[IEEE Access] Decoding the "Why" Behind Social Unrest: A Psychology-Based Framework for Human Need Detection
Recognizing Human Needs During Critical Events Using Machine Learning Powered Psychology-Based Framework
The paper introduces a psychology-based framework for human need detection using machine learning on social media data. It employs three specific models (RNC, INT, MNS) to identify "need-bearing" content, classify the specific type of need (Autonomy, Competence, Relatedness), and measure its satisfaction level, achieving State-of-the-Art (SOTA) results with accuracies up to 93.56%.
TL;DR
Researchers have developed a machine learning framework that doesn't just analyze what people are saying on Twitter, but why they are saying it by identifying fundamental psychological needs. By mapping social media text to the Self-Determination Theory, the system can detect if a population's needs for Autonomy, Competence, or Relatedness are being met. Tested against the 2018 Florida shooting, the model proved that unmet psychological needs are the primary precursors to social conflict and organized protest.
Background: Moving Beyond Sentiment Analysis
While standard sentiment analysis tells us if a tweet is "happy" or "angry," it fails to explain the motivation. An individual might be angry because their freedom is restricted (Autonomy) or because they feel disconnected from their community (Relatedness). This paper bridges the gap between Natural Language Processing (NLP) and Human Needs Theory (HNT), positioning social media users as "soft sensors" for the emotional health of a city.
Problem & Motivation: The Failure of Traditional Metrics
The authors point out three major flaws in current methods:
- Scalability: You cannot interview an entire city during a crisis.
- Surface-Level Mining: Existing AI tools often focus on consumerism (e.g., wanting to buy a product) rather than psychological well-being.
- The Conflict Link: Psychological theories suggest that violence is almost always a result of unmet needs. Without a way to measure these needs, authorities remain reactive rather than proactive.
Methodology: The Three-Layer Multi-Model Approach
The framework operates in a pipeline of three specialized models:
- Recognizing Need Content (RNC): Distinguishes between general noise and tweets that actually convey psychological states.
- Identifying Need Type (INT): Classifies the content into one of three universal needs:
- Autonomy: The need to self-endorse behavior (choice).
- Competence: The need to interact effectively with the environment.
- Relatedness: The need to feel connected to others.
- Measuring Need Satisfaction (MNS): Determines if the expressed need is satisfied or dissatisfied.
Feature Engineering & Architecture
The authors didn't just rely on words; they used a sophisticated feature set:
- LIWC & LCM: Lexicons that track social and cognitive processes.
- Emoji Analytics: Analyzing the frequency, sentiment, and even color of emojis to capture sub-textual nuance.
- SMOTE (Synthetic Minority Over-sampling Technique): To handle the fact that certain needs (like Competence) are expressed less frequently than others (like Relatedness) in crisis data.
Figure 1: The offline training phase showcasing the integration of psychological lexicons and machine learning.
Case Study: The Florida Shooting & "March for Our Lives"
The most compelling part of the research is the application to the 2018 Florida school shooting.
- Initial Reaction (Wednesday): There was a high satisfaction in Relatedness despite the tragedy. Why? Because the community was bonding through "thoughts and prayers" and shared grief.
- The Shift (Thursday-Saturday): As the conversation moved to gun control, Autonomy (the need for safety/policy change) satisfaction plummeted.
- The Prediction: The model showed a massive spike in "Dissatisfied Autonomy," which directly preceded the birth of the #MarchForOurLives movement. Once the protest began, Competence satisfaction rose because people felt they were finally "doing something" (effectively interacting with their environment).
Figure 2: Tracking the satisfaction levels of universal needs across the timeline of a critical event.
Critical Analysis & Conclusion
The results are impressive, particularly the 93.56% accuracy in measuring satisfaction. However, the study has limitations:
- Data Source: It is currently limited to English Twitter text. Sarcasm and cultural nuances in other languages remain a challenge.
- Multimodality: While the paper discusses images, the core classification is still heavily text-weighted.
The Takeaway: This framework successfully demonstrates that machine learning can quantify the "Invisible Hand" of psychology. For urban planners and public policy makers, this is an early warning system. By identifying which fundamental human needs are being suppressed in real-time, we can address social grievances before they escalate into violence.
Future Work: The authors plan to integrate Topic Modeling and Named-Entity Recognition to specify exactly what (e.g., a specific policy or person) is causing the dissatisfaction.