Haven't looked at facebook, but there have been a few papers that have tried to predict the big 5 personality traits from twitter.

Golbeck et al. (2011) took 50 twitter profiles, had them complete personality surveys, and tried to train regression algorithms to automatically score them against the big 5. Most of the features came from applying the LIWC tool to tweet content. Unfortunately their best approach was no better than ZeroR (which basically assumes everyone has the same personality).

Quercia et al. (2011) do a larger study with 300+ profiles, using only 3(!) features: counts of you followers and people you follow, and of people who ‘favourite’ you. They claim pretty good results, which seems crazy...how can people's complex personalities boil down to these 3 statistics?

Sumner et al. (2012) have done the biggest study so far, with about 3000 profiles. They applied the LIWC tool to each user’s tweets to obtain the same features as Golbeck, and put them up Kaggle where they got 1,700 submissions and some hugely complex ensemble classifiers. Despite all this effort they got pretty poor results and conclude that the previous papers were a bit misleading about how difficult the task is. A big problem was that previous papers only reported error rates, which hide a lot of problems. Anyway, this paper is definitely a good read.

Another good read is Pennacchiotti & Popescu (2011). They aim for a general purpose algorithm for classifying twitter users, rather than one tailored to the big 5. But they explore a very wide array of features.

References:

Golbeck, J., Robles, C., Edmondson, M., & Turner, K. (2011) Predicting personality from twitter. In Privacy, security, risk and trust (passat 2011) and social computing (socialcom 2011)

Quercia, D., Kosinski, M., Stillwell, D., & Crowcroft, J. (2011) Our twitter profiles, our selves: Predicting personality with twitter. In Privacy, security, risk and trust (passat 2011) and social computing (socialcom 2011)

Sumner, C., Byers, A., Boochever, R., & Park, G. J. (2012) Predicting Dark Triad Personality Traits from Twitter usage and a linguistic analysis of Tweets. In Machine Learning and Applications (ICMLA 2012)

Pennacchiotti, M., & Popescu, A. M. (2011). A machine learning approach to twitter user classification. In ICWSM 2011.

I think what you are looking for are apps that take a person's interests (likes) and applies them to the market place for shopping. They become very popular around the holidays for when people are trying to figure out what to buy their family and friends. I know you are looking for something more detailed than that but that is about the only thing I know so far, besides reading a person's profile manually that gathers information for you quickly. Wolfram also has a very interesting ability as well. I think you will be impressed with that option too.

Here is the app I was talking about. (only one I can find right now for some odd reason, but during the holidays tons of articles come out about holiday shopping apps)

http://www.pickie.com/gifts

Another site you want to look at is

http://www.socialbakers.com - they analyze social media behavior on all the major platforms.

Read this article carefully to see if this will help you

http://mashable.com/2012/09/03/this-app-knows-more-about-your-facebook-account-than-you-do/

To start with take a look at NODEXL http://nodexl.codeplex.com/ it's not specifically designed for this but it has the capability of extracting the data you want. Since it's based on MS Excel it's easy to add your own macros for "doing something" with the data.

So it looks like this is something that will help with what you are looking for. excellent

NODEXL: excellent resource. Excel tends to be way too limited for any serious pattern recognition, and NODEXL is only for Windows and only for (very nice) graphs. I have access to Windows, but it's all Mac for me when I can. But without an SPSS license, there's always the chance I'll be excelling once in a while.

Wolfram Alpha: also an excellent resource, and good fun - corrupting the "findings" manually by inserting facebook data is a good way to reveal to boundaries of these analyses. However, it's still a blackbox and self-centred.

I want something to help me turn many consenting "respondents'" many facebook "samples" into computer-friendly descriptions (the features) that I can plug into the kind of pattern recognition pipelines (like SVM classifiers) that I'm familiar with. I expect to be able to train classifier models where the classes (e.g. "I prefer foreign holidays" versus "I prefer domestic holidays") are reasonably straightforward and taken from (simple or sophisticated) self-reporting questionnaires.

