Shuang Wu

1
Name That ClusterText vs. Graphics

• Shuang Wu
• REU-DIMACS, 2010
• Mentor James Abello

2
Talk outline

• Project description
• Our research project
• Input time data recorded from the
  Name That
• Cluster web page.
• Output statistic results of
  participants different
• behaviors under
  corresponding situations
• when using
  computing interfaces.
• Main challenges
• a. formed the data for
  statistics analysis.

3
Description

• There are three different interfaces shown to
  users under each query Textual ( ) ,
  Graphical ( )
• and Hybrid ( ).
• Generally, the process consists of selecting,
  exploring, and finally rating and naming
  clusters.
• The interface gives detailed instruction about
  its usage.

4
The use evaluation of clusters proceeds as
follows

• Step 1 Choose a phrase as well as one of the
  interface
• buttons next to it.
• Step 2 Select a group of related phrases.
• Step 3 Inspect the contents of the selected
  group of
• related phrases.
• Step 4 Enter answers about the selected group,
• which includes,
• a. Group description
• b. Group relevance
• c. Description relevance
  

5
Project Purpose

• One of the most interesting things in this study
  is that users are given the choice of three
  different interfaces textual, graphical and
  hybrid.
• Currently, most clustering are shown in a pure
  textual form. However results (or search engines)
  to enhance current interfaces with graphical
  representations.
• Our main task is to get a statistical analysis
  based on the data we collected from the Name
  That Cluster web page, to see whether
  participants exhibit different behaviors when
  using different interfaces.

6
Project process

• Order the raw data for each query.

• Ex. Timing Information
• Note T, G, H identify the three interfaces.

• The numbers identify
• the time for a user to
• evaluate either T, G or
• H.

7
We will transform the raw data into table below.
The main process is to get triples of users that
evaluate a query with the three different
interfaces, and leave the leftovers at the end.

• Note leftovers are the remained data from T, G
  or H after grouping in triples.

8
2. Initial analysis

• After having the three groups of users of the
  same size corresponding to an interface per every
  query.
• We will perform statistical analysis. To answer
  questions like
• Which interface is preferred by participants
• Peoples variety of behaviors per query
• The max, min, and average number of queries
• evaluated per user
• etc.

9
Challenges

• Need to have meaningful statistics.
• Leftovers
• Because the number of triple groups (TGH)
  depends on the minimum numbers of triples over
  all queries, we need to actually find this
  minimum number. In order to minimize the
  leftovers and support the most effective
  statistical analysis .

10
Conclusion

• We expect to get statistic results from this
• project, in order to determine what would be
• the preferred interface.

11
References

• Name That Cluster Text vs. Graphics, J. Abello,
  
• B. Gaudin, C. Tominski, H. Schulz.

12
Thank you

• THE END