Geri Gay

The paper analyzes the reliability of clickthrough as implicit feedback in web search. Eyetracking and comparisons with manual judgments show that while clicks are biased and cannot be treated as absolute relevance labels, relative preferences derived from clicks are on average accurate. The results guide how to use click data for learning to rank and evaluation in information retrieval systems.

Using eye‑tracking with college students, the study shows substantial trust in Google’s ranking: users’ click choices are strongly biased toward higher‑positioned results even when abstracts are less relevant. When retrieval quality was artificially degraded, participants scrutinized more but still failed to match normal success rates. The findings highlight how positional bias and trust in ranking influence user behavior, with implications for culture, society, and web traffic.

This paper examines how reliably implicit feedback signals—especially clicks and query reformulations—indicate relevance in web search. Combining eye‑tracking to analyze decision processes and comparisons with manual relevance judgments, the authors show that clicks are informative yet biased (e.g., trust and context biases). While absolute interpretations of clicks can be problematic, relative preferences inferred from click behavior are reasonably accurate, including across chains of query reformulations.

This empirical study of 31 distributed learners examined relationships among communication styles, social networks, and learning performance in a CSCL setting. Using social network analysis and longitudinal surveys, the authors found that both individual factors (e.g., willingness to communicate) and initial positions influenced development of collaborative networks. Centrality in emergent networks predicted higher final grades, underscoring the importance of communication and social structure in distributed learning.

The study investigates how domain expertise and feedback affect search term generation and reformulation across 10 trials. Comparing experts and novices under two feedback conditions, the authors coded behaviors such as unique term count and time per trial, identifying nine strategies. Differences emerged by expertise and feedback, and strategic behavior depended on prior search conditions. Implications include digital library usability, metadata generation, and query‑expansion system design.

An eye‑tracking study investigated how users view and evaluate search results pages to understand browsing patterns and selection of links. Initial results report time spent per abstract, total abstracts viewed, and how thoroughly searchers evaluate results. The findings inform better interface design and interpretation of implicit feedback (e.g., clickthrough) for machine learning approaches in information retrieval.

With 30 participants viewing 22 pages from 11 popular sites, this eye‑tracking study examined how individual differences, site types, page viewing order, and task instructions affect ocular behavior on single web pages. Gender, viewing order, and interactions between order and site type significantly influenced behavior; task instructions did not. Scanpath analysis suggested that page design complexity affects the variability of viewing patterns across users.

Two experiments in an upper‑level Communication course examined how in‑class laptop use for browsing/search/social computing affected learning. Students allowed to use laptops during lecture showed decrements on traditional measures of memory for lecture content versus those with laptops closed. A second experiment replicated the effect. Analyses by “browsing style” and interpretation via Lang’s Limited Capacity Model suggest multitasking reduces encoding resources and contributes to poorer immediate learning outcomes.

E‑graffiti is a campus‑based, context‑aware application that detects user location and displays text notes tied to that location; users can also create location‑linked notes. Deployed to 57 students using campus wireless, usage logs and a post‑semester questionnaire revealed challenges for ubiquitous systems including misleading conceptual models, reliance on explicit input, the need for highly relevant contextual focus, and benefits of rapid, iterative prototyping with user evaluation.