Textual Genre Analysis and Identification - Ambient Intelligence for Scientific Discovery - page 142

examined Edison's personal papers, by contrast, he relied instead on his rhetor-
ical intuition to sort texts by genre within an archive containing a wide variety
of genres.
Large-scale bibliographic databases pose challenges for these traditional
methods of identification. ProQuest's ABI/INFORM Global database, for in-
stance, provides access to the texts published in over 1800 business periodicals.
For the rhetorician interested in a particular genre, this abundance can be highly
problematic. Over the past few years, for example, one of us has been interested
in the rhetoric by which personal digital assistants (or PDAs) and other mobile
technologies have been circuated through culture [15]. One particular genre of
texts, the technology review, was of special interest. A search in the ProQuest
database showed, however, that between 1996, the year one of the earliest and
most successful PDAs was introduced, and 2002, 1,988 articles appeared contain-
ing the term “personal digital assistant” in their citations or abstracts; 1,472 of
these were available with full text. Furthermore, a few hours of perusal were
enough to show that only some of these articles were actually tech reviews. Thus
neither selective search nor rhetorical intuition was robust enough to identify
the tech reviews in this archive. Could DocuScope help?
To answer this question we undertook a procedure called discriminate analysis.
To build the sets of tech reviews from which DocuScope would build discriminate
functions, we began with an existing set of 34 reviews of mobile technologies by
Stuart Alsop [16] and added a set of 78 by a second tech columnist, Stephen
Wildstrom. Alsop wrote the tech review column for Fortune from 1996 to 2003.
Wildstrom served as the tech columnist for Business Week from 1995 through
2003. Combined, this gave DocuScope a set of 103 tech reviews dealing with
mobile technologies, tech reviews identified through the traditional method of
selective search. To provide a comparable set of articles that were not tech re-
views, we selected 52 articles from the ProQuest database that mentioned PDAs
across the same time period (1996 through 2003). Rhetorical intuition told us
these articles were not tech reviews. To insure a wide range of features occurred
in these so-called non-reviews, we made sure to include some of all the major
types of texts we had seen mentioning PDAs, in particular:
- articles about specific companies associated with personal digital assistants
(17),
- articles about industry trends in personal digital assistants, (15)
- articles discussing an application of the technology of personal digital assis-
tants, (8)
- articles profiling specific CEO's in companies associated with personal digital
assistants, (5)
- articles reporting academic research on personal digital assistants (4), and
- articles providing social commentary on phenonmena like time management
using personal digital assistants (3).
The multivariate weights for the two discriminate functions built from DocuS-
cope's analysis are shown in Table 4. As this shows, only some of the 146 vari-
ables available come into play in discriminating the genre of tech review from