Machine learning for socionics

[gohakubisu]

Greetings!
Im working on a project of neural network that will analyze human faces in relation to their sociotypes and will look for patterns. The purpose of this project if to have a neural network that will be able to define sociotype of a person by just looking at a photo. The way that is done is by analyzing face features of many humans with known sociotype.
The problem here is lack of data, so in order for the system to accurately define sociotypes it needs to look at MANY-MANY pictures first.

I have made a website which allows for uploading of the image+specification of the sociotype for that image.
If you could help my project by uploading images of people who's sociotype is known to you (you can also upload a picture of yourself if you dont mind).
I will really appreciate your help and contribution! And together we can make the neural network smart enough to get the sociotypes right!

If you have any questions I will be more than happy to answer them!
Will keep you posted on the progress of this project and thank you in advance!

Oh, btw, here is the website for uploading photos:
http://socionics.co/

[Cosmic Teapot]

Awesome project


My Questions:
What if the person who uploaded his image chases his type later?
What quality (px, dpi) do you need for the photo and are there other requirements like lighting, background etc?
Who will have have access to the photos and where do you plan to publish the results?
What about subtypes?

[gohakubisu]

Thank you, Horatio!

So the answers:
You can only choose the sociotype when uploading the image, than it is saved to the website database in a format of "sociotype"+"#of picture", thus there is no way to change the sociotype afterwards. The only way to do it if if I download the image from the database manually and rename it. So yeah, its best if you are sure about the sociotype from the beginning, otherwise if the sociotype does not fit to the face type the neural network will learn wrong pattern.

In regard to the quality. Pretty much everything will do. The images are cropped to 512x512px around the face ( after the face has been detected). So as far as the image is not super small it will fit. The neural network is already trained to recognize faces on the images, so it does not matter what the background is, if the image is in color or B&W, etc. Its pretty flexible, so no extra effort needed to make a "Special" picture that will be recognized. It just works

Regarding access to the files. All the images are stored in the websites database that only i can enter. The images will stay in private access and i am not planing to publish them anywhere. If you are concerned about the privacy issues there are 2 things: 1) i will not share anything from the participants with the 3rd parties. 2) i do not posses any information of the users,but the picture and the sociotype correlated to that picture. More than this i have no way of knowing if you uploaded picture of yourself or somebody else, so i would say that it is very safe and its not faceboook

On this point the first thing i want to see/prove is that the sociotype theory is actually working and if the features of teh human face tell something about the character. And if so is it possible to tell who is who just by seeing somebody's face. If the results are positive (and i really hope they are) than i might publish the results somewhere or write an article about it, but i will provide only statistical data in it (such as #of pictures used, how accurate the predictions are etc. so only numbers not pictures of participants).

About subtypes. What do you mean exactly? Currently the system is designed in a way that it can distinguish 16 sociotypes like here :
https://image.slidesharecdn.com/68e7...?cb=1437752809

[Cosmic Teapot]

Thank you, Horatio!

So the answers:
You can only choose the sociotype when uploading the image, than it is saved to the website database in a format of "sociotype"+"#of picture", thus there is no way to change the sociotype afterwards. The only way to do it if if I download the image from the database manually and rename it. So yeah, its best if you are sure about the sociotype from the beginning, otherwise if the sociotype does not fit to the face type the neural network will learn wrong pattern.

In regard to the quality. Pretty much everything will do. The images are cropped to 512x512px around the face ( after the face has been detected). So as far as the image is not super small it will fit. The neural network is already trained to recognize faces on the images, so it does not matter what the background is, if the image is in color or B&W, etc. Its pretty flexible, so no extra effort needed to make a "Special" picture that will be recognized. It just works

Regarding access to the files. All the images are stored in the websites database that only i can enter. The images will stay in private access and i am not planing to publish them anywhere. If you are concerned about the privacy issues there are 2 things: 1) i will not share anything from the participants with the 3rd parties. 2) i do not posses any information of the users,but the picture and the sociotype correlated to that picture. More than this i have no way of knowing if you uploaded picture of yourself or somebody else, so i would say that it is very safe and its not faceboook

On this point the first thing i want to see/prove is that the sociotype theory is actually working and if the features of teh human face tell something about the character. And if so is it possible to tell who is who just by seeing somebody's face. If the results are positive (and i really hope they are) than i might publish the results somewhere or write an article about it, but i will provide only statistical data in it (such as #of pictures used, how accurate the predictions are etc. so only numbers not pictures of participants).

