Introduction
Digital libraries (DLs) have traditionally functioned as static repositories where information retrieval is governed by a rigid, predefined hierarchy of hyperlinks. This paper proposes a paradigm shift: the automatic generation of flexible, personalized interfaces that support "out-of-turn" interactions. By allowing users to provide unsolicited information via text or voice, the system dynamically restructures the library's navigation to match specific information-seeking goals.
The Limitations of Static Navigation
In the early 2000s, DLs typically relied on "hardwired" structures. To locate a document, a user had to follow a specific path through categories and subcategories. This creates a significant mismatch between the system's structure and the user's mental model. For instance, a researcher looking for papers by a specific author might not know which technical category those papers reside in, leading to a frustrating search through deep, unfamiliar hierarchies. The paper addresses this by proposing a "mixed-initiative" dialog, where the user can jump ahead or provide partial information at any time to reshape the interface on the fly.
Technical Framework and Methodology
The authors developed a generator that transforms a digital library model (such as an XML representation or the 5S framework) into a customized interface. The architecture relies on several key technical components:
- Staging Transformations: A framework used to model the user-DL dialog as a series of functional specifications.
- Program Slicing: Borrowing from software engineering, the system uses "slicing" techniques to identify and retain only the navigation paths that match user input, effectively pruning irrelevant branches of the hierarchy.
- OTML (Out-of-Turn Markup Language): A specialized XML-based language that allows developers to define how these interactions should behave.
- XSLT Compilation: The system uses XSLT to compile OTML into functional interfaces, specifically XUL toolbars for the Mozilla browser and SALT (Speech Language Application Tags) for voice-enabled interaction.
Key Insights: Pruning vs. Searching
A critical distinction of this system is that it prunes the navigation structure rather than just returning a list of search hits. Unlike a standard search bar, if a user provides a specific keyword, the system removes all categories that contain no documents matching that keyword. This leaves the user with a simplified, personalized version of the original library hierarchy.
Furthermore, the system is designed for multimodal flexibility. Because the underlying logic is based on transformations, the same "out-of-turn" logic can be applied to a graphical toolbar or a voice-recognition system. To assist the user, the system includes a "What may I say?" feature (meta-enquiry), recognizing that users need to know the system's vocabulary to interact effectively.
Implementation and Results
The primary testbed for this system was the Computing and Information Technology Interactive Digital Educational Library (CITIDEL). Using the classification schemes and metadata inherent in the CITIDEL repository, the authors demonstrated several personalized interaction types:
- Basic/Generalized Interaction: Pruning the classification tree based on user input.
- Collect Results: A "dialog termination" technique that flattens the remaining hierarchy into a simple list of relevant documents.
- Restructure Classification: Allowing users to change the "facet" of navigation (e.g., switching from an Author-Title hierarchy to a Journal-Author-Title hierarchy).
Qualitative studies suggest that users are adept at recognizing when to use these out-of-turn features to save time and reduce the cognitive load of navigating complex repositories.
Limitations and Future Work
Despite its successes, the system faces challenges in voice complexity; voice-enabled interfaces are currently limited to shorter phrases to maintain high recognition accuracy. Additionally, the basic version of the tool requires user input to match existing labels in the classification scheme, though the "generalized" version attempts to mitigate this by analyzing leaf-level metadata.
As "end-user programming" becomes more prevalent, the authors suggest these tools will become essential for librarians to build sophisticated, user-friendly digital repositories without requiring deep computer science expertise.
Summary updated