Supporting Orientation of People with Visual Impairment: Analysis of Large Scale Usage Data
Hernisa Kacorri, Sergio Mascetti, Andrea Gerino, Dragan Ahmetovic, Hironobu Takagi, Chieko Asakawa · 2016 · Proceedings of the 18th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '16) · doi:10.1145/2982142.2982178
Summary
This paper analyzes large-scale remote usage data from iMove, an iOS GPS-based orientation app for people with visual impairments, to understand how users interact with assistive navigation technology in real-world conditions. Traditional assistive technology user studies are limited by geographically sparse participants, small sample sizes, the Hawthorne effect, and difficulty conducting longitudinal research. By analyzing anonymized log data collected from December 2015 to April 2016, the researchers studied 771,975 log records across 17,624 unique users. After filtering out "incidental" users (many confused iMove with Apple's iMovie video editor), they focused on 4,055 users with two or more interaction sessions, generating 255,004 logs. iMove provides current address, nearby Points of Interest (POIs), and user-defined geo-notes (audio or text annotations associated with locations), with a "Notify me" mode that delivers periodic location updates while walking. The analysis examined four log categories: preferences (3.41%), screen navigation (66.23%), actions (0.76%), and notifications (29.55%). Users were split into VoiceOver users (VO-group, 1,025 users) and non-VoiceOver users (NVO-group, 3,030 users).
Key findings
Stark differences emerged between VO and NVO users. VO-users (25.28% of users) generated 56.34% of all logs, with a mean of 140 records per user vs. 37 for NVO-users. VO-users used the app for significantly longer periods (mean 53.95 days vs. 20.45 days), preferred smaller temporal and spatial notification thresholds (wanting more frequent and proximity-triggered updates), and used speech significantly more than text for creating geo-notes (confirming that typing while mobile is particularly challenging for VI users). Users strongly preferred to modify default settings to receive more detailed location information including city, speed, heading, and course — parameters whose defaults should be revised. Clustering of 1,025 VO-user interaction streams using n-gram feature extraction and Louvain community detection revealed four distinct user groups: C1 (370 users) — POI explorers who frequently check nearby POI lists and details in short sessions; C2 (247 users) — continuous navigators who keep the app active during long sessions while moving, receiving frequent location and POI notifications; C3 (215 users) — quick location checkers who open the app briefly to read their current address then close it; and C4 (154 users) — brief notification listeners who start the app, listen to one or two notifications, then stop. The target user group the app was designed for (C2, continuous navigation during travel) represented only 25% of VO-users, while 75% used the app in patterns not originally anticipated.
Relevance
This study demonstrates the value of large-scale remote usage analytics for assistive technology research, overcoming the persistent challenge of small sample sizes in accessibility studies. For AT developers, the most striking finding is that 75% of the target user population used the app differently than intended — discovering actual usage patterns is essential for tailoring features to real needs rather than designer assumptions. The four user clusters provide actionable design implications: C1 users would benefit from POI lists shown on the first screen; C3 users might be served by an accelerometer-triggered interface that reads the address when the phone is pulled from a pocket; and C4 users may benefit from a simplified notification-only mode. The methodology — treating interaction logs as text streams and applying NLP techniques (n-gram analysis, cosine similarity, community detection) — is transferable to other assistive technology applications. The finding that default settings needed revision (users consistently changed notification preferences to receive more detailed, more frequent information) underscores that AT defaults should be calibrated through usage data, not designer intuition.
Tags: visual impairment · blindness · navigation · orientation and mobility · mobile accessibility · data collection · machine learning · user research · VoiceOver