Introducing the New TweetPsych and TweetPsych for Lists

After I first launched the Twitter psychological profiling tool TweetPsych, some of the most common feedback I got was that it was hard to understand the results. So I designed a new reporting mechanism and design to solve that problem. The new TweetPsych uses “meta dimensions” which are combination of related factors from the two linguistic algorithms (RID and LIWC) the application uses. Each of these comes with a description and is represented on a bar graph. Each user’s profile is compared against the average user and the report explains which dimensions occur more or less frequently than the average.

I also launched a new feature for the site. TweetPsych for Lists allows you to do the same kind of psychological profiling, but of entire lists. Curious to know what the inside of Zappo’s employees’ heads looks like? Here you go.

If you have other ideas on how to make TweetPsych even better, let me know.

If you liked this post, don't forget to subscribe to my RSS feed or my email newsletter so you never miss the science.


Tom Moullet January 4, 2010 at 11:11 pm

Very awesome. If I could make one change, I would make it so that when a trait goes negative (ie 70% less than average etc) have the tense/tone of the description take into account the absence of that trait instead of having the trait descriptions remain constant.

John Paul January 4, 2010 at 10:59 pm

Very cool tool.. don't know if I wanna know what it tells me :)

PhilRichards January 5, 2010 at 10:02 am

Thanks Dan, I enjoyed using this and think that it has the potential of being a great help.

I need a little more help on interpretation though…. My report said:-

“This user Tweets about anxiety 46% less than the average user. This includes uncertainty, nervousness and apprehension. It may indicate a stress and fear. “

So does that mean I am exhibiting stress and fear 46% less than the average user ?

It is probably obvious to you – but I fail to be clear what the system is telling me.



znmeb January 5, 2010 at 3:16 pm

I stumbled across “” a couple of days ago. That appears to be a “pure LIWC” approach. I was disappointed in its results. I'm curious if you think adding RID makes any difference.

I'm skeptical about TweetPsych and AnalyzeWords for two reasons:

1. Tweets aren't sentences. They *can* be, but are often merely links to larger bodies of text outside of Twitter, not necessarily written by the tweeter. For example, a blogger will often post a link to a blog post. To really analyze the tweet(er), you'd have to follow the link, determine that the tweeter was the author, and analyze the blog post. I suppose you could filter out all of a person's tweets that weren't stand-alone sentences and those tweets that contained links, but …

2. Tweets aren't English. There's no English equivalent for a hashtag, an @reply, or retweets either using the built-in retweet button or manual editing.

In short, tweets are a new (meta)language, with syntax and semantics evolving in near real time. That's going to pose a challenge for those who would automate the extraction of meaning from the tweet stream. It is a challenge I see being addressed, and TweetPsych seems to be a piece of the puzzle. But we aren't there yet.

Bryan Bliss January 5, 2010 at 3:47 pm

I Like the idea and i'm interested in the science of what you've put together here.
The zappos report seems like a pretty thorough analysis but when i consider some of the things i use in psych profiling.
It may take some noodling and analysis on my part but i'm missing some key ingredients that i consider.
what about generosity: I look to see how generous in spirit in action a person is. Generous in an impactful, empowering and resourceful way.
i look to see if someone shares resources of value, answers questions, leverages his network to help others make connections, inspires and encourages others to give and share.
Richard davis of U Wisconsin has done some fascinating studies about generosity, empathy brainwave activity and meditation, cool stuff you will like Dan.
what about Listening and responsiveness?
I look to see just how “tapped in” the person is to their community. how proactive and responsively are they communicating. its quite a different thing to Respond only to inquiry or conflict yet another to initiate conversation designed to enlighten and resolve an issue.

some of my most insightful profiling data comes from looking at their media preferences, both modality of media and specifics like do they like Bach or Billy Ray Cyrus?.
Demographics of links tell alot, hobbies, politics, you get the picture.

Thanks dan for your insight and work and i'll take a closer look at how this all works
Bryan Bliss

bkjrecruiter January 5, 2010 at 3:56 pm

Dan- Great post… I enjoyed reviewing what I already know… I tweet about things that are important in the business world…. Keep pressing… Brian-

Rebecca Leaman January 5, 2010 at 5:31 pm

Dan, this new incarnation of TweetPsych is a great improvement! Not only does it seem to be presented in a way that most of us should be able to get our heads around a bit better, it certainly addresses the question about how to interpret the numbers. I can't comment on the science behind it, of course, but suddenly TweetPsych is looking much more like a useful tool to help people see their Twitter activity in a new light. Thanks for what you do. I'm looking forward to playing around with TweetPsych and TweetPsych for Lists. January 6, 2010 at 10:09 pm

Nice tool, thanks for sharing

dana allen-greil January 6, 2010 at 6:58 pm

Dan- I wish there was a way to get more information on how determinations are made for each category. For example, I manage the @amhistorymuseum account. Somehow, TweetPsych determined that we tweet about the past 41% less than the average user. !!?!! 5-10 times I day I tweet historical facts with this format: “Today in 1920: NY Yankees announce purchase of Babe Ruth from Boston Red Sox for $125,000. Autographed Ruth ball:“ So I thought, maybe it will think I talk about the present a lot because I'm always using “today”… but no, TweetPsych says we tweet about the present 55% less than the average user.

It seems like it could be a really fascinating tool but I'm very skeptical of the accuracy… Thoughts?

Mark McCulloch Success Coach January 9, 2010 at 2:56 pm

Great blog with top quality content.

Everytime I visit your blog I do enjoy my time as the quality of your content really is second to none and I will be back again soon for some more.

Great tool by the way.

Mark McCulloch

tom deeter January 12, 2010 at 4:10 pm

Very interesting app! How about a feature that shows profiles/lists with similar TweetPsych scores once the data has been crunched?

erikabarbosa January 13, 2010 at 2:34 am

Interesting stuff – thanks Dan! It would be great if the tool provided some examples that attribute to the percentage as well.

{ 2 trackbacks }