Gravi++ : an interactive information visualization for high dimensional, temporal data

Thesis
Author
Advisor
Reviewer
Abstract

Tracking and comparing psychotherapeutic data derived from questionnaires involves a number of highly structured, time-oriented parameters. Because of the small amount of data items, traditional statistical analysis are not suitable. Therefore, we decided to analyze this special kind of data with Information Visualization techniques. As a first step we searched for existing visualization methods that could be applied to our problem characteristics. We found out that there are no techniques that are able to perfectly cope with our kind of data and support the medical personnel in analyzing the psychotherapeutic data and, thus, we decided to develop a new visualization technique for this reason.

In this thesis we propose an interactive Information Visualization called Gravi++ to visualize highly structured, temporal, categorical data. It integrates a spring-based core visualization to display the multidimensional data set. To cope with the temporal dimension, we integrated animation and an additional view called Traces that maps the animation on a single image. Furthermore, the Star Plot view, Attraction Rings view, and the tooltips can give information on the exact values of each dimension. In addition to this, we added two overview visualizations called ListVis and TableVis to handle larger amounts of data and enable the user to select a subset. The different views are coordinated using linking and brushing and provide many different interaction possibilities.

In two different use cases we present some typical steps when analyzing data with Gravi++. One use case demonstrates interactions on the medical data set we used in our evaluation and the other one demonstrates the use of Gravi++ when visualizing a larger data set of Eurobarometer data (questionnaires to assess the attitude towards the European Union).

To get more information on the usefulness of Gravi++ we conducted an extensive evaluation of our visualization. The first step was a usability evaluation to preclude that usability problems are mixed up with problems with the visualization method as such. After that we conducted an insight study, where we wanted to find the strength and limitations of supervised Machine Learning (ML) techniques, Exploratory Data Analysis (EDA), and Gravi++ for our problem domain. It indicates that Gravi++ is a very good method for analyzing this kind of data and that using ML, EDA and visualization methods in conjunction will very likely contribute to a deeper comprehension of the data to explore.

Keywords
Year of Publication
2006
Academic Department
Institute of Visual Computing and Human-Centered Technology
Number of Pages
133
University
TU Wien
City
Vienna
reposiTUm Handle
Download citation