Sentiment and Emotion-aware Multi-criteria Fuzzy Group Decision Making System

Basic Information

• Paper ID: 2408.11976
• Title: Sentiment and Emotion-aware Multi-criteria Fuzzy Group Decision Making System
• Authors: Adilet Yerkin, Pakizar Shamoi, Elnara Kadyrgali (Kazakh-British Technical University)
• Classification: cs.AI
• Publication Date: July 2024
• Paper Link: https://arxiv.org/abs/2408.11976

Abstract

This paper proposes a sentiment and emotion-aware multi-criteria fuzzy group decision making (GDM) system designed to enhance the effectiveness of consensus achievement in group environments. The system integrates natural language processing techniques to analyze sentiments and emotions expressed in textual data, considering not only explicit numerical preference inputs but also participants' underlying perspectives. The system aggregates individual preferences into a collective preference matrix, then inputs sentiment, emotion, and preference scores into a fuzzy inference system to obtain comprehensive evaluation scores. The system's effectiveness is validated through a small-scale decision process involving a group of friends selecting a vacation hotel. Results demonstrate that integrating sentiment and emotion analysis into group decision systems significantly improves consensus among participants.

Research Background and Motivation

Problem Definition

1. Core Issue: Traditional GDM systems require participants to provide explicit numerical opinions, but in reality, people often express their views through natural language text (reviews, social media, messages, etc.), which contains uncertainty and fuzziness.
2. Significance: Group decision-making is ubiquitous in daily life (selecting restaurants, vacation destinations, etc.), requiring consensus promotion among diverse preferences.
3. Limitations of Existing Approaches:
   • Traditional GDM systems rely on explicit numerical inputs and cannot process natural language expressions
   • Lack consideration of sentiment and emotion factors
   • Difficulty handling uncertainty and fuzziness in the decision process
4. Research Motivation: Combining sentiment analysis, emotion detection, and fuzzy logic to construct an intelligent group decision system capable of processing natural language inputs.

Core Contributions

1. Proposed the first sentiment and emotion-aware multi-criteria fuzzy group decision framework, organically combining natural language processing techniques with fuzzy inference systems
2. Designed an integrated architecture of voting systems, sentiment analysis, and fuzzy inference systems, achieving a balance between quantitative assessment and qualitative insights
3. Constructed a comprehensive consensus evaluation mechanism, including feedback systems and consensus measurement metrics
4. Validated system effectiveness in practical decision scenarios, demonstrating high consensus levels through a hotel selection case study

Methodology Details

Task Definition

Given a set of alternatives X = {x₁, x₂, ..., xₙ}, a set of experts E = {e₁, e₂, ..., eₘ}, a feature set F = {f₁, f₂, ..., fₚ}, and corresponding weights Z = {z₁, z₂, ..., zₘ*ₚ}, the objective is to rank alternatives based on preference values and textual discussions provided by experts.

Model Architecture

1. Preference and Voting System

The preference value of expert eⱼ for alternative Xᵢ is calculated as:

Pref_{E_j}(X_i) = ∑_{i,j,k=1}^{i=n,j=m,k=p} W_k(X_i) × Z_k(E_j)

where preference assessment employs a three-level scale: "-1" for opposition, "0" for indifference, and "1" for agreement.

2. Sentiment Analysis Module

VADER from the NLTK library is used for sentiment analysis, providing quantitative insights into participant discussions. The final preference score is calculated as:

Preference Score = α × Joint Sentiment Score + β × Emotion Score

where α and β are weight parameters balancing the impact of sentiment and emotion.

3. Fuzzy Inference System

• Input Variables:
  • Voting Preference: aggregated voting scores of participants for each alternative (0-100)
  • Sentiment Preference: average sentiment scores derived from qualitative discussions (-1 to 1)
• Output Variable: Overall preference score (0-10)
• Fuzzy Rules: 15 IF-THEN rules combining input levels to produce output levels

Technical Innovations

1. Multimodal Fusion: Innovatively combines numerical voting, textual sentiment analysis, and emotion detection
2. Fuzzy Logic Processing: Effectively handles uncertainty and fuzziness in human opinions
3. Natural Language Interface: Allows experts to participate in decision-making using everyday language
4. Sentiment Weighting Mechanism: Through weight allocation of α=0.6, β=0.4, emphasizes the contextual relevance of sentiment analysis

Experimental Setup

Dataset

Hotel Selection Dataset:

• 7 hotels, each with 8 attributes: price, rating, meal type, room size, star rating, beach access, city distance, swimming pool
• Preference data provided by 4 participants
• Participants debated the advantages and disadvantages of each hotel through chat messages

Evaluation Metrics

• Consensus Measure: Interquartile range (IQR) and average consensus level
• Feedback Score: Fuzzy inference based on agreement and confidence levels
• Preference Score: Overall preference integrating voting and sentiment analysis

Implementation Details

• Sentiment Analysis: VADER (compound score range -1 to 1)
• Emotion Detection: Text2Emotion library detecting happiness, anger, surprise, sadness, fear
• Feature Normalization: Continuous features normalized through mean comparison; binary features used directly
• Weight Settings: Sentiment analysis weight 0.6, emotion analysis weight 0.4

Experimental Results

Main Results

Hotel3 achieved the highest comprehensive score of 7.52, becoming the recommended choice.

