- The new project achieved 94% accuracy.
- The AI based system can choose "Winners" and "Losers."
- There are specific features and criteria that were used to train the system.
Russian economist Ivan Smirnov from the Higher School of Economics created an artificial intelligence system that determines a student’s academic performance based on their posts on the social media platforms. Smirnov received a Russian grant to facilitate the project. The result is 94% accurate.
The work titled “Estimating educational outcomes from students’ short texts on social media” is available here. The project used posts from the social media to train a neural network. A majority of the gathered posts were from the Russian alternative to Facebook, VKontakte.
The reports of 2,468 subjects who took the PISA test in 2012 were selected. Overall, more than 130,000 texts and posts were included in the training sample. Smirnov tested its skills on the posts of students from hundreds of major universities in Russia.
Computer forecasts were compared with the average use score of applicants and graduates of the educational institution and official general information on academic performance. In total, more than 1 million posts of almost 39,000 users were analyzed.
Thus, the neural network has learned to distinguish participants with good academic performance from the poor performers. For example, well-performing students tend to write long texts with a variety of words.
They also use longer words, often borrowed from other languages. Another marker is the love of the verbs “I think,” “I believe,” and other indicators that denote the thought process.
Those who cannot boast of success in their studies prefer short posts. Also, “losers” can be recognized by the abundance of exclamation marks, emoticons, words written in caps, and very poor grammar. Additionally, the post topics differed.
In the posts of excellent students, the credit is given from authors and heavy use of scientific terms. Low-performing students prefer to discuss horoscopes, road accidents, and military service (in Russia military service is mandatory for one year for men).
“Based on these rules, our model identified students with high and low academic performance using Vkontakte posts with an accuracy of up to 94%. We also tried to apply it to short texts on Twitter – successfully,” says Smirnov.
Smirnov is hoping to teach artificial intelligence to identify the state of depression by posts on social networks. Previously, some scientists have shown that this is possible. At the same time, Smirnov notes that the development of new technologies also promises alarming prospects.
“On the one hand, this approach can be useful in identifying depression that affects academic achievement,” the expert notes.
“On the other hand, our results once again showed how vulnerable user privacy is in a social network. People worry about the ubiquitous cameras and facial recognition systems, but even a seemingly insignificant digital footprint, such as a short text, can become a source of information that a person did not even intend to disclose.”