("follow this link") to an indicator of intention ("I'm trying to click near here"), which better matches the capabilities of users with limited motor control. Second, by implementing as a browser plug-in rather than modifying individual websites, it works universally without requiring web developers to make changes. Third, the configurable parameters allow personalization for different motor abilities and input devices. For web developers and accessibility practitioners, this work highlights that even when websites meet technical accessibility standards, the interaction model of precise pointing remains a significant barrier for motor-impaired users. Future plans to use machine learning for automatic parameter tuning and crowdsourced usage data collection could further improve the system's adaptability.
Tags: motor accessibility · web accessibility · mouse replacement interface · eye tracking · assistive technology · pointing assistance · browser extension · input methods