MULTIDIMENSIONAL INFORMATION VISUALIZATION

TECHNIQUES

Evaluating a Taxonomy of Tasks with the Participation of the Users

Eliane Regina de Almeida Valiati, Josué Toebe, Antonio Flávio Gomes, Milene Andréa Guadagnin

Leandro Luis Bianchi and João Roberto Telles

Faculdade de Tecnologia SENAC de Passo Fundo, Avenida Sete de Setembro n 1045, Passo Fundo, Brazil

Keywords: Information Visualization, Taxonomy of tasks, Usability evaluation.

Abstract: Multidimensional information visualization techniques has the potential to assist in the analysis and

understanding of large volumes of data by detecting patterns, clusters and trends which are not obvious,

when using non-graphical forms of presentation. When developing a visualization technique, the analytic

and exploratory tasks that a user might need or want to perform on the data should guide the choice of the

visual and interaction metaphors implemented by the technique. Usability tests of techniques for

visualization also need a clear definition of tasks of the user. The identification and understanding of these

tasks is a matter of recent research in the area of visualization of information and some works have

proposed taxonomies to organize them. This paper describes an experimental evaluation of a classification

based on the observation of different profiles of users performing tasks in exploratory data using

multidimensional visualization techniques.

1 INTRODUCTION

Techniques of visualization have been developed to

support the navigation, manipulation and

information extraction from large data sets.

The identification and understanding of the

nature of the tasks of the user in the process of

acquisition of knowledge in visual representations is

a matter of recent research in the area of

visualization of information (Stasko, 2006).

This article aims to present the evaluation of a

classification of tasks the user describing two

experimental procedures involving different user

profiles, reporting and discussing the different

results.

2 RELATED WORK

Weherend and Lewis (1990) and Springmeyer

(1992) in the early 90’s were among the first ones to

explicitly address user operations and tasks

characterizing the data analysis process in order to

facilitate the selection of adequate visual

representations.

With the goal of facilitating the choice of visual

representations, Weherend and Lewis (1990)

classified operations that a user might need to exe-

cute to analyze data.

Later on, Zhou and Feiner (1998) introduced

another categorization of tasks. They separated

presentation intents (goals a user has when using a

visual representation) from low-level visual

techniques (the exact operation performed on a

given object presented in the display) by means of

an intermediate level, the visual tasks.

Amar and Stasko (2004) proposed a new

taxonomy with higher level tasks, that can provide a

better support to visualization systems designers and

evaluators. In a very recent work, Amar et al. (2005)

proposed a taxonomy of 10 low level tasks based on

196 analytic questions found by students when

analyzing data with commercial visualization

systems.

3 TAXONOMY OF TASKS

This section presents the taxonomy of specific users’

tasks we used to guide the selection of tasks of our

experiment. The taxonomy was designed to support

the design of different scenarios for the evaluation of

multidimensional visualization techniques.

The taxonomy comprehends seven tasks:

identify, determine, visualize, compare, infer,

configure and locate. Five of these tasks can be

considered as goals a user might have when using a

311

Regina de Almeida Valiati E., Toebe J., Flávio Gomes A., Andréa Guadagnin M., Luis Bianchi L. and Roberto Telles J. (2009).

MULTIDIMENSIONAL INFORMATION VISUALIZATION TECHNIQUES - Evaluating a Taxonomy of Tasks with the Participation of the Users.

In Proceedings of the International Conference on Knowledge Discovery and Information Retrieval, pages 311-314

DOI: 10.5220/0002305903110314

 SciTePress

visualization technique for either visually exploring

or analyzing the data set through some statistics

(identify, determine, compare, infer and locate). The

other two tasks (visualize and configure) are typical

intermediate level tasks that support the analytical

ones.

4 EXPERIMENTS

This section describes two experimental procedures,

in which different profiles of users (actual and

experimental) used the same techniques for

visualization of information for the use of

multidimensional data.

The main objective of the experiments was to

identify the interactive tasks performed by users

during the data analysis and exploitation, and verify

that experimental procedures would be more

appropriate to evaluate the classification of tasks

consistently.

4.1 Visualization Techniques

In experiments used two different implementations

of techniques for geometric visualization. One of the

component implementations Xmdvtool package

(available at http://davis.wpi.edu/ ~ xmdv). The

other implementation is part of an application,

developed by Hoffman (1999) and his group,

available at (http://ivpr.cs.uml.edu/ ~ hoffman /).

4.2 Experiment I: Trial Users

In the first experiment were tested for interaction

with users 11 students of the discipline of HCI, the

Course of Computer Science.

4.2.1 Data Sets

The classical data set containing information about

American, Japanese and European cars

manufactured between 1970 and 1982 was used.

This data set was selected due to the familiarity

all the students would have with the domain,

facilitating the understanding of questions as well as

their accomplishment. Moreover, it has been used in

many data mining and visualization systems for

evaluation purposes, making easier further

comparison of results.

