[Tech Review] Mapping the Digital Ego: Neural Network Classification of Social Media Behavior
Analysis of Formats of Young People’s Communicative Behavior in Social Network
The paper presents a neural network-based framework to classify the communicative behavior of young people on social networks into 18 psychological scales derived from G. Ammon’s Ego-structure test. By leveraging a heterogeneous feature set including profile statistics, text posts (Word2Vec), and images (ImageNet labels), the authors achieved state-of-the-art classification accuracy across various psychological dimensions using Support Vector Machines and Ensemble methods.
TL;DR
Researchers in St. Petersburg have developed a sophisticated method to detect the psychological state of young social media users. By aligning social media activity with the G. Ammon Ego-structure test, the team used neural networks and multi-modal data (text, images, and stats) to categorize users across 18 psychological scales, identifying destructive behavioral patterns with significant accuracy.
Background: Why Social Media Monitoring is Now a Psychological Science
The rise of destructive online influences—ranging from cyberbullying to self-harm provocation—has made it critical to understand the "communicative behavior" of the younger generation. Traditional sentiment analysis (detecting "happy" vs. "sad") is too shallow. This paper shifts the focus to Ego-functions, looking for deeper personality indicators like narcissism, anxiety, and aggression in their constructive, destructive, or deficient forms.
Methodology: From Profiles to Psychological Scales
The core innovation lies in the transformation of messy social media data into structured psychological features.
1. The Ammon Framework
Instead of generic labels, the study uses 18 specific scales. Each function (e.g., Aggression) is split into:
- Constructive: Healthy expression.
- Destructive: Harmful expression.
- Deficient: Complete lack of the function.
2. Feature Extraction Breakthrough
The researchers didn't just look at text. They engineered a three-pronged feature vector:
- Statistical: Number of friends, subscribers, and photos.
- Linguistic: Using a semantic correspondence algorithm. Most frequent words from specific psychological groups are compared with a user's posts using Word2Vec.
- Visual: User photos are passed through an ImageNet-trained CNN to extract category labels, which are then analyzed for semantic psychological patterns.
Table 1: The diverse feature set showing correlations with psychological class labels.
Experimental Insights
The study tested various classifiers, including Linear Regression (LR), Neural Network Models (NNM), and Support Vector Machines (SVM).
- The Winner: SVMR (Radial Basis Function SVM) consistently outperformed basic neural networks on the testing sample.
- Highest Accuracy: Achieved in the "Constructive Anxiety" (72.64%) and "Deficient Sexuality" (69.35%) categories.
- Social Indicator: Interestingly, the "Number of Friends" was the most influential numerical feature (Correlation: 0.172), suggesting that social connectivity is a primary external indicator of internal Ego-structure.
Table 2: Breakdown of training and testing accuracy across all 18 psychological scales.
Deep Dive: The Correlation of Destructiveness
One of the most profound sections of the paper explores the interaction between different Ego-functions. Using Spearman’s rank correlation, the authors found that Destructive Narcissism is a central hub for other negative traits. It correlates positively with destructive/deficient anxiety and external delimitation, acting as a potential "early warning sign" for overall psychological instability.
Table 3: The Spearman Correlation Matrix revealing how different psychological traits cluster together.
Critical Analysis & Conclusion
Limitations
- Dataset Size: With 460 unique entries, the model is a strong proof-of-concept but would benefit from "Big Data" reinforcement to improve the generalization of lower-performing scales (like Destructive External Delimitation, which hovered around 47% accuracy).
- Contextual Nuance: While Word2Vec captures semantics, it may miss the sarcasm or slang prevalent in youth subcultures.
Future Outlook
This work lays the groundwork for creating "Psychological Digital Twins." By automating the Ammon's test through social media monitoring, platforms could potentially provide "constructive anti-stimuli" or alerts to mental health professionals before a user’s behavior turns self-destructive.
Final Takeaway
The integration of clinical psychological theory (Ammon's Test) with modern NLP and Computer Vision represents a significant leap from simple sentiment analysis toward a true digital diagnosis of human personality.