14 8
c h a p t e r6 :SOCIAL ANALYTICS, METRICS, AND MEASUREMENT■The challenge with automated sentiment—and one of three primary motivators
for the consideration of workflow tools as a part of your listening program (the other
two being data conditioning and noise elimination, along with scalable and trackable
routing and follow-up) is that “meaning” is almost entirely context driven. One of the
shortcomings of the automated listening tools is that they tend to consider the post—in
which keywords of interest were embedded—in isolation: In other words, they see the
immediate conversation, but not the back-story.
Consider Twitter as a channel: Because of its highly fluid and distributed nature,
the short posts that define Twitter are actually interlinked conversations. One person
says, “I bought a new Dyson—I love it,” and another, seeing that post, replies “Dyson
sucks!” as a humorous affirmation of the product, resulting in the “negative” post refer-
enced earlier. Because replies are not always linked (in the technical sense) to their origi-
nating posts, the listening tool sees “Dyson sucks” in isolation and then applies its rule
base accordingly. Somewhat humorously—or maybe by design—with the launch of the
new Dyson bladeless fan referenced in the prior chapter, a whole new round of sentiment
issues arises for Dyson as posts like “Dyson really blows!” start circulating. Oh well.
More seriously, none of this is a knock on sentiment analysis as an idea, nor is it
intended as commentary on listening tools that include sentiment analysis in particular.
The fact is that the current generation of tools—with all their attendant shortcomings—
still provide more value than not using them at all. The challenge—and your responsibil-
ity—is to ensure that you are, at some level, tying the conversations you discover back to
their original context so that you can actually deduce the intended meaning. This is also
important from the perspective of workflow: the elimination of “noise,” those results
that while they match your keywords are not related to your actual search. Referencing
the Dyson products example, conversations mentioning Esther Dyson are probably unre-
lated. The effort required to winnow the results to those which are relevant to your spe-
cific interests must be considered as an integral part of any listening program.
Here’s a twist on sentiment analysis: You can, if you’re savvy, get your fans to
do at least some of the work for you. Jake McKee, while at LEGO, often turned to key
people within the communities he worked with for insights into “brand sentiment.”
Fans would often let Jake know—via IM, where Jake maintained an active, open pres-
ence with fans—that there was something or other that he needed to pay attention to,
adding within their posts whether this was a good thing needing more attention, or
otherwise. This is one more reason, as if you needed one more, to actively build a base
of loyal and alert supporters.Know Your Influencers
Parallel to traditional PR and the associated marketing and advertising concepts relat-
ing to “influentials,” there are metrics related to the social graph—the connective links,
profiles, and updates connecting people on the Social Web—that define the sources of