TL;DR
This paper introduces Myriad, a novel architecture for information retrieval and delivery that dynamically adapts to a user's evolving context. By integrating adaptive hypermedia and information retrieval principles, Myriad aims to provide users with the right information, in the right form, at the right time, overcoming the limitations of traditional systems.
What Problem Does This Paper Address?
In today's information-rich environment, knowledge workers constantly grapple with finding and processing relevant information. Traditional information retrieval systems often treat each query in isolation, delivering raw results without considering the user's current situation. Conversely, adaptive hypermedia systems, while good at tailoring content, often assume a closed, pre-defined set of documents, making them difficult to scale to the vastness of the web. This paper addresses the challenge of building systems that can effectively retrieve information from diverse sources and deliver it in a contextually appropriate manner, acknowledging that user needs are dynamic and evolve throughout an interaction.
Methods
Myriad is a generic architecture built around a Virtual Document Planner (VDP). This VDP is the core component, responsible for receiving user input, understanding the user's communicative goal, and orchestrating retrieval and delivery. It leverages several key components:
- Context Models: Myriad maintains five dynamic context models to represent the user's situation:
- User Model: Captures long-term profiles (e.g., job description) and short-term information goals.
- Domain Model: Defines concepts, properties, and relationships within the application domain.
- Task Model: Describes common user tasks, including sub-steps, conditions, and effects.
- Device Model: Details the characteristics of the user's current input/output device.
- Discourse Model: Tracks the user's interaction history, including queries and document access.
- Retrieval Agents: These agents are responsible for fetching information from various data sources. The paper describes agents like
GoogleAgent,IMAPAgent,JDBCAgent, andLocalPersonAgent, showcasing a range of scopes from general web search to specific local database queries. - Delivery Agents: These agents take the retrieved content and present it to the user in an appropriate format, potentially organizing or clustering it.
The VDP uses declarative plan operators to determine how to satisfy a user's communicative goal. These operators define the effects, constraints, and actions needed for retrieval and presentation, allowing for flexible adaptation to different applications.
Datasets
Not discussed in the paper. The paper focuses on the architecture and its application in various systems (Tiddler, PERCY, DFDMSA, Skil, Tiger), implying these systems would utilize existing data sources relevant to their domains rather than relying on specific, pre-defined datasets for training the architecture itself.
Key Insights and Ideas
- Context is King (and Dynamic): The fundamental insight is that user information needs are not static but are deeply intertwined with their evolving context (who they are, what they're doing, what device they're using). Myriad's strength lies in its ability to continuously track and exploit this dynamic context.
- Bridging the Gap: Myriad elegantly combines the strengths of adaptive hypermedia (tailored delivery) and information retrieval (access to open, heterogeneous sources). This hybrid approach is crucial for real-world applicability.
- Modularity and Reusability: The architecture is designed to be generic, with components like retrieval and delivery agents, and plan operator libraries, being configurable for different information applications. This promotes reusability and reduces development effort.
- Implicit vs. Explicit User Models: The paper highlights the trade-offs between the implicit user models of traditional IR (just a query) and the explicit, rich user models of adaptive hypermedia. Myriad's context models offer a more nuanced approach than either extreme.
Key Results
The paper doesn't present quantitative results from a single, large-scale evaluation. Instead, it showcases the Myriad architecture's application in several systems:
- Tiddler: A tailored information delivery system for personalized travel information.
- PERCY: An application for corporate knowledge memory, generating personalized e-brochures.
- DFDMSA: A system for aviation applications, aiming to maximize operator efficiency by providing task-relevant information.
- Skil: A computer-assisted training environment in a medical domain, using haptic virtual environments and natural language generation.
- Tiger: A prototype system demonstrating actionable information delivery, exemplified by retrieving and presenting information about a student (grades, emails, personal details) to a professor. Figure 2 shows a sample output from Tiger, illustrating the integrated delivery of diverse information types.
The success is demonstrated through the successful implementation and ongoing development of these diverse applications, each leveraging Myriad's contextual adaptation capabilities.
Limitations and Future Work
- Scalability of Context Models: While the context models are designed to be dynamic, the paper doesn't deeply explore the computational cost or scalability challenges of maintaining and querying these models in extremely large-scale or high-throughput scenarios.
- Generalization of Plan Operators: The paper mentions the need for a more general set of VDP plan operator libraries that can be easily adapted to new information application domains. This suggests that current libraries might be somewhat application-specific.
- Future Work: The authors emphasize the need for:
- Further development of existing Myriad applications.
- A greater variety of retrieval and delivery agents to handle diverse information sources and presentation needs.
- More generalized VDP plan operator libraries to facilitate adaptation to new domains.
Summary updated