From Chat to Consensus: Bridging Social Network Sentiment and Group Decision Making

TL;DR

Researchers have developed a framework that transforms informal social network discussions into structured group decisions. By combining Sentiment Analysis, Fuzzy Preference Relations, and K-means Clustering, the system extracts expert opinions from hashtags and comments to calculate a "Consensus Degree." A case study involving 40 experts proved that this method could effectively prioritize corporate investments while maintaining a high agreement rate (over 80%).

The Evolution of Decision Making: Why Numbers Fail

In the era of Web 3.0, decision-making is no longer confined to boardrooms. While social platforms are ripe with expert opinions, a fundamental gap exists: computers require precise numbers (0.1, 0.5, 0.9), whereas humans communicate through nuance ("slightly better," "unprofitable," "much more important").

Traditional Group Decision Making (GDM) forces experts into uncomfortable mathematical formats, often leading to data fatigue or shallow evaluation. The authors argue that the future of decision-making lies in Passive Extraction—letting experts talk naturally and using NLP to do the heavy lifting.

Methodology: The Logic of Digital Agreement

The proposed model follows a rigorous four-step pipeline: Extraction → Preference Calculation → Consensus Assessment → Ranking.

1. Sentiment-to-Fuzzy Mapping

The core innovation lies in the transformation of comparative expressions. By using three lexical lists (Positive, Negative, and Neutral), the system maps phrases like "needed more" to numerical values. These are then normalized into a fuzzy preference matrix where $p_{ij}=0.5$ represents indifference and $p_{ij}o1$ represents total dominance.

2. The Consensus Hierarchy

Agreement isn't binary. The paper introduces a mathematical hierarchy to measure consensus at three levels:

• Pair Level ($c{p}_{ij}$): How much do experts agree on Alternative A vs. Alternative B?
• Alternative Level ($c{a}_i$): What is the overall agreement on a specific investment?
• Global Level ($GCP$): Does the group as a whole agree enough to move forward?

(Image: The conceptual flow of integrating web interactions into GDM)

Case Study: Where Should the Money Go?

The authors tested their model on a company deciding between 7 investment alternatives, ranging from "Staff Development" ($x_1$) to "Saving Money" ($x_7$). Using a private corporate portal and the hashtag #GDM, 40 employees provided qualitative feedback.

The Ranking Results

Using the GDD (Dominance) and GNDD (Non-Dominance) operators, the system produced a definitive ranking:

1. Staff Development ($x_1$) - Clear Winner
2. New Product Development ($x_4$)
3. New Equipment ($x_3$) ...
4. Saving Money ($x_7$) - Least Preferred

(Image: Visualization of the final alternative ranking based on expert consensus)

Deep Insight: The Demographic Divide

By applying K-means clustering, the study revealed fascinating demographic patterns in consensus. Experts were split into four clusters based on gender and age.

The data showed that Cluster 1 (Women ~40 years old) and Cluster 2 (Women ~25 years old) often showed higher internal consensus than their male counterparts. Specifically, older experts (Cluster 0 and 1) tended to be more unanimous in their strategic vision for the company, whereas younger male experts (Cluster 3) displayed the highest variance in opinion.

(Image: Diamond diagram showing alternative-level consensus across different demographic clusters)

Critical Perspective & Conclusion

While the study successfully bridges the gap between text and math, it has notable limitations:

• Lexical Sensitivity: The model relies on pre-defined keyword lists. Sarcasm or highly context-dependent language in social media might skew preference values.
• Scale: While 40 experts is a "large-scale" start for GDM research, true Web 2.0 social networks involve thousands of users, necessitating more robust automated sentiment engines.

The Takeaway: This work proves that we no longer need to choose between the "wisdom of the crowd" and "mathematical rigor." By treating social network publications as raw data for GDM, organizations can reach faster, more democratic, and transparent decisions.