This thesis presents the design and implementation of an interactive, web-based visualization tool for analyzing sentence-level annotation logs. The system is developed to address key limitations in current annotation workflows, including the lack of visual integration between temporal activity, annotation category effort, and self-reported confidence. Using log data from the SKILL Project, the tool transforms raw CSV entries into structured JSON datasets and visualizes them through a React and D3.js frontend.

Three core research questions guide the analysis: (1) how temporal factors affect annotation efficiency, (2) how different annotation categories vary in cognitive demand, and (3) how confidence levels relate to annotation time. To explore these dimensions, the tool features session-level Gantt charts, bar and radar plots for category comparison, and multivariate scatter and matrix plots for analyzing confidence-efficiency relationships.

The findings reveal that annotation performance varies significantly by time of day, category type, and annotator strategy. Peak efficiency tends to occur during late morning hours; domain-level categories often require more effort than discourse-level ones; and lower confidence is frequently associated with longer durations. These patterns support previous research and underscore the value of combining behavioral data with subjective metrics.

The tool’s interactive design facilitates visual exploration, pattern detection, and workflow optimization. It contributes to research in human-computer interaction, annotation analysis, and visual analytics, offering a reusable framework for improving transparency and efficiency in annotation-based projects.