Personalization of User Interfaces for Browsing XML Content Using Transformations Built on End-User Requirements
Benoît Encelle, Nadine Baptiste-Jessel · 2007 · Proceedings of the 2007 International Cross-Disciplinary Conference on Web Accessibility (W4A) · doi:10.1145/1243441.1243459
Summary
This paper from IRIT at the University Paul Sabatier in Toulouse presents a model for generating personalized multimodal user interfaces for browsing XML content, driven by end-user requirements rather than developer-imposed defaults. The authors identify a core accessibility problem: content providers dictate how content is presented and navigated, leaving users with no control over output modalities or scanning possibilities. For people with visual impairments, for example, this means they cannot specify that RSS feed titles should be read via text-to-speech while descriptions appear on a Braille display, or that they want to navigate items by speaking a number. The proposed solution introduces a profile-based architecture with two layers: stereotype-based profiles (created by experts for user groups like blind users, wheelchair users, or elderly people) and personalized profiles (refined by individual users). Each profile contains application policies that govern two aspects of content browsing — presentation policies (which content parts to show and through which modalities) and navigation policies (how the user scans within content using input events). Users specify these policies through user-friendly description languages built with LL(1) grammars, which are then automatically converted into XSLT transformation rules that produce the adapted interface. The system leverages XPath to target specific XML elements and SMIL timing features for coordinating multimodal output.
Key findings
The utility evaluation using the CARE framework (Complementarity, Assignment, Redundancy, Equivalence) demonstrated that the policy description languages could express all CARE multimodal interaction properties except one, confirming they can describe a wide range of multimodal user interface configurations. The usability evaluation with 20 subjects (10 experts, 10 novices based on programming experience) showed that novice users rapidly closed the gap with experts: by the third exercise, the difference between expert and novice averages dropped to just 2.5% of the initial gap, with novice scores averaging 5.875 out of 6 compared to expert scores of 5.960. This suggests that end users without programming skills can learn to specify their own browsing preferences after minimal training (a 15-minute tutorial). The transformation pipeline from user-friendly policy specification to syntax tree to XSLT templates was fully automated using the visitor design pattern, meaning users never need to interact with the underlying XSLT complexity. The implemented XML Multimodal Browser (XMB) system, built in Java for platform independence, used an eXist XML database to store User Global Profiles and could apply personalization policies to any XML-conformant resource.
Relevance
This paper tackles a fundamental tension in web accessibility: the gap between what assistive technologies provide by default and what individual users actually need. Rather than treating all blind users or all elderly users identically, the approach recognizes that accessibility requirements are deeply personal — one blind user may prefer speech for titles and Braille for body text, while another may want the reverse. The profile architecture that combines expert-defined stereotypes with user-customizable overrides is a practical pattern that modern accessibility tools could adopt. The finding that novice users quickly learn to specify their own preferences challenges the assumption that personalization must be handled entirely by developers or accessibility specialists. For current practitioners, the concept of separating content structure (XML) from presentation and navigation policies resonates with modern approaches like CSS, ARIA, and progressive enhancement. The work also anticipated the growing importance of multimodal interfaces — combining speech, Braille, visual display, and gesture input — which is increasingly relevant as voice assistants, haptic devices, and screen readers continue to evolve.
Tags: personalization · multimodal interaction · user profiles · content adaptation · XML · model-based user interfaces · braille · text-to-speech
Standards referenced: XSLT · XPath · SMIL · CC/PP