PROJECT SOCIOMETR

[Yaaroslav]

Please support my project!

Litecoin
LcqjZGwLv8HcchTFgFDud8boS8emAcowhw

USDT
TXX2xXQHA4K7NmPD8SdXSPAJ8mzkfoDdxr


https://isocionics.com/projects/sociometr15

https://isocionics-com.translate.goo..._x_tr_pto=wapp

https://boomstarter.ru/projects/1087...bvi_sotsiometr


https://boomstarter-ru.translate.goo..._x_tr_pto=wapp

[Beautiful sky]

How can we translate the page into English?

[Yaaroslav]

How can we translate the page into English?

Didn't you follow the link with the translation?

[Yaaroslav]

PROJECT "SOCIOMETER 15"


The program "Sociometer" uses content analysis of the text taken from the typed speech, and on the basis of this, it makes a decision about the probability of belonging of the typed person to one or another type.

Keywords : Reinin features, typing, exact typing, neural networks, socionics, C#, Reinin features calculator, PR calculator, Sociometer, content analysis.





Who most often says - the word "I want" - Prudent. Who most often uses the preposition "to" - Statics. Who most often says the word "absolutely" - emotivists! These and other studies form the basis of the sociometer project.

User instruction:
Main file: Sociometry.exe

The program works with Russian speakers. Chat with the target person on an arbitrary topic in a moderately friendly mood and at the same time capture video / audio. Turn the first 15 minutes of communication into text. To turn words into text, be sure to use the service of professional typesetters. Modern programs do not provide the required quality of translation. Only the text of the typed is needed. Convert to file with txt extension . Then, in the program, click on the "Calculate TIM" button - specify the path to the file and get the program's response in the form of version distribution. The program also gives a ranking in descending order of typing versions in a separate text file, and detailed probability calculations. The output file will be placed in the Logs folder.

You can also test the work of the sociometer on current texts by selecting a file from the "Processed" folder and see where it will type this or that person.

Program core:
The calculation algorithms are based on the program "PR Calculator". First, the sociometer calculates the probability of assigning the person being typed to one or another polarity of the dichotomy, and then enters the obtained values ​​into the PR calculator and calculates the final probability of referring to one or another socionic type.



Program operation methods:
The program uses content analysis, analyzing the text of the typed speech, according to the words, analyzing their specific concentration in terms of 10,000 characters of the text, as well as by the number of characters (extraversion-introversion dichotomy: criterion: talkativeness per unit of time).



Sources of the training sample:
Interviews of people whose socionic types were confirmed by at least 3 socionics (including me) in the amount of 113 people were taken as the basis for training. At the moment there are programs "Sociometer 15" - a program that analyzes at an interval of 15 minutes, and "Sociometer 45" - a program that analyzes a text from a person's speech at an interval of 45 minutes. - Currently in the project.


How dichotomies are calculated:


When calculating the polarities of certain dichotomies, 2 types of distributions were taken as a basis:

Normal distribution:

[Yaaroslav]

Exponential Distribution

The density of the normal distribution is calculated by the formula

* k

Where k is the correction factor. Because the distribution is not normal. This is due to the fact that the normal distribution density function captures the negative area a little, which ultimately does not give the total area of ​​​​the normal distribution density function equal to 1. The normal distribution density function will be multiplied by this coefficient so that the total value in the area of ​​\u200b\u200bthe normal distribution integrated from zero to infinity gives 1

x is our resulting frequency number per 10,000 characters

is the standard deviation for our dichotomy.



.

Where a is the arithmetic mean. The subscript 1 is the first polarity of the dichotomy. The subscript 2 is the second polarity of the dichotomy.

x is our received frequency per 10,000 symbols

is the standard deviation for our dichotomy. The subscript 1 is the first polarity of the dichotomy. The subscript 2 is the second polarity of the dichotomy.





The probability in this situation will be calculated by the formula:







P (1| B) – Probability of assigning the typed to the first polarity of the dichotomy
P (2| B) – Probability of assigning the typed to the second polarity of the dichotomy

Exponential Distribution:

f (x,a)=a*exp(-ax)

a is equal to 1/M, where M is the arithmetic mean

x is our received frequency



The probability in this situation will be calculated according to the formula according to Bayes' theorem:

For exponential distribution:







For mixed distribution:





Or:





f (x) - can be calculated by different methods, but they will be calculated in the same way after that according to the formula according to Bayes' theorem:

p 1= f 1( x )/( f 1( x )+ f 2( x ))



Note to the program: the program uses the number of characters spoken by the typed as a way to recognize the Extraversion-Introversion dichotomy, working on the principle of normal distribution

Step 7. Calculate the probability of assignment to a dichotomy based on the obtained probabilities of all words of the keys using Bayesian method. According to this formula:



Step 8: Calculate all dichotomies in the PR calculator.

About how calculations are carried out in the PR calculator - read the article "PR Calculator"

We bring to your attention the version: "Sociometer 15" (other versions are possible, but the core and calculation algorithms will be the same)

In this version, studies were conducted on an interval of 15 minutes from 113 people: everything was turned into text and patterns were studied about the frequencies of the words used, which the sociometer uses to solve the inverse problem: based on the frequencies of words, to conclude that the typed person belongs to one or another sociotype evaluating the probability in parallel.

