Overview
The 2007 Workshop on Web Information Seeking and Interaction (WISI) served as a critical bridge between academic information-seeking theory and the practicalities of industrial search engine design. This report synthesizes the discussions from that gathering, highlighting a fundamental shift in perspective: search is no longer viewed as a simple query-response event, but as a complex, interactive task. The workshop advocated for moving beyond traditional relevance metrics toward deeper measures of user engagement and behavioral modeling.
Addressing the Interaction Gap
By the mid-2000s, the rapid expansion of the Web created a significant disconnect between user behavior and system design. While researchers studied information-seeking in controlled environments, engineers at major firms like Google and Yahoo! were building massive systems that often treated users as "black boxes."
The workshop identified several key friction points:
- User Diversity: The challenge of designing a single interface that serves both "power users" and infrequent searchers with complex needs.
- The Relevance Ceiling: Traditional evaluation methods focused solely on whether a result was "relevant," failing to capture the nuances of exploratory search, learning, or user frustration.
- The Interaction Gap: The difficulty of introducing new features, such as search assistants, without cluttering the interface or impeding the user's workflow.
Methodological Frameworks
Rather than presenting a single algorithm, the workshop synthesized various methodologies to better understand and improve the search experience. Key approaches included behavioral characterization, where users were profiled as "cognitive misers" who are risk-averse and prone to "tunnel vision" (focusing only on top results).
One of the most significant conceptual contributions was the "Improvement Spectrum," which categorizes strategies for enhancing the search process.
This model ranges from Manual interventions, such as teaching users better information literacy, to Automatic solutions like system adaptation and task modeling. Between these lies Focussed Teaching, where the system provides active help, query feedback, and manages user expectations by exposing its underlying model.
Key Insights: Beyond Relevance
Several "aha moments" emerged from the workshop's cross-disciplinary dialogue:
- The Mouse as a Reading Aid: Researchers discovered that users often use their mouse pointer as a physical marker, moving it across text as they read or using it to "mark" a result while scanning others. This suggests that mouse tracking can serve as a low-cost proxy for user attention.
- Engagement over Relevance: The workshop argued that "Engagement" is a higher-order metric than simple relevance. A session without a click might still represent a successful learning task, while a click on a relevant document might still result in a poor overall user experience.
- The Academic-Industry Divide: A recurring theme was the mutual need between sectors: academics require access to large-scale industry data, while industry practitioners seek general, coherent principles of interaction that can be implemented at scale.
Results and Future Directions
Experimental results presented at the workshop showed promise for new interaction models. For instance, studies by Rodden and Fu found a strong correlation between mouse and eye movements, validating mouse tracking as a scalable research tool. Additionally, adaptive personalization models demonstrated success in helping users re-find information by prioritizing results based on predicted tasks.
However, several challenges remain:
- The Intrusion Paradox: Balancing the need for user data to improve the experience with the user's desire for a low-friction, non-intrusive interface.
- Information Literacy: Recognizing that since most users have no formal training in search, "teaching" the user may be as vital as improving the algorithm.
- Defining the Science: The community continues to grapple with whether web information seeking is an applied science and how to ensure its models are robust and generalizable across different environments, from e-commerce to digital libraries.
Summary updated