A team of scientists from the National Research Nuclear University "MEPhI", the National Research Center "Kurchatov Institute" and the Voronezh State University have developed a method that teaches a computer to recognize the sex of a person from a text written by him with an accuracy of 80 percent. Scientific development belongs to the field of computational linguistics. The research was carried out with a grant from the Russian Science Foundation. The results are published in the journal Procedia Computer Science.
Numerous scientific studies show that a written text inevitably reflects the characteristics of its author - gender, psychological characteristics, level of education. Speech is a valuable psychodiagnostic tool used by human resources specialists of large companies, as well as security services.
Based on the analysis of speech, it is possible to diagnose the presence of certain diseases in a person (dementia, depression) and a tendency to suicidal behavior. The need to establish the characteristics of the author of the text also grows with the development of Internet communications: it is important for companies to know which groups of people like their products and services.
Scientists working in this area (linguists, psychologists, information technology specialists), on the basis of the numerical values of various parameters of the text, build mathematical models for diagnosing certain personality parameters.
A team of experts analyzed the effectiveness of various machine learning technologies using neural networks for text analysis.
In the course of the study, they compared the accuracy of solving the problem of gender identification of texts based on two approaches to data-based modeling: on the one hand, machine learning algorithms (support vector machine and gradient boosting), on the other hand, deep learning neural networks (convolutional neural networks and recurrent neural networks with long short-term memory).
“We have achieved high results in determining the gender of the author of the text thanks to advanced neural network models, in conditions when the author does not hide his gender. The next task is to determine the sex in terms of its intentional concealment,”says Alexander Sboev, associate professor at NRNU MEPhI.
So, in the following texts, which were initially posted on a dating site, the neural network easily finds a catch in ten out of ten cases, moreover, the author deliberately puts the name of the opposite sex in the signature.
Promotional video:
The text was written by a girl: “I am a handsome, muscular man of 30 years. I work for a large oil and gas company in a good position with a decent salary. I live in my own apartment in Moscow. The property also has a small but pretty house in one of the villages in Italy. I am fond of sports, in particular, football. I love going out for the weekend, I hate stay-at-home. A girl who would suit me should have a modest disposition, good looks and an attractive figure by modern standards. She should share my interests, should not be jealous, and should not try to make me feel jealous. I'm not going to support the girl, because I believe that both should work in the family. I also prefer to keep the budget separately. I will not tolerate treason."
The text was written by a man: “Hello! I am extremely unhappy, extremely! Why are you behaving like that with us ?! We are people too, we are all equal! Are you sexist? I won't take this anymore! I’ll break your car all over, paint it. Wait, inhuman. I will finish this way."
The results of this study showed that an approach based on the use of convolutional neural networks and deep learning methods for recognizing the gender of the person who wrote the text is the most optimal.
Now a group of researchers is working on the problem of age recognition.