Why We Chose
“Facts Before Feelings”
Over the last several years, we
have seen an emergence in companies that aim to mine the net’s
feelings. Labeled sentiment mining by most, the general idea is to
understand people’s thoughts and feedback around a particular subject.
Most of the time, these results
are divided into positive, negative, or neutral feedback. While
this is an important practice, it is something we have always seen taking place
after first measuring the amount of related conversations taking place.
So for the past several years
Trendrr’s focus has been around getting the best possible quantities view
across as many data sets as possible. While it’s great to know what
people are saying, we first wanted to know on which channels people are
discussing these brands, products, etc. and at what volume. By
understanding sample size as well as frequency of discussion, we can achieve a
holistic view of the level of conversation or mind share.
Now that our quantitative data is
aligned and tracking, we are ready to put a qualitative lens around this
information to better understand the tone of the conversations.
While we were working on ‘best in
market’ quantitative views, we were also doing diligence around what the most
effective sentiment algorithm was and who was doing the best work in the
To be honest, I have always
viewed this type of data collection as a big bag of hurt. The accuracy
around the language, and the mining of it, yielded a view that was ambiguous at
The marketplace is demanding it
so we naturally address it in our own product development.
To get an understanding around
accuracy, 70-80% is considered the current standard. As we roll-in new
feature sets like sentiment analysis and predictive views, we want to rely on
who is doing the best work in each area worldwide. With this in mind, we
are reviewing algorithms that is on par or hope to be better than the
current marketplace benchmarks and standard.
The differentiator for us is not
the algorithm. Technology is always commoditized at some
point. As C. Dixon points out in a recent blog post titled: To make
smarter systems, it’s all about the data
breakthroughs come from identifying or creating new sources of data, not
inventing new algorithms.”
We are confident that our
holistic view to key data sets coupled with lenses like sentiment and
predictive trends will provide the greatest insights, despite media (NYT
“Mining the Web for Feelings, Not Facts “) and competitive hype
to the contrary.
Over the next few weeks, we will
demonstrate the integration of these lenses into the Trendrr dashboard, discuss
which algorithm we used and why and demonstrate an overall group trend formula
that takes into account both qualitative and quantitative data sets.
This is our current marketplace
strategy. It is practical and pragmatic which we believe to be two good
character traits of any marketplace analysis. Stay tuned.