Actually in the first place the kind of personality, values and play-pattern correlates I'm after are laid out in a pilot study I published a while ago if you're interested: https://www.researchgate.net/publication/215448321

And my hope is pretty much described in that paper: to model individuals to predict frustrations, difficulties, comfort-zones etc. when interacting with certain kinds of gamified and group tasks; and thus both experiment with easing group interactions in shared goals, and experiment with implanting cognitive dissonance and other psychologically-activating "tricks" from games design into non-games contexts.

But first I need a simple means to profile users/players with a low entry threshold to participation and that doesn't enforce a group size: hence mining facebook profiles.

Conference Paper Introducing game and playful experiences to other applicatio...

As a benchmark...

...from these initiatives and the social media commentary alone...

http://mypersonality.org/wiki/doku.php

http://research.microsoft.com/pubs/163535/FacebookPersonality_michal_29_04_12.pdf

http://www.pnas.org/content/early/2013/03/06/1218772110.abstract

https://www.onlineprivacyfoundation.org/research/personality-facebook/the-big-5-experiment/

https://www.onlineprivacyfoundation.org/research_/PredictingdarkTriadPersonalityTraitsfromTwitter.pdf

...we get the following suggestions for "human understandable pseudo-features"...

- device, location, activities, status's, interests and passions

- weather, location and even budgets, schedules

- size and density of their friendship network,

- number uploaded photos, number of events attended, number of group memberships, and number of times user has been tagged in photos

- Frequency with which you post to your wall, Number of photos and albums, Number of friends, Whether you arrange your friends into groups

...with great results (from youarewhatyoulike in this case)...

male or female (93%), sex of user (93%), African American or Caucasian (95%), sexual orientation (88%), political beliefs (85%), Christian or Muslim (82%), age (75%), smoked cigarettes (73%), intelligence according to an IQ test (40%), extrovert or introvert (43%) and whether the user’s parents had split up by the time he or she was 21 (60%).

...as long as you have 50000+ samples :)

BIG NOTE: fancy maths is needed and no such thing akin to SIFT or SURF seems to exist in this facebook domain (to allow just about anyone to do decent analysis without a university research group of grads+phds).

(Not sure if it was just an anomoly, but those links are weighted to the UK and to UoCambridge+Microsoft - I would have expected a wider distribution of interest in this area).

Haven't looked at facebook, but there have been a few papers that have tried to predict the big 5 personality traits from twitter.

Golbeck et al. (2011) took 50 twitter profiles, had them complete personality surveys, and tried to train regression algorithms to automatically score them against the big 5. Most of the features came from applying the LIWC tool to tweet content. Unfortunately their best approach was no better than ZeroR (which basically assumes everyone has the same personality).

Quercia et al. (2011) do a larger study with 300+ profiles, using only 3(!) features: counts of you followers and people you follow, and of people who ‘favourite’ you. They claim pretty good results, which seems crazy...how can people's complex personalities boil down to these 3 statistics?

Sumner et al. (2012) have done the biggest study so far, with about 3000 profiles. They applied the LIWC tool to each user’s tweets to obtain the same features as Golbeck, and put them up Kaggle where they got 1,700 submissions and some hugely complex ensemble classifiers. Despite all this effort they got pretty poor results and conclude that the previous papers were a bit misleading about how difficult the task is. A big problem was that previous papers only reported error rates, which hide a lot of problems. Anyway, this paper is definitely a good read.

Another good read is Pennacchiotti & Popescu (2011). They aim for a general purpose algorithm for classifying twitter users, rather than one tailored to the big 5. But they explore a very wide array of features.

References:

Golbeck, J., Robles, C., Edmondson, M., & Turner, K. (2011) Predicting personality from twitter. In Privacy, security, risk and trust (passat 2011) and social computing (socialcom 2011)

Quercia, D., Kosinski, M., Stillwell, D., & Crowcroft, J. (2011) Our twitter profiles, our selves: Predicting personality with twitter. In Privacy, security, risk and trust (passat 2011) and social computing (socialcom 2011)

Sumner, C., Byers, A., Boochever, R., & Park, G. J. (2012) Predicting Dark Triad Personality Traits from Twitter usage and a linguistic analysis of Tweets. In Machine Learning and Applications (ICMLA 2012)

Pennacchiotti, M., & Popescu, A. M. (2011). A machine learning approach to twitter user classification. In ICWSM 2011.