ITMO University student Kseniya Buraya and her colleagues taught an algorithm to predict users’ marital status with a 86% precision by means of combining the data of three social media platforms instead of one. The researchers believe that these results will help make human psychological portraits in future. Buraya will report about her findings at the most important scientific event in the field of artificial intelligence – AAAI Conference on Artificial Intelligence that will take place at the beginning of February in San Francisco.
The AAAI Conference on Artificial Intelligence (AAAI-17) is held in North America for the 31st year. During this time, the event gathered a pool of outstanding scientists and IT companies that work on both applied and theoretical research in the field of AI creation and learning. This year from 4th to 9th of February, the Hilton hotel in San Francisco, California, USA, brings together representatives of business, developers, scientists and students. Such industry leaders as IBM Research, Baidu, Amazon, Tencent, Microsoft and Facebook give speakers to the conference.
Among the participants, there is a Russian student Kseniya Buraya, undergraduate of the Computer Technology Department and researcher at the International laboratory “Computer technologies” of ITMO University. The article she co-authored won a student section contest, thus the scientist had an opportunity to submit a 3-minute thesis about the research. “This conference is one of the most affluent in AI field, the performance there opens significant prospects in the science world for me. Of course, it was not easy to become a speaker: among ten papers sent from our lab, only one succeed,” says the researcher.
Now Buraya does an internship at the National University of Singapore, where she studies approaches for describing human personality through social networks. She collects and analyzes user data, and then adapts it to Myers-Briggs Type Indicator (MBTI), a scale of psychological types based on Jung theses.
User profiling, according to the student, can be useful in a wide range of areas. For example, recruiters can learn more about people who apply for a job. More generally, characterizing personality through activity in social media will help discover threat groups as well as to find people prone to depression or suicide and support them.
For the project, the programmers found out that user profiling through several social networks allows one to specify his individual features. In particular, researchers focused on such a characteristic as the marital status, and combining data from Twitter, Instagram and Foursquare, taught the algorithm to predict this parameter with 86% of precision that is for 17% higher than single network provides. These results will be presented at the conference.
The MBTI scale describes a person in terms of how it interacts with the world, that in turn is easiest to learn from social media. Kseniya Buraya explains: “Many scientific sources associate human psychological type with marital status. So we decided to check how precisely we can predict this parameter to use it for psycho determining in the future”.
To train the algorithm in understanding the data input, the scientists turned the activity of users from New York, Singapore and London into sets, or vectors, of such parameters as an average tweet size, the most frequent objects in the photo, check-in distribution and so on. Then programmers used these vectors in basic machine learning models.
The authors regard the report on the AAAI-17 as a kind of “reserve” for further research. They are going to continue the study to achieve publication in peer-reviewed international journal.
The conference is organized by the Association for the Advancement of Artificial Intelligence (AAAI). Founded in 1979, this international non-profit scientific society promotes machine learning research, and assists to improve AI public understanding and education quality.
Kseniya Buraya, Aleksandr Farseev, Andrey Filchenkov and Tat-Seng Chua (2017), Towards User Personality Profiling from Multiple Social Networks. In Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence. AAAI