Nobel prize laureate Herbert Simon once popularized the term “satisfice”. Simply put, “satisfice” means “make something good enough.” It’s what we do when optimizing — making something the best it can be — costs too much in terms of time, money or other resources.
I was reminded of this concept the other day while experimenting with Twitter search. In particular, I was attempting to figure out how the Twitter search “sentiment analysis” works. And the reason I was doing that is that I was struggling to understand why my screen name, “znmeb”, was returning a 6 percent “negative” sentiment rating.
So, why does a nice guy like me have a negative sentiment rating of 6 percent? As far as I can tell, It looks like Twitter search determines whether a tweet is positive, negative or neutral solely by the presence or absence of smileys! That is, if a tweet has a 
or or 
, it will be counted as a negative. If it has a 
it will be counted as
positive, etc. Incidentally, if it has a “?” it will be counted as a question. That, apparently, is the scope of its “sentiment analysis” capabilities.
OK … that’s part of the story … people who put frowny-faces in tweets about me will give me a negative rating. So … who are the folks frowning on yours truly? Well, in the words of Pogo, “We have met the enemy and he is us!” It turns out that most of the tweets counted were tweets I made! So, if I frown on Apple, Twitter search frowns on both Apple and me. If someone else frowns on AT&T and addresses it to “@znmeb”, Twitter search frowns on AT&T, me and the sender.
To test this, I posted a tweet that said, “I hate Hoover vacuum cleaners — they really suck 
”. Now that’s an example of a sentence more or less deliberately designed to embarass any natural language processing sentiment analysis algorithm. As you probably have guessed, Twitter search counted it as a positive reference to both me and Hoover.
Why does this matter? Because it looks to me like many — perhaps most — of the social media monitoring tools available use Twitter search as their primary way of gathering input data. If these tools are passing on the sentiment analysis provided by Twitter search without any further processing, they may misrepresent the true sentiment about the search target. And that in turn may foster errors in decisions made using the dashboards these tools provide.
So … I put the question to my readers: “Is Twitter search good enough?”