4.2.2 Procedure

Before the experiment, users received training

regarding the use of techniques. At the beginning of

the test of interaction, each user received a list

containing 4 high-level analytical issues to be solved

using two techniques of visualization and was

instructed to verbalize all actions taken and

problems encountered (“think aloud” method).

The experiments were conducted individually, in

the laboratory, in the presence of an observer noted

that the sequence of tasks involved in resolving each

issue.

Users were randomly selected to use two

techniques in alternating order, so that 5 users and

then used Radviz Parallel Coordinates (the

application of Hoffman) and 6 users using Parallel

Coordinates and then Matrix ScatterPlots (the

package Xmdvtool), totaling 22 comments.

Completed the tests of interaction, the scenarios

observed were compared to scenarios estimated by

the evaluator (ie, the sequences of tasks to achieve

the answers to questions).

4.2.3 Results

Looking up to the 88 real scenarios (8 real scenarios

performed by each of the 11 users in the solution of

the 4 questions) observed during this experiment and

comparing them to estimated scenarios was possible

to observe that independent of the techniques used

for all users execute the resolution of each issue

basically the same tasks, with very few variations.

The only differences relate to the way that the

iterative sequence of actions (subtasks) occurred

during the analysis and exploitation of data between

users.

Due to the exploratory and iterative nature in

search of solutions: the use of views and subsequent

analysis of data users conducted repeatedly return to

certain tasks, in different ways, in an attempt to

understand the issues and solutions proposed.

Considering all real scenarios, therefore, was to

perform all tasks of the proposed classification (see

Table 1) except the task "infer" which may not occur

due to the type of questions proposed to users.

Table 1: Tasks observed by Experiment 1.

Tasks Number of

incidents per user

Types of incidents

Identify 9 groups, data distribution, similarities,

differences, patterns, correlations.

Determine 1 values, average

Visualize 7 n dimensions, n items, data

Compare 3 groups, data, figures, graphics primitives

(color, shapes, sizes)

Infer 0

Configure 8 Filtering, primitive graphics

Locate 4 items, data, figures, groups, primitives

graphics

KDIR 2009 - International Conference on Knowledge Discovery and Information Retrieval

312

4.3 Experiment II: Real User

The second experiment was a case study involving

one geographer that work in a research project

related to Urban Area Sociospatial Diagnostic. He

was an expert user in Geography domain and having

also a good experience in information analysis

activity.

The main goal of user in this case study was to

verify the relation between socioeconomic data

about habitation, employment, education level and

revenue of boroughs and residential areas of a city.

4.3.1 Data Sets

The data set for this case study have appertained to

researcher herself, containing socioeconomical data.

The data set used had 241 items and 12 dimensions.

4.3.2 Procedure

We use the same methodology for all longitudinal

case studies, according guidelines for MILCs

described in (Shneiderman, 2006), focusing on

participatory observation and interviews, like

adopted by (Seo, 2006). Nevertheless, for clarity, we

describe betimes the main proceedings for each case

study.

Before the beginning of experiment (the data

analysis), each user was trained on visualization

techniques to be used.

There was no a priori rigid and fixed protocol

defined for users behaviour: the number of meetings

with experimenter, the time of observation, data sets

for analysis and also the high-level analytical

questions for data exploration were defined by the

users themselves, as answers to real work questions.

Likewise, we requested for each user to use the

visualization techniques as far as he hasn’t seen any

additional understanding about the data in analysis

(considering the number of meeting with

experimenter and the time of duration each session).

During the sessions (always occurring weekly

with the presence of experimenter), each user was

observed and stimulated to speak (“think aloud”

method) while doing data analysis and exploitation

using visualization techniques.

After the session end, all the registers were

reorganized in order to allow a better analysis of

collected data..

The case study had 5 user-experimenter

meetings, completing 12 observation hours. In the

first meetings, user have used the Parallel

coordinates and then ScatterPlots Matrix (from

Xmdvtool package), and for the remainder the

techniques available in Hoffman’s application.

4.3.3 Results

Through the analysis of records made in each

meeting, the information collected were categorized

into: 23 high-level analytical issues, different

instances of tasks and subtasks.

Therefore, during the process of visual analysis

and exploration of data, the user made several

analytical issues related to the factors observed.

Examples of high-level analytical issues, which

could be observed and recorded, one can cite: (1)

"What is the profile of the neighborhood X? Which

districts have a similar profile?” (2) Are there

significant socioeconomic differences between

neighborhoods and lots?".

Table 2 summarizes the observations with

respect to the tasks performed, verifying that the

tasks of greatest occurrence were, respectively,

Configure, Visualize. Compare and Identify as the

resolution of almost all the tasks / issues analytical

high-level they appear many subtasks at different

times and levels.

There is also the various types of occurrences of

each task, which in this experiment were to detect

possible.

Table 2:Tasks observed by Experiment 2.

