SMART-TMT: Bridging Social Trust and Semantic Web for Specialized Medical Tourism

Executive Summary

TL;DR: This paper presents SMART-TMT, a prototype recommender system designed for the Tunisian medical tourism sector. By extracting user interests from Facebook and weighting them based on the temporal trust of their social circle, the system overcomes traditional data sparsity issues. It transforms raw social interactions into a structured User Interest Ontology, which is then matched against a specialized domain ontology to deliver highly personalized travel and healthcare advice.

Positioning: This work sits at the intersection of Semantic Web mining and Social Recommender Systems. It moves beyond simple collaborative filtering by introducing an ontological layer that accounts for who you trust and when you last interacted with them.

The Problem: The "Cold Start" and Semantic Vague

Traditional recommendation engines are often "blind" to the deeper meaning of user preferences. They either rely on keywords (which are ambiguous) or collaborative filtering (which fails when a new user has no history—the Cold-Start problem).

Furthermore, while "Social Trust" has been explored, most systems treat a "Like" from five years ago the same as a "Comment" from yesterday. The authors argue that trust is a dynamic, time-sensitive reality. Without accounting for the temporal factor, a user model becomes a static graveyard of past interests rather than a reflection of current needs.

Methodology: From Social Graphs to Ontologies

The core innovation lies in a three-layered architecture that converts social "noise" into semantic "signal."

1. Trusted Friends Calculation

The system doesn't just look at a user's friends; it calculates a Level of Trust. It analyzes Facebook interactions (tags, likes, comments on videos/photos) and applies a temporal decay function.

• Insight: Interactions are not perceived the same way over time. Recent interactions carry higher weights in defining the current "trusted circle."

2. User Interest Ontology Construction

The system aggregates:

• Explicit Data: Direct answers to health-related questions.
• Implicit Data: Mined from the user's social profile and the preferences of their top-tier trusted friends.
• Weighting: Uses TF-IDF to measure term density and PCA (Principal Component Analysis) to reduce the dimensionality of friend preferences, ensuring the ontology focuses on the most significant interests.

3. Semantic Matching

The User Interest Ontology is matched against the Tunisian Medical Tourism (TMT) Ontology. The system calculates semantic similarity between the user’s weighted concepts and the available medical services (e.g., Plastic Surgery, Dental Cures, Thalassotherapy).

Experiments and Results

The system was implemented as a Java prototype. By using the Facebook Open Graph API, the authors demonstrated that the system could automatically populate a user's profile even with minimal direct input.

Key Findings:

• SOTA Comparison: Unlike traditional systems that use binary trust (trust/no-trust), SMART-TMT's continuous trust model ([0,1]) allowed for finer-grained filtering of social influences.
• Medical Domain Adaptation: The TMT ontology specifically modeled Tunisian infrastructure (clinics, hotels, thalasso centers), allowing the recommendation logic to bridge "health needs" with "leisure preferences."

Critical Analysis & Conclusion

Takeaway

The study proves that Social Trust + Ontology is a powerful combo for specialized markets. By letting "trusted friends" fill in the gaps of a user's profile, the system provides a seamless experience that feels personalized from the first click.

Limitations

• Data Privacy: The reliance on Facebook's Open Graph API is a double-edged sword; changes in social media privacy policies can significantly impact data availability.
• Complexity: Maintaining and updating two interlinked ontologies (User vs. Domain) in real-time is computationally expensive.

Future Work

The authors plan to implement dynamic ontology updating to reflect real-time shifts in user interests, ensuring the system evolves as the user’s social circle and health priorities change.