Receteame.com: Integrating Persuasion and Trust into Social Food Recommendations

TL;DR

receteame.com is a novel social recommendation platform that goes beyond simple keyword matching for recipes. By synthesizing content-based filtering with social argumentation techniques, it addresses the "Cold Start" problem and provides "Persuasive" recommendations backed by trust, reputation, and friendship metrics.

Background Positioning

In the spectrum of AI-driven discovery, this work bridges the gap between traditional Collaborative Filtering (CF) and Human-Centric Social Computing. It shifts the focus from optimizing a mathematical error function (like RMSE) to maximizing user satisfaction through social context and rational justification.

Problem & Motivation: The "Black Box" of Big Data

The authors argue that Web 2.0 has made users producers, not just consumers, yet recommendation engines haven't kept pace. Existing systems suffer from:

1. The Explainability Gap: Users don't know why a recipe was suggested.
2. The Trust Deficit: In open networks, it is hard to gauge if a review is authentic.
3. The Cold Start Problem: New users with zero history receive poor or no suggestions because traditional algorithms lack data points.

Methodology: The Power of Social Logic

The core innovation lies in its hybrid algorithm that triggers two parallel search streams when a user clicks "Recommend me":

1. The Content-Based Stream

Matches user profile footprints (dietary restrictions, past likes) against recipe metadata.

2. The Social Persuasion Stream

This is where the magic happens. The system queries the target user's social circle (and some random users to offset cold starts). It ranks recipes based on three high-level social parameters:

• Direct Trust: How much the target user (and their friends) trusts the recommender.
• Global Reputation: The recommender's average trust rating across the whole network.
• Friendship Strength: Calculated via predictive variables like interaction frequency on platforms like Facebook.

(Note: Please refer to the paper for the internal argumentation and dialogue game flowcharts.)

3. Argumentation & Agreement

Unlike a simple weighted average, the system uses an internal agreement procedure. It employs Argumentation Techniques, allowing the algorithm to prioritize recommendations from users who can provide better "justifications." This mimics human social dynamics where we trust a friend who explains why a dish is good over one who just says "try this."

Experiments & Results: Real-World Deployment

The system was validated through its deployment on receteame.com. Key highlights include:

• Scale: Successfully supported 2,000+ users and 30,000+ views.
• Robustness: By mixing social queries with random user sampling, the system maintained functionality even for new users with sparse profiles.
• Integration: Seamless extraction of nutritional data and dietary restrictions (e.g., gluten-free, vegetarian) as hard constraints before the social ranking layer.

(Note: See Figure 2 in the paper for specific trust evaluation distributions.)

Critical Analysis & Conclusion

Takeaway

The project proves that social layers are essential to modern AI. By treating recommendation as a "persuasive dialogue" rather than a database query, receteame.com creates a more human-like experience.

Limitations

While the system uses Facebook for friendship strength, its dependency on external social APIs can be a bottleneck for privacy-conscious users. Furthermore, calculating global reputation and multi-agent argumentation in real-time for millions (instead of thousands) of users would require significant computational optimization.

Future Outlook

The authors envision evolving this into a Group Recommender System—capable of planning a weekly family menu or a dinner party that balances the conflicting preferences and allergies of multiple people simultaneously. This moves AI from individual assistance to collective coordination.