Yes Qualrus is an exception. I didn't know exactly what was behind the
Qualrus feature to suggest new codes, but it was clearly some form of
machine learning algorithm.
Ed Brent wrote:
>I think I might part company with Normand a bit regarding the analysis
>tools. He is right that existing data mining software packages rely heavily
>on statistical tools such as cluster analysis, multidimensional scaling, and
>correspondence analysis. But I would argue those tools may be more useful
>for quantitative data than they are for textual data.
Do you mean numeric rather than text data? If yes, then I would have to
disagree since all those tools have already showed great potential for
analyzing text. You only have to look at the research area of automatic
text categorization to see that many statistical and machine learning
algorithms are used successfully. Of course, new algorithms are being
developed specifically to analyze textual data, such as support vector
matrix or latent semantic analysis, but comparison between the newer
algorithms and the older one shows that those old statistical algorithms
fair pretty well, and sometimes even better than newer ones developed
specifically for the analysis of textual data.
As for the analysis of manually assigned codes, I have made the bet that
those tools would also be useful to the analysis of codes so I included
those tools in the new QDA module of WordStat 4.0. You can get a lot of
information with them that would be difficult to obtain otherwise unless
you export your qualitative data to a statistical program. For example, you
can find easily which codes co-occurs frequently, you can quickly find
which people are similar in terms of the relative importance of various
codes and identify empirically subgroups of people. You can also quickly
identify differences between groups of individuals and map those
differences using correspondence plots. I see no reason why data mining
algorithms could not be used with qualitatively coded documents to find new
relationships, explore data, etc.
I must say that it is too soon for me to tell whether qualitative
researchers will like those features and find them useful. I did not came
with this idea after interviewing numerous qualitative researchers and
identifying their expressed needs, but from my own vision on what I would
like to have in a QDA software. Will researchers or software developers
follow me in this direction, I currently have no idea. I would be glad to
hear comments from qualitative researchers about this issue
([log in to unmask]).
Regards,
Normand Péladeau
Provalis Research
www.simstat.com
|