Nonnegative Matrix and Tensor Factorization for Discussion Tracking - Text Mining: Classification, Clustering, and Applications - page 106

3−way Results

1

0.8

Weekly pro & college football
betting pool at Enron

0.6

0.4

0.2

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

4−way Results

1

From the list of 197 recipients
3 individuals appeared one week

0.5

0

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Month

FIGURE 5.4 : Weekly betting pool identified in the three-way (top) and
four-way (bottom) analyses.

but not any examples that spanned some number of days. Here we present
several examples of the latter type and also show what is gained in going from
a three-way to four-way analysis.
Figure 5.4 shows a series of email messages announcing the results of a
weekly betting pool based on the number of winning teams chosen correctly
out of all pro and college football games for the week. Most of the top terms
were names, but after a dozen terms more interesting terms, such as games ,
score , picked ,and prize , start to appear. Each email lists all of the names
entered in that week's pool and their record, which explains why the names
appear high in the list of terms for the group.
The unusual feature of this group is that the time profile is so regular. This
is because the discussion took place weekly for one day. Results of the betting
pool were sent out after the conclusion of all games in the pro and college
football schedules.
The four-way analysis identified this discussion but only found a single
spike in time. The group showed that the organizer only sent this message
to four recipients (out of 197 email addresses) in this case. Presumably the
four recipients did not participate in other weeks, and none of the remaining
193 addresses participated in other weeks. If the recipient list were expanded
to include others in the betting pool, then the four-way analysis might have
picked up other days and recipients as well.
As a second example, Figure 5.5 shows the temporal activity for a discussion
involving FERC and its rulings on RTOs. From one of the newsfeeds from
[e-mail] o n May 4, 2001 there was this description:

“For background, an RTO is a regional entity that is designed to
consolidate control and delivery of electricity across various types
of transmission systems within a particular region. The origins
of FERC's RTO policy dates back to its December 1999 Order
2000, in which it strongly encouraged all transmission-owning util-