Sentiment Analysis: How Do People Feel About You On Twitter?
November 4th, 2009 (1:00pm) Dawn Foster 7 CommentsTweet This (168)
Do you know how people feel about you on Twitter? Are the tweets about you or your product mostly positive, negative or neutral? While things like feelings and sentiment can seem fuzzy and “touchy-feely,” there are sentiment analysis tools available for Twitter that attempt to classify tweets into either positive, negative or neutral categories automatically using algorithms and lists of keywords. For example, using words like “sucks,” “sad” and “hate” would be classified as negative, while “awesome,” “great” and “love” would be positive, with a neutral rating given to anything not falling into one of the other two categories.
As you can probably guess, the results from sentiment analysis tools vary widely, with many tweets ending up in the wrong category. Each tool is only as good as the list of keywords and the algorithms it uses, and they are easily confused by imperfect human beings who send mixed signals into their algorithms. For example, this tweet from @PDXrox was classified by one tool as positive and negative by another, based on confusion over using both “dang” and “:(” while also using the word “love:”
“Dang! Both @backfencepdx and @igniteportland have scheduled their events for the same night – Nov 19th. I love them both – wah! :-(“
You have a few options for sentiment analysis. If you are doing work for a company with budget for analysis tools, you should seriously think about purchasing a tool like Techrigy’s SM2, Scout Labs or other social media monitoring packages with sentiment analysis, since they have more robust features than some of the free tools. For those of us doing personal monitoring or working at small companies with limited budgets, there are some free tools that you can use to get at least a rough idea of how people feel about you or your products and services.
twendz
twendz is certainly the best free tool that I used, which isn’t really surprising since it was created by Waggener Edstrom, a large public relations firm with a vested interest in having accurate sentiment analysis for its clients. It handled my complex search query with ease when other tools did not, and it constantly updates in the browser window with new tweets. It also has some really nice features, including sentiment summary per tag and the ability to see exactly which tweets are positive or negative overall or for each tag.
twitrratr
I have some fundamental issues with twitrratr: It doesn’t handle complex searches; it doesn’t pick up new tweets very quickly and it doesn’t have additional analysis by tag or other parameters. However, I like how you can easily see at a glance the tweets that are positive, negative and neutral with words highlighted to show you why the tweet was put in the positive or negative category.
As I mentioned before, these tools aren’t perfect. Any time you are dealing with human beings and our imprecise languages, there will be plenty of opportunities for putting tweets into the wrong categories. However, sentiment analysis does provide a starting point and a rough idea for assessing how you people feel about you or your product.
What have you experienced when using sentiment analysis tools?
Web Work 101: How To Escape The Cubicle
That’s just strange. I’m not generally a fan of Twitter, but some developers are coming up with some really cool ways of using it.
Social Mention also provides sentiment, with other metrics such as passion, strength and reach, across all social media including microblogs. I’m still not absolutely convinced these are useful. I tested Twendz when Jade Goody died, and tweets such as “So sad Jade Goody died” were classed as negative when, to me, they should have been positive.
This is an excellent and timely article. I also want to bring to your attention the first and only truly semantic search engine that currently works on Twitter data, TipTop, now available in a beta version at http://FeelTipTop.com TipTop’s powerful engine understands each and every message on Twitter just like a human being would. As a result, it can discover from within the data the very best tweets organized nicely along a variety of categories and concepts learned dynamically. In fact, the entire platform learns from data as data flows through the engine. You can now see in real time the sentiment associated with anything in the world that people are talking about. Please give it a try.
IMO http://tweetfeel.com has been the most accurate free tool. They even just recently launched http://tweetfeel.biz which is subscription based and takes into consideration, far more complex analysis.