Phonebook-Centric Social Networks: Solving the Identity Matching Puzzle

TL;DR

Building a contact list from scratch is the primary friction point in joining new social networks. This paper proposes a phonebook-centric approach that treats your mobile contacts as the foundation of your social graph. By utilizing a weight-based similarity algorithm and dynamic "customized contacts," the authors solve the problem of identifying and merging duplicate identities across private and public domains with a false-positive rate under 10%.

Context: The Phone as a Social Anchor

In the traditional social network model (Facebook, MySpace), you join a walled garden and manually "add" friends. The authors argue that your mobile phonebook is already a high-fidelity social structure. The challenge lies in Similarity Management: how does the system know that "Dad" in your phone is "John Doe" on the server?

The Problem: Why Exact Matching Fails

Most contact sync tools rely on exact string matching. If one record has "Joseph" and the other "Joe," or if two people share a central office phone number, traditional systems either miss the match or create a false duplicate.

• Prior Work Bias: Basic tools ignore the context of "social dimensions" (e.g., distance, common friends, or name synonyms).
• The Merging Dilemma: Once a private contact is linked to a network member, how do you allow the user to keep their nickname ("Dad") while still receiving updates when John Doe changes his official profile picture?

Methodology: The Weight-Based Engine

The core innovation is a heuristic algorithm that calculates a similarity score based on field-specific weights.

1. Architectural Overview

The system, named phonebookmark, uses a Drupal-based server with XML-RPC access for mobile and desktop clients, keeping the local phonebook and social graph in sync.

2. The Algorithm Logic

Instead of a binary "match/no match," the system calculates a weighted sum:

• Name Match: +30
• Private Mobile Number Match: +10
• Email Match: +15
• Conflicting Birthday: -40 (Significant penalty)

The raw weight ($w$) is then converted into a probability percentage ($pp$) using an arc-tangent function: $pp=80+\arctan \left(\frac{weight}{100}\right)\frac{40}{\pi }$ This ensures that while the score increases with more evidence, it never hits a "100% certainty" too easily, acknowledging the inherent uncertainty in social data.

3. Dynamic Name Handling

The system "learns" synonyms. If users frequently merge "Katharine" and "Kate," the system stores these in a similarity table, allowing the algorithm to automatically suggest Kate-to-Katharine matches in the future.

Experimental Results

The authors validated the system using 400 internal users managing 70,000 contacts.

• False Positive Rate: Suggested matches were correct in over 90% of cases.
• Scalability: As the number of users grew, the number of detected similarities per person remained stable, proving that the algorithm filters noise effectively as the database expands.

Critical Insight: Customized Contacts

The most overlooked but valuable contribution is the Customized Contact definition. It creates a "triangular" link:

1. Private Contact: Your raw phone data.
2. Member Profile: The global user data.
3. Customized Copy: A bridge that lets you rename a contact locally while subscribing to global updates. This provides the "best of both worlds"—personalization and real-time accuracy.

Conclusion and Future Outlook

While the current algorithm relies on manual weight tuning, it provides a robust framework for similarity management. The authors plan to implement machine learning models to adjust weights automatically based on user behavior logs. This work paves the way for advanced location-based recommendations and automated social graph expansion that feels seamless to the end-user.