We didn’t look at that in this study, but I think that would likely provide more clues.

I’d love to see this reversed in an attempt to identify real factors or false factors. i.e. Start with an established twitterer and do a real personality profile, then look for key indicators in tweets which support the personality profile and see if those hold/stack across a sample.

I’ll look into the resources you posted.

Additionally i definitely agree with Davien (regarding the context issues) and I think its a massive challenge trying to automatically/automagically write things for em. However with that being said its also a case of ‘judging a book by its cover’ and sadly, that is the world we live in (ask me, I had a Mohawk). People make massive leaps in assumptions all the time and make decisions based on those (we know who you are marketing/advertising people).

Davien, we agree (mostly/generally). Over a sample set like our and by making the methods open public to scrutiny, our hope is that this will generate future research and much more discussion. Rightly or wrongly, people are making judgements of others based on Social Media, one of the things we want to do is really examine how justified they are. There are a number of associate questions regardless of how justified they are, such as an individuals right to privacy, not to mention the discrimination issues.

However, it wouldn’t be surprising to find that people who engage in Twitter rather than Read-Only are slightly more socialable (Extraverstion) when viewed as a group? Our previous post does a good job (we hope) of highlighting the potential for error in these sorts of judgements. This post is here… https://onlineprivacyfoundation.org/just-how-accurate-is-facebook-personality-prediction/

The excellent book “Virtually You” also does a great job of highlighting how different people can be (or appear) in different contexts. My Twitter self may be radically different from my work and home self for instance.

Appreciate the discussion and look forward to more.

For example, number of followers vs. number of followed could be an expression of newness to Twitter, or to dedication (or not) in blocking bot follows.

Lack of engagement (replies to conversation threads) could mean you use Twitter read only, not that you’re aloof.

Retweets vs. conversations could be a reflection of an account used for news aggregation or business promotion rather than a measure of narcissism.

The words used will even have some skew based on geographical origin, cultural norms, or even education. There are probably some conclusions you could make related to those factors moreso even than personality.

I guess I would say anyone attempting to gather this data or conduct this analysis should be very cautious about how these conclusions are applied.

Andrew. Many thanks, these are great (and do feel free to add more when you think of them).

It’s safe to assume that a prize will be winging itself to you (in the future).

From the tweets:
* Number of shared followers between target and targets friends, ie Andrew is following James and Chris and Chris is also following James (is it circular), this shows more community based network perhaps indicating likability/social skills rather than a straight follow relationship

* Number of links to ‘self image’ related work – is the user only interested in posting things about him/herself or are they also sharing various other things they find interesting.

* With RT’s and similar links, are they posting from a small subset of the users they follow or does it cover many of them – determining whether they are ‘friends’ with the people that follow them

* Conversation lengths – does the person have multiple to and froms with people (sociability?), what sentiment can be extracted from this (are the conversations mostly agreeing or disagreeing) – possibly confrontational/submissive personality

* How often does the user change his/her handle/alias – what similarities happen prior to this (confrontation, slowing of updates)

* How often does the person tweet, are the tweets looking for responses? (high/high could indicate a need for attention or high/low a more arrogant streak, low/high or low/low could indicate a producer )

* Number of people mentioned in tweets who do not follow the person could indicate attention seeking

* Hashtags that match trending topics (trending both within friends/followers, locally, and internationally) – following the crowd?

* Retweets of company accounts vs people

I think some other things you can look at on twitter are the things that touch on the real world:

* What times is the user tweeting – office hours? Weekends? Saturday nights?

* Where are they tweeting from – phone,pc? Correlating this with the times

* Do tweets at different times change, for example does the person mostly speak to woman at night, RT at night, link to self promotion things at night?

I’m sure I could think up a few more but soon i’d be taking on a post the same size as yours :/

Gabe –
– I like producer/consumer angle.

– For sentiment analysis, we will be running Tweets through James Pennebakers LIWC program.

– The hashtag thing is interesting for sure, feel free to submit pseudo code to tackle that. It could be done compare the hashtag in a tweet with hashtags taken out of a stream of data. DataSift’s new historical search could be very interesting here.

– There is a verified flag in user information, so that’s a doddle (english for easy). 🙂

– ReTweets celebs…yep. Also interesting.

Thanks Gabe, some good stuff. That hashtag one is a good one. Might also be useful in detecting spam bots