Number of

incidents per

user

Types of incidents

Identify 59 groups, correlation,

properties, characteristics,

similarities, differences,

dependency, independence,

changes in data

Determine 16 values, average, variance,

range, amounts, proportions,

differences, probability

Visualize 81 n dimensions, n items, data

Compare 55 dimensions, items, data,

figures, groups, properties,

proportions, positions,

distances, primitive graphics

(color, shapes, sizes)

Infer 17 hypotheses, rules, trends,

probabilities, causes / effects

Configure 83 rating, filtering, zoom, order

of size, attributes derived,

primitive graphics

Locate 47 items, data, figures, groups,

properties, positions,

distances, primitive graphics

MULTIDIMENSIONAL INFORMATION VISUALIZATION TECHNIQUES - Evaluating a Taxonomy of Tasks with the

Participation of the Users

313

5 DISCUSSION OF RESULTS

AND CONCLUSIONS

The two experiments reported were designed to

identify tasks performed by users during the data

analysis and exploitation, and verify that

experimental procedures would be more appropriate

to evaluate the classification of tasks consistently.

Thus, in each experiment were different profiles of

users (actual and experimental) and fields (data set

and contexts of use). In common, several techniques

were used visualization of multidimensional and

different implementations of the same techniques.

Tasks the user could be observed through the

two experiments. However, by the Experiment 2 it

was possible to observe greater number of tasks,

number and type of occurrences of each task,

certainly, because of the type of user involved and

the procedures adopted in this experiment.

The experiment 1 was based on four questions of

analysis proposed by the evaluators to users.

Already the results of experiment 2 were obtained

with a considerable body of 23 high-level analytical

questions, formulated by the user through the

process of exploration and analysis of their data.

Still, despite the different situations in terms of area

and issues of analysis, tasks were not detected in the

classification proposed not only new occurrences of

the same tasks. Moreover, except for "infer" in

experiment 1, all tasks of classification were

observed in real situations of use, which indicates

that they are necessary for the performance of the

analytical process by users.

However, tests of interaction and case study

showed that different results can be obtained when

actual users (experts in the field of data ) are

involved in the assessment in comparison to

experimental users (not specialists).

This study is an effort to systematize the process

of evaluating the usability of visualization

techniques, whereas part of this process should be

focused on the specification of tests of interaction

covering the diversity of tasks that users of this class

of systems must perform.

As future work is to conduct experiments with

new procedures based on field studies, as used in

experiment 2 and strongly suggested in

(Shneiderman, 2006), including other areas, so it is

possible to identify high-level analytical issues that

address consequently, different tasks and using

different techniques for viewing data, and different

implementations of the same techniques for viewing.

REFERENCES

Amar, R., Stasko, J., Knowledge Task-Based Framework

for Design and Evaluation of Information

Visualizations, Proc. IEEE Symposium on Information

Visualization (2004), 143-149.

Amar, R., Eagan, J., Stasko, J., Low-Level Components of

Analytic Activity in Information Visualization. Proc.

IEEE Symposium on Information Visualization (2005),

111-147.

Hoffman, P.E. Table Visualization: A formal Model and

Its Aplications. Computer Science Dept., University of

Massachusetts at Lowell, 1999 (Doctoral

Dissertation).

Stasko, J. Evaluating Information Visualizations: Issues

and Opportunities. Proc. BELIV’06 – Beyond Time

and Errors: Novel Evaluation Methods for Information

Visualization - a Workshop of AVI’06, (2006), 5-8.

Seo, J. and Shneiderman, B. Knowledge discovery in high

dimensional data: Case studies and a user survey for

the rank-by-feature framework. IEEE Trans. On

Visualization and Computer Graphics 12, 3

(May/June, 2006), 311-322.

Shneiderman, B., The Eyes Have It: A Task by Data Type

Taxonomy for Information Visualization, IEEE

Symposium on Visual Languages, 12, (1996), 326-343.

Shneiderman, B., Plaisant, C. Strategies for Evaluating

Information Visualization Tools: Multi-dimensional

In-depth Long-term Case Studies. Proc. BELIV’06 –

Beyond Time and Errors: Novel Evaluation Methods

for Information Visualization - a Workshop of

AVI’06, (2006), 38-43.

Springmeyer, R., Blattner, M. and Max, N.L., A

characterization of the scientific data analysis process,

Proc. IEEE Visualization’92, IEEE Computer Society,

(1992), 235 – 242.

Weherend, S., Lewis, C. A Problem-oriented

Classification of Visualization Techniques. Proc.

IEEE Visualization’90. IEEE Computer Society

(1990). 139-143.

Zhou, M., Feiner, S. K., Visual Task Characterization for

Automated Visual Discourse Synthesis, Proc. CHI’98

Conference, ACM Press, (1998), 392-399.

KDIR 2009 - International Conference on Knowledge Discovery and Information Retrieval

314