Carpoolap: Building Trust in the Sharing Economy via Social Graph Analysis

TL;DR

Despite the environmental and economic benefits of carpooling, many people avoid it because of the inherent risks of traveling with strangers. Carpoolap addresses this "trust gap" by integrating professional social networks and a weighted reputation algorithm that prioritizes real-world friendship degrees over easily-manipulated star ratings.

The Problem: The "Stranger" Barrier and Sybil Vulnerabilities

The fundamental obstacle to carpooling isn't a lack of vehicles or routes; it's a lack of trust. Prior works typically rely on simple feedback systems—the classic "5-star rating." However, these systems are fundamentally flawed:

• The Sybil Attack: A user with poor ratings can simply delete their account and start fresh with a clean slate.
• Privacy Leaks: Most platforms expose phone numbers and emails too early, leading to potential harassment or data misuse.

The authors argue that a rating system is only as good as the identity verification behind it.

Methodology: The Trust Computation Engine

The core innovation of this paper is its Reputation Algorithm. Instead of treating all "5-star" reviews as equal, it segments trust into two components: Friendship Level (lvFs) and Historical Ratings (rat[]).

1. The Social Weighting (0-7 Points)

The system utilizes the "Six Degrees of Separation" theory, specifically leveraging research that shows Facebook's average degree of separation is roughly 3.9.

• Direct Friend: 7 points.
• Mutual Friend: 6 points.
• Chain of 2+ Friends: 4 points.
• No Connection: 0 points.

2. Experimental Validation of Ratings (0-3 Points)

Ratings are secondary to social ties. Even a driver with perfect 5-star reviews from every trip can only gain a maximum of 3 points from the rating component.

Fig 1: The Client-Server Architecture utilizing Google Cloud Messaging (GCM) and a dedicated DB server.

The Formula for Trust

The reputation score is calculated as: $Reputation=lvFs+\frac{\sum rat[i]}{n}$ By setting a threshold of 7.5 points for automatic data visibility, the system ensures that only people with some degree of real-world social overlap can automatically see private contact details.

Defeating the Sybil Attack

A "bad" user (B0) might create dozens of fake accounts (B1, B2...) to give themselves 5-star ratings. In a traditional system, they would appear highly trustworthy.

Fig 2: A Sybil attack attempt where a malicious user tries to inflate their score via bogus accounts.

Why Carpoolap survives this: Because the social component (lvFs) is capped at 0 for these fake accounts, the total score for a malicious user would top out at 3.0 (from fake ratings). Since the "trust gate" is 7.5, the attacker remains isolated and untrusted by the wider network.

Implementation & Results

The authors developed a full-stack solution:

• Mobile: Android app (Carpoolap) using Google Maps 3D and Facebook SDK.
• Backend: Node.js with MongoDB (NoSQL) for flexible data handling of social graphs.

Fig 3: The Android application interface showing the rating and route selection process.

Critical Insight & Conclusion

This paper shifts the carpooling paradigm from "What is your rating?" to "Who do we both know?"

Takeaway: The true value of social networks in the next decade isn't just communication; it's identity and reputation as a service. By mathematically weighting social distance higher than user feedback, Carpoolap creates a self-policing community that is structurally resistant to the most common types of online fraud.

Future Outlook: While effective, the system currently relies heavily on centralized social giants (Facebook/Twitter). Future iterations could benefit from decentralized social protocols (like Lens or Farcaster) to ensure that reputation is portable and not owned by a single corporation.