Comparing Similarity Perception in Time Series Visualizations.

Area of Science:

• Data Visualization
• Human-Computer Interaction
• Time Series Analysis

Background:

• Identifying similar patterns in time series data is crucial for tasks like anomaly detection and clustering.
• Current methods often require visual inspection of automatically generated similarity results.
• The impact of different visualization techniques on similarity perception remains under-explored.

Purpose of the Study:

• To investigate how visualization techniques (line charts, horizon graphs, colorfields) affect human perception of time series similarity.
• To compare perceived similarity with results from automatic similarity measures.
• To understand if similarity perception is visualization-dependent.

Main Methods:

• Conducted two experiments using electroencephalogram (EEG) signals.
• Participants evaluated similarity of temporal patterns presented via line charts, horizon graphs, and colorfields.
• Compared human judgments with similarity scores from automatic measures like dynamic time warping and z-normalization.

Main Results:

• Horizon graphs showed alignment with similarity measures allowing temporal variations (e.g., dynamic time warping).
• Line charts and colorfields aligned better with measures insensitive to amplitude and y-offset scaling (e.g., z-normalization).
• Perceived similarity varied significantly across the tested visualization techniques.

Conclusions:

• The choice of visualization technique influences the perception of time series similarity.
• Similarity in time series is not an inherent property but is influenced by the chosen visual representation.
• Findings have implications for selecting appropriate visualizations in time series analysis and pattern recognition.