Database Reference
In-Depth Information
Almost invariably, these representations of sentiment from huge volumes
of social media messages are very shallow. We found that most consist of a
simple set of rules that assign positive weights for positive words, and nega-
tive weights for negative words. This might work for some tweets, but it
can't be considered reliable. For example, consider this sentence:
“The
amazing
number of lip synched performances during the 2012
closing ceremonies
inspired
me to take a walk and have my own lip
performance at the local pub with a pint.”
Even though the two underlined words are considered positive, this tweet
is overwhelmingly negative. What this sentence illustrates is that context is
extremely important, and that any serious analysis of free-form text has to
derive meaning through more sophisticated approaches, as opposed to a
simple lexicon of classification. We've made this point in a couple of chapters
throughout this topic, and this is one of the differentiating pieces of the IBM
Big Data platform.
Moreover, sentiment itself isn't enough. If you work in a car company's
marketing department, you want to know if someone's interested in buying
your company's product. Even better, you'd probably like to see market seg-
mentation details so you can get a breakdown of how people are responding
to your campaigns.
Building on the capabilities of BigInsights, Streams, and the Advanced
Text Analytics Toolkit, IBM has developed a Social Data Accelerator (SDA for
short, but officially known as the IBM Accelerator for Social Data Analytics),
which provides a rich set of text analysis rules so that you can get an accurate
and precise sense of what people are saying online.
The SDA includes extractors that are focused on lead generation and
brand management for a number of industries. For lead generation, the focus
is on discovering potential customers, while for brand management, how
people feel about the brand and any competitors is the focus. In addition,
each set of extractors for a particular use case is divided between rules that
determine the meaning of a social media message, and rules that help to
build profile information for a particular user. To pull the output of these
extractors together, the SDA includes a workflow that helps you manage the
flow of data through these extractors all the way from ingest to visualization.