About subtypes. What do you mean exactly? Currently the system is designed in a way that it can distinguish 16 sociotypes like here :
https://image.slidesharecdn.com/68e7...?cb=1437752809

Thank you for your detailed response. I am not planning to upload a photo because I'm not 100% sure of my type and I don't want to screw up your sample data. I thought these questions might be helpful for others who want to contribute to you project.


As for subtypes I mean for example the Se-ESTp and Ti-ESTp types. You can research the descriptions here:http://www.wikisocion.org/en/index.p...e=ILE_subtypes
They are often relevant because Subtypes differ in terms of facial expression, behavior, physique and other details.


I whish you success!

[JoshuaHilderberg]

I see on your submit page that you advise to take the personality test from https://www.16personalities.com/ which is rather a MBTI site.

If you mix personalities from MBTI and Socionics, you will probably get mixed results, because there is a significative part of people that don't type the same in both systems, since there are differences that are quite fundamental between them.

I am not saying MBTI is bad, just that if you mix both systems, neither Socionics, nor MBTI can use your results.

Otherwise, your project looks very interesting. I wish you good luck !

[JoshuaHilderberg]

I wonder if the traits could become more pronounced with age, and if it would be interesting to associate the age with the photo...

[gohakubisu]

I see on your submit page that you advise to take the personality test from https://www.16personalities.com/ which is rather a MBTI site.

If you mix personalities from MBTI and Socionics, you will probably get mixed results, because there is a significative part of people that don't type the same in both systems, since there are differences that are quite fundamental between them.

I am not saying MBTI is bad, just that if you mix both systems, neither Socionics, nor MBTI can use your results.

Otherwise, your project looks very interesting. I wish you good luck !

I see, you have a good point! Thanks for mentioning this to me, because that is something I did not consider indeed!
Do you, by any chance, know any good article that describes the differences? This is actually complicating the task, since i have no way to check if the uploaded picture is within MBTI or Socionics domain... Hmm, something to think about

I wonder if the traits could become more pronounced with age, and if it would be interesting to associate the age with the photo...

Yeah, thats a very interesting idea! But implementing it will require more time+ extra effort to train the neural network to recognize the ages.Though if the basis of the system work (recognizing sociotypes from a photo) than the age detection can be implemented as the next step! Thank you for the insight

[JoshuaHilderberg]

Do you, by any chance, know any good article that describes the differences? This is actually complicating the task, since i have no way to check if the uploaded picture is within MBTI or Socionics domain... Hmm, something to think about

I abandoned MBTI a long time ago, so I am very biased towards socionics ;-)
I often sum it up by saying that MBTI is more oriented towards observable behavior, while socionics is more about the internal structure of the information processing. MBTI is better to categorize people by their observable behavior, while socionics is better to understand the internal workings of people, especially for interactions, even-though you get more variety of behavior per type than in MBTI.
A few articles:
- MBTI article on Wikisocion
- MBTI vs Socionics – The Conversion Dilemma
- Socionics vs MBTI revisited (a more objective critique of both)

As for your project, the mere fact that socionics tries and gets interesting results by identifying people's type through VI (Visual Identification) is encouraging. There are people from MBTI who tried to do the same, but they never got interesting results and gave up (but that was a few years ago, so maybe it has evolved since then).

Unfortunately for your project, many people don't know that there is a difference between both systems and often mix them up, jumping from one website to another without paying attention. Another important difference, is that finding one's type is easier in MBTI (well, comparatively to socionics) because it is more focused towards behavior. So with one or two tests, you can get a good confidence in the results. Whereas in socionics you usually need to correlate different methods (behavior tests, profiles comparison, dichotomies preference, VI, self typing through understanding of the theory, identifying intertype relations with other people, etc.), as you can see in this list of tests.

But implementing it will require more time+ extra effort to train the neural network to recognize the ages.Though if the basis of the system work (recognizing sociotypes from a photo) than the age detection can be implemented as the next step! Thank you for the insight

I see. Maybe using the neural network to identify the age is not needed. If the age was given with the photo, we could see if the neural network results give stronger scores to older people. That could be interesting to find out in (maybe in a later experiment).

[JoshuaHilderberg]

I am not very interested VI, but I have seen studies that used a good collection of pictures. I don't know if they were digitalized, but it could provide an interesting preliminary test. Also, because they were taken in a single country, you wouldn't get much ethnicity differences, which could help (?).

Just an example: http://www.the16types.info/vbulletin...ortraits-Alpha

Or maybe find the original pictures that were mixed to create those composites: https://www.pinterest.com/heatherado/socionics-faces/

[mclane]

This is a good idea but for it to flourish I see some requirements that have to be met:

1. The picture uploader must be 100% sure of their typing (this includes subtype).
2. The image analysis program must be 100% accurate in catching the pattern between people of the same alleged type and subtype.