How AI predicts student performance through social media subscriptions

A team of Russian researchers from the Higher School of Economics (HSE University), Skoltech, and Tomsk State University has developed a groundbreaking method to forecast academic performance using artificial intelligence. By analyzing students’ VK (VKontakte) subscriptions, the AI system was able to predict academic success or struggle with high accuracy. The study was published in the journal IEEE Access.

Digital Footprints as Predictors of Success

Modern internet users leave behind extensive digital footprints — likes, group memberships, comments, music choices, and even link clicks. These seemingly minor data points often reflect deeper aspects of personality and interests. The researchers leveraged these traces to uncover patterns linking online activity to academic achievement.

The study analyzed 4,445 university students with publicly accessible VK profiles. Using Natural Language Processing (NLP) techniques, the researchers examined the themes of the groups students followed, the linguistic complexity of the texts in those communities, and the emotional tone of the content. Each participant’s digital behavior was then compiled into a personal profile, which was processed using machine learning algorithms.

What Did the AI Discover?

The algorithm revealed clear distinctions between high-performing and low-performing students:

• High achievers tended to follow communities focused on science, education, and technology. They engaged with complex, analytical content, often participating in discussions and preferring materials with a positive or neutral emotional tone.
• Lower-performing students were more likely to subscribe to entertainment-focused groups — memes, humor, music, and gaming. This content was generally less cognitively demanding, often carried a negative tone, and lacked educational value.

The AI was able to predict student performance with over 80% accuracy, suggesting strong potential for use in education and beyond.

How Does It Work?

The method involves several key steps:

1. Data Collection: Public subscription data from VK profiles is gathered.
2. Content Classification: NLP tools identify each group’s topic category, text complexity (vocabulary, structure), and emotional sentiment (positive, neutral, or negative).
3. Profile Building: A digital portrait is created for each student, reflecting their interests and engagement patterns.
4. Prediction: The algorithm cross-references these portraits with academic records (grades, attendance) to find correlations.

For example, a student following pages on programming and scientific discoveries is statistically more likely to excel than one engaged primarily with memes or short-form video content.

Applications and Ethical Considerations

The method opens up new opportunities in educational settings:

• Early Detection: Universities can identify students at risk and offer support before performance declines.
• Personalized Learning: Course materials could be adapted based on student interests and engagement styles.
• Increased Motivation: Teachers might use this data to encourage students to explore research and academic content aligned with their interests.

However, the use of AI in social media analysis raises ethical concerns. Even though data is public, some may see it as an invasion of privacy. The researchers emphasize that only anonymized profiles were used, and informed consent was obtained. Widespread use of such methods would require clear privacy regulations.

What’s Next?

The team plans to expand the analysis to include likes, comments, and music preferences, potentially boosting prediction accuracy. They also aim to adapt the method for other platforms like Telegram and Instagram, though their more closed nature makes data collection more challenging.

In the future, similar AI models could be applied not only in education, but also in HR, marketing, and psychology — helping to forecast job performance, consumer behavior, or even emotional well-being based on online habits.