Main file for setting parameters

keys . txt

This file contains:





The response threshold is the threshold from which the sociometer makes a decision about the type of type being typed. Below this threshold, the sociometer considers that the type is not defined. Specified by the first line as a percentage. "It's better to abstain than to mistype"

maxrange =4 Parameter that counts the number of typed in the first four. You can put a different number. Works when the bulk typing button is pressed. Relevant for such tasks: “If the sociometer is wrong, then at what level is the true version?”

symbol - the number of symbols parameter applicable to the dichotomy

Extraversion-Introversion

Set in style:

symbol;5864.18;2111.57;1;4705.04;1840.62;1

5864.18 - the average number of characters spoken by extroverts in the first 15 minutes of communication

2111.57 - Sigma for extroverts

1 - correction factor due to the fact that the distribution differs from the normal one. If it's a unit, then it doesn't affect anything.

4705.04 is the average number of characters spoken by introverts in the first 15 minutes of communication

1840– Sigma for introverts

1 - correction factor. If it's a unit, then it doesn't affect anything.

Other words:

well; 35.74; 18.13; 1.03; 45.30; 25.20; 1.04;

35.74 - The frequency of the words "Well" extroverts say in 15 minutes in terms of 10,000 characters.

For example: someone can say the word "Well" 40 times, but at the same time they will say 5000 characters. Then the frequency of the word "Well" he said is 80.

18.13 - Sigma for extroverts

1.03 - Correction factor for extroverts

45.30 - The frequency of the words "Well" say introverts in 15 minutes in terms of 10,000 characters.

30;25 - Sigma for introverts

1.04 - Correction factor for introverts

good;3.38;0;0;5.62;0;0

If the correction factor is 0, then the density function is calculated using the exponential distribution formula.

3.38 - Frequency of "Good" words spoken by extroverts in 15 minutes per 10,000 characters.

5.62 - The frequency of the words "Good" say introverts in 15 minutes in terms of 10,000 characters.

The program implements the functions of both single typing and mass typing.





The mass typing button allows you to analyze the effectiveness of the sociometer by the following parameters:

Threshold - the percentage of confidence of the sociometer, below which it is recognized that "there is no version"





Number of versions - the number of versions typed. Because there is a trigger threshold - some versions of the sociometer admit that "NO version" = Better to abstain than mistype

Versions NO - number of typing abstentions

Matched - the number of matches with the original typing version:

The files are named something like this:

ENFJ Kamenskaya.txt

- the first 4 letters the program reads and remembers the socionic type.

Recognition: the percentage of versions that the program has identified correctly divided by the total number of types typed



Convergence: The number of correctly typed divided by the total number of typed of those with versions

Included in the 1st maxrange=4

A parameter that counts the number of typed hits in the top four. You can put a different number. Works when the bulk typing button is pressed. Relevant for such tasks: “If the sociometer is wrong, then at what level is the true version?”

Results of work:

Many experiments were carried out with the selection of the number of words of the keys. The most successful version is the version where the first 100 words ranked in the list by frequency were taken as a basis and loaded into the sociometer for recognition.

An important parameter of work is the correlation of the parameters "Threshold" and "Convergence". If the difference between them is small, it means that the sociometer is not lying. In the previous slide, the trigger threshold and convergence are both 70%, which indicates that the sociometer's confidence is supported by real statistics of correct conclusions. This means that if, for example, a sociometer concluded that you are Hamlet with a probability of 90%, then most likely you are. Congratulations: the sociometer was able to correctly determine your TIM!






Here, on this slide, my interview with the socionician Victoria Kamenskaya, officially known for the socionic type Hamlet, is uploaded during the first 15 minutes.

If you still doubt this version, then with a probability of 88% your version will be in the list of the first 4 versions. This is implemented in the output file











Recognition was 42% - which already exceeds the accuracy of typing, achievable by me in the first 15 minutes. I usually need 1-3 hours to reliably determine the socionic type. In my experience, I manage to determine the type with an accuracy of about 20-30% within the first 15 minutes.



Conclusions:
The first program has been created that can type based on content analysis and objectively evaluate accuracy without misleading users. The proximity of the parameters: "Threshold" and "Convergence" - guarantees for the user that if, for example, he scored more than 90% typing accuracy, then he received a reliable version of his socionic type
This program is able to recognize in the interval of the first 15 minutes of communication already in 42% of cases, which exceeds the accuracy of typing the author himself
Machine learning or neural networks have a huge potential to achieve high-precision recognition of anything. Now it's the turn of socionics in the field of high-precision typing!




PS:

I admit that I could be wrong. I am donating this Sociometer to the socionic community, along with the source code written in C#, for further improvement.

Thanks

The author thanks Alexander Vladislavovich Rybak for the formulas for the density function of the normal distribution and the exponential distribution. Also programmer Ilya Kulpin for writing a program based on the TOR.



Literature:

Polozov Ya.V. "Reinin Signs Calculator"
Augustinavichute A. Reinin's Theory of Signs. Essay on socionics // "Socionics, mentology and personality psychology", 1998, No. 1-6.
Karpenko O.B. "Position of an expert" SMiPL N 1, 1999.
Mironov "Napoleon of Reinin's Signs: Results of Practical Research" SM&PL No. 1 2003
Morozov M., Proshkin I. Probabilistic typing based on Reinin's features // "Socionic newspaper", 2003, No. 14 (17).
Reinin G.R. A group of bipolar traits in Jung's typology. // "Socionics, mentology and personality psychology", 1996, No. 6.
Bayes formula http://nsu.ru/mmf/tvims/chernova/tv/lec/node15.html