The brands tracked by Tweet Satisfaction are heavily weighted toward consumer and enterprise software, internet services, computers and electronics. This update expands coverage to include travel, education and personal finance.

Tweet Satisfaction will continue to expand coverage of brands based on the interest of current beta users. New brands will include vehicles, airlines, hotels, and select companies from TechCrunch 50.

In the end, it improved the signal-to-noise ratio of happy and mad tweets about brands. But, it wasn’t an easy decision.

Why are : ) and : ( poor indicators of emotion or sentiment?

The emotional response to brands (positve or negative) is difficult to infer from 140 character messages. When Tweet Satisfaction started, the library of “happy” and “mad” match terms was pretty small. We decided to include emoticons or attitude to increase the volume of happy and mad tweets.

Soon emoticons like : ), : D, : ( began to dominate the Twitter analytics for popular brands. During the iPhone 3GS launch, we had dozens of messages like “UPS confirmed delivery of my iPhone. I can’t wait to get home to try it! : )” and “UPS confirmed delivery of my new iPhone. I wish I was home so I can try it out! : (“

Tweet Satisfaction‘s natural language processing engine would interpret both messages as happy tweets if we ignored the : ) and : ( symbols. But with these symbols, how would a reasonable person interpret the sentiment? The first message is  happy. But, the second message is debatable.

In other cases, the : ) was used in messages where someone  had a bad experience with a product, brand or particular store. “After arguing with the sales clerk at XYZ, I gave up and decided to dispute the charges. We’ll see who has the last laugh!  : )”

Finally, the attitude symbols were often used in messages that had no actionable information about why the person was happy or mad.

What does this mean for you?

Tweet Satisfaction‘s Twitter Analytics will now return fewer “happy” and “mad” tweets about a particular brand. It will be easier to find actionable insights to improve product, customer and brand experiences. This is especially true for brands that get lots of buzz on Twitter.

If people rarely tweet about your favorite brands, this change will make it harder for the brand to rise on Tweet Satisfactions Popular Picks list. If you manage one of those brands consider adding viral features the product or website. ShareThis has a nice widget that might help.

We set out to answer two questions: 1) What good and bad things were written about “xyz” on Twitter? 2) What is the trend of  “happy” and “mad” tweets about “xyz” over time?

Like many seemingly simple questions, mining the data to find an answer was not easy.

• Twitter‘s character count limit encourages users to save time and key strokes. This produces many 140 character run-on sentence fragments and creative ways of shortening proper names.
• Twitter‘s search API limits the number of tweets returned per API call  and does not return search results beyond an every shrinking time frame.
• Most tweets about “xyz” are neutral. Most tweets (roughly 95-99%) share links to news and announcements, post questions, or provide updates on what someone is doing, did or hopes to do.   Tweet Satisfaction is only interested in the rare tweets about “xyz” that include “happy” or “mad” emotions.

Working nights and weekends, our small team experimented with a variety of solutions to these challenges. Some worked better than others, and we’ll continue to refine the algorithms that identify whether a tweet is about “xyz” and whether the tweet is “happy” or “mad”.

The biggest challenge (other than attracting a large number of happy users) is avoiding the temptation to add more features.

There are a number of enterprise solutions that attempt to answer our questions and do a lot more. We view the best of these solutions as potential strategic partners (not competitors). For example, Scout Lab helps a few hundred enterprises use “social media to build better products and stronger customer relationships.” You might also consider Radian6 or Sysomos if you have the need and budget.

To stay focused on simplicity and value, we draw inspiration from Survey Monkey,  Basecamp, Highrise, Mint, and DropBox.

We hope to follow in their footsteps and benefit several thousands (or millions) entrepreneurs, small businesses, journalists, bloggers, analysts, policy makers and – last but not least – people who are simply curious.

Imagine the perfect storm for a growing start-up. The company’s first iPhone App was released several days after a major upgrade to the online platform. At the same time, it launches massive search engine marketing, social media engagement, and other viral marketing campaign. The company was just featured in the Wall Street Journal and several leading blogs. Oh yeah, and it’s the peak sales season.

If this was your company, would you grab a surfboard and ride the wave? Or would you baton down the hatches and hope it blows over without major damage?

In this case, the company got serious about Twitter analytics. It hopes to make the most of the approaching storm.

“My biggest concern is not that these initiatives fail to deliver the number of new accounts we project.” The CEO explains, “My biggest concern is that we hit our growth numbers and all of those new customers a unhappy with the experience.”

The EVP of Marketing  asks the Customer Delight team to do more. “We have great reporting on support ticket volumes and categories  from our call center. But we need more insight into unhappy customers that turn to social media instead of calling support.”

At first, the company tried to use desktop applications like Seesmic and TweetDeck. The Customer Delight team was, “more than happy to delight the customers who contact us. But, we aren’t set-up to manually sift through all the tweets to find and address customer concerns.”

The company used the alpha version of Tweet Satisfaction to pinpoint customers that were “tweeting mad” and escalate the issue to support, quality assurance, product development

“We hope twitter analytics would do more than measure the buzz for marketing campaigns,” says Sean Brady, Product Guy at Tweet Satisfaction. “We hope it can be an early warning signal that identifies issues with new product releases, server outages and customer experiences that under perform expectations.”

“We also hope that the entreprenuers and start-ups that rely on Tweet Satisfaction are successful and reach a scale that requires an enterprise class solution.” Sean continues. “Scout Labs, Radian6 and Sysomos provide robust solutions for many more forms of social media (not just Twitter). They offer a valuable service for companies with budgets for social media monitoring.”

In March, Tweet Satisfaction began storing tweets that matched certain keywords. At the time, the Twitter Search API would return tweets for the last 3 months.  Searching in the online application would produce similar results.

Today, the both twitter search methods only produce results for the last 21 days.

Why reduce the search “look back” period by an order of magnitude?

In a word…   Oprah. In two words… Swine Flu.

Twitter has gone mainstream (well… technically it started it’s leap across the chasm to attract the early majority). As the volume of tweets (and searches) accelerates, Twitter is likely taking preventive steps to avoid the dreaded fail whale and unscheduled downtime.

The search limitation probably isn’t a problem for 98% of twitters users. [Update: many readers would like to find old tweets for links to articles or find a movie/club/restaurant review].

“On a personal level, I have trouble keeping up with a handful of CEOs, bloggers and friends. And, I don’t really care if your son beat you at xBox last Christmas.

But, this represents a huge loss to policy makers, marketers, product managers and entrepreneurs who hope to mine Twitter for competitive intel, customer satisfaction, public sentiment, or other business insights.

That leads to a bigger question: What’s Twitter doing with the data?

Is it throwing it away like yesterday’s garbage? Or, will it be charging a fee to data mine your tweets?

Post comments with your theory of the case.

PS> I tweeted this post.