Consensus Evaluation Results

Consensus Measure: IQR=2.19, Average Consensus=7.34, Consensus Level: High

Experimental Findings

1. Importance of Sentiment Analysis: Although hotel3 did not have the highest voting score, it achieved the highest comprehensive score due to positive sentiment feedback
2. System Effectiveness: High consensus level (7.34) validates the method's effectiveness
3. Emotion Impact: Negative emotions significantly affect final scores (e.g., hotel7's fear emotion -0.63)

Related Work

Main Research Directions

1. Fuzzy Group Decision Making: Since the 1970s, fuzzy logic has been widely applied to handle uncertainty in decision-making
2. Application of Sentiment Analysis in Decision Making: An emerging research direction including SAR-MCMD methods, sentiment-driven fuzzy cloud multi-criteria models, etc.
3. Application of Natural Language Processing in GDM: Allowing experts to use everyday language rather than specific formats or numbers

Advantages of This Work

Compared to existing work, this paper is the first to comprehensively integrate sentiment analysis, emotion detection, and fuzzy inference systems, providing a more comprehensive group decision-making solution.

Conclusions and Discussion

Main Conclusions

1. Integrated Framework Effectiveness: Integrating sentiment and emotion analysis into multi-criteria fuzzy group decision systems is feasible
2. Consensus Improvement: The system significantly enhances consensus levels among participants
3. Practical Validation: The hotel selection case demonstrates the system's applicability in real-world scenarios

Limitations

1. Case Simplicity: Currently validated only in simple hotel selection scenarios with limited complexity
2. Text Data Dependency: System effectiveness heavily depends on text data quality; incomplete or biased data may affect results
3. Implementation Complexity: Multi-component integration makes system implementation challenging
4. Context Adaptability: Effectiveness may vary across different decision contexts

Future Directions

1. Extended Application Scenarios: Explore more complex and diverse decision scenarios
2. Method Comparison: Conduct comprehensive comparisons with traditional multi-criteria decision methods such as TOPSIS, ELECTRE, PROMETHEE
3. Sensitivity Analysis: Perform comprehensive sensitivity analysis to evaluate method robustness
4. Component Optimization: Further refine components to improve generalizability

In-Depth Evaluation

Strengths

1. High Innovation: First to propose a sentiment and emotion-aware fuzzy group decision framework with significant technical innovation
2. Complete Methodology: Forms a complete decision process from preference collection to consensus evaluation
3. High Practical Value: Addresses the practical problem of traditional systems' inability to process natural language inputs
4. Solid Theoretical Foundation: Based on mature fuzzy logic and sentiment analysis theories

Weaknesses

1. Limited Validation Scale: Validated only in a simple 4-person group scenario, lacking testing in large-scale and complex scenarios
2. Insufficient Comparative Experiments: Lacks comprehensive comparison with other advanced group decision methods
3. Lack of Theoretical Justification for Parameter Settings: Weight parameters α=0.6, β=0.4 lack theoretical support
4. Unknown Generalizability: Applicability across different cultural backgrounds and decision types requires further verification

Impact

1. Academic Contribution: Introduces a new sentiment-aware dimension to group decision systems with certain theoretical value
2. Application Prospects: Has potential applications in social media decision-making, online collaboration, and similar scenarios
3. Reproducibility: Method description is relatively clear but requires more implementation details for complete reproducibility

Applicable Scenarios

1. Small-scale Group Decision Making: Informal scenarios such as friend gatherings and family decisions
2. Decisions Where Sentiment Factors Matter: Travel destination selection, entertainment activity planning, etc.
3. Multi-criteria Evaluation Scenarios: Product selection, service evaluation, and other decisions requiring comprehensive multi-factor assessment

References

The paper cites 26 relevant references covering important works in fuzzy logic, group decision-making, sentiment analysis, and other fields, providing a solid theoretical foundation. Particularly noteworthy are Zadeh's foundational work on fuzzy set theory, Herrera-Viedma et al.'s consensus models for group decision-making, and the VADER sentiment analysis tool.

Overall Assessment: This paper proposes an innovative sentiment-aware group decision framework with clear technical methodology and reasonable experimental design. Although the validation scale is limited, it provides new ideas and directions for the development of group decision systems, possessing certain academic value and application potential.