Defining Patterns for a Conversational Web

Title of the Paper

Defining Patterns for a Conversational Web

Paper Information

• Subject Area: Human-Computer Interaction, Accessibility Technology, and Conversational AI Design
• Keywords: Conversational UIs, Conversational Web Browsing, Design Patterns, Accessibility Technology, Human-Computer Interaction, User Research, Voice Assistants, Web Navigation, Content Segmentation, Cognitive Assistance
• Publication Date and Venue: CHI '23, Hamburg, Germany

Research Background and Problem

• Problem or Challenge:

  • Current web design remains overly visual, creating significant navigation and information access challenges for blind and visually impaired (BVI) users.
  • Screen readers, a common assistive tool, face the following issues:
    • Websites that do not adhere to accessibility design guidelines limit screen reader parsing capabilities.
    • Website content is primarily designed for visual consumption, making it difficult to convey information through voice.
  • Existing conversational AI systems are useful for search or FAQs but lack sufficient support for in-page navigation and information retrieval.

• Significance:

  • Leveraging conversational AI can provide a more natural and accessible web interaction experience for visually impaired users and other groups, making the web truly open to everyone.
  • There is a need to establish design guidelines that ensure consistent experiences across websites, optimizing the conversational web experience.

• Research Motivation and Related Work:

  • Addressing the limitations of existing conversational AI design guidelines, particularly their insufficient consideration of BVI user needs.
  • Developing patterns and guidelines specifically for web browsing tasks.

Solution

• Proposed Solution: Through a user-centered design process, the authors propose design patterns for conversational web browsing (Conversational Patterns). These patterns are based on an in-depth investigation of the information access and navigation challenges faced by visually impaired users, aiming to improve the interaction experience with conversational AI.

• Innovations:

  • A model called the "Conversation-oriented Navigation Tree (CNT)" is designed to organize and represent the hierarchical structure of web content and navigation, enabling natural language interaction with web pages.
  • Specific patterns are proposed, including navigation space mapping, hierarchical content browsing, content reading optimization, and conversational control intents.

• Key Techniques and Steps:

  • User Research: Identifying specific challenges in web browsing through surveys, interviews, focus groups, and co-design experiments, and proposing improvement suggestions.
  • Design and Validation:
    • Introducing techniques such as content segmentation, voice labels, and link prediction.
    • Developing the "ConWeb" platform to support Wikipedia pages, validating the feasibility and effectiveness of the design patterns.
  • Technical Implementation:
    • Automatically generating conversation-oriented navigation trees.
    • Using natural language processing (NLP) techniques for dialogue parsing and generation.

Research Outcomes

• Specific Outcomes:

  • Defined design dimensions for the conversational web, including navigation structure mapping, quick navigation mechanisms, page content segmentation, and summarization mechanisms.
  • Implemented various conversational patterns:
    • Hierarchical content navigation
    • Quick keyword Q&A
    • User bookmarks and categorization
    • Content segmentation and reading through paragraph scrolling
    • Conversational control mechanisms, including help, history rollback, and global flag nodes

• Advantages Compared to Existing Solutions:

  • Improved the shortcomings of existing conversational AI in web navigation, providing a more natural and cognitively lightweight interaction experience.
  • User-focused design patterns capture the needs and pain points of BVI users in detail, making them more practical and specific.

• Experimental and Evaluation Results:

  • Preliminary validation shows that the design patterns reduce users' cognitive load and help them quickly build a mental model of the website.
  • Participants appreciated the intuitiveness and consistency of the conversational web presentation, especially the ability to directly access content through quick Q&A.

• Limitations and Future Directions:

  • Limitations:
    • Did not deeply address interaction pattern design for dynamic web components and complex forms.
    • Limited coverage of a broader user base and diverse website types.
    • Small sample size, with younger participants, which may affect the generalizability of the results.
  • Future Research Directions:
    • Expanding design patterns to more dynamic and interaction-intensive websites.
    • Further integration of screen readers with conversational AI technology.
    • Investigating the applicability and evaluation of conversational web for other user groups.

This research provides new design theories and technical practices for improving conversational AI in web browsing. It opens a new chapter in accessible web design, better serving a wider range of users, including BVI individuals.