Social Conversation Monitoring platforms generally use keyword matching algorithms to decide sentiment, a few also leverage natural language processing (NLP) techniques. Crimson Hexagon has gone a different route, leveraging algorithms created by Gary King to essentially convert text snippets into binary information that is faster,and some would argue more accurate, to search. Most companies claim 75-80% accuracy with NLP, Crimson claims 97% accuracy with its algorithm.

Who is Crimson Hexagon?

They are headquartered in Cambridge, Massachusetts, not far from me actually.  Gary King created the core algorithms used by Crimson Hexagon, algorithms based upon his work at Harvard University. As is the case with dozens of companies of in this space, they have been around for only a couple of years, quietly focusing on building great technology, investing little in marketing efforts.  However, even without a focus on marketing they are getting buzz, check out this video from CNN’s John King.

What excites me about Crimson Hexagon, and about the direction this market is going, is a focus on taking the large volume of data and turning it into actionable information.  Even a relatively small brand could have thousands of mentions across Twitter, Facebook, blogs, forums, etc, pouring through that to find the important data points is daunting. Crimson focuses on making that a little less challenging by identifying relevant information, I’ll tell you more about their approach as we go along.

How much will this cost me?

It’s not cheap, but neither is rent in Cambridge and someone has to pay the bills, right?  The core platform costs $25K per year, plus charges for each monitor.  Your average company that leverages 25 monitors will be paying about $150K per year (no charge for named users).  Crimson does give away its Buzz product for free, however, and its basic keyword search capabilities seem solid.

How does it actually work?

At a high-level, the approach is roughly:

• Define objective.  The only reason to use these tools is to support corporate or agency goals, to make sure that the strategies and tactics you are employing are resulting in the changes you need, or to inform decision-making before making changes in direction.  You do know your goals, right?
• Define filter.  Define, using simple keywords or robust boolean logic,  a set of filters that make sure you are looking at the most meaningful information.  For example, want to find out information to help you decide if Tiger Woods is a practical brand spokesman?  Filter out the noise (e.g. commentary about his PGA video) games to focus in on the relevant conversations.
• Define the questions to ask.  Simple enough, right?
• Calibrate algorithm.  This is where Crimson earns its money.  To train the system, for each question, you literally walk through social mentions and decide which class it falls into (or skip it).  For example, if you question is “Has Tiger Wood’s brand been tarnished” you find tweets, blog posts, etc., that are examples of these types of messages.  The more messages you define as examples of each category/question the better the algorithm will do.
• Ongoing monitoring.  Get results, make decisions, grow your business.
  • Note that Crimson Hexagon attempts to only show you relevant mentions by default.  It is much easier to review 1000 relevant mentions vs. scanning through tens of thousands of mentions.

A few more details

• The algorithms are language-independent.  You train the system in a specific language and it performs its magic against the data set regardless of the source language.
• Information is not translated on the fly, however.  If you have trained a monitor in Hebrew it will not find mentions in English.
• A point of confusion.  When you review results you are shown the results of two algorithms.  The summary information is using the “secret sauce”, the detailed data simple keyword searches.  This can lead to some confusion as you may actually see posts/tweets that were not identified by the magic formula when reviewing the detailed data set.  Crimson Hexagon includes indicators next to each items (red, yellow, green) to show the likelihood that a piece of data was included in the main formula’s algorithm.  However, I still felt this could be confusing for some users.

I do recommend checking these folks out.  The tools are impressive and, while pricey, could be a good addition for those companies/agencies looking to go deeper in their analysis.

John

Republished with author's permission from original post by John Moore .