Proof of Quality

Overview

The Proof of Quality (PoQ) system is designed to evaluate the quality, consistency, and authenticity of Amazon user data contributions. It processes various data types, including retail order history, digital purchases, Audible activity, and Prime Video usage. By assigning a quality score to each dataset, PoQ ensures that submitted data is both genuine and valuable, incentivizing meaningful contributions.

Scoring Methodology

The PoQ system employs a sophisticated scoring model that integrates category-specific weights, logarithmic scaling, and time decay. Each data category, such as retail or digital items, is assigned specific weights to assess relevant components, with a combined weight of 100. Logarithmic scaling rewards larger datasets while ensuring diminishing returns for excessive volume, while time decay prioritizes recent data contributions, boosting their scores.

The scoring process also incorporates validation mechanisms, combining metadata analysis (60%) with large language model (LLM) validation (40%) to ensure data authenticity. This hybrid approach enables robust evaluation of datasets, with scores ranging from 0.0 for low-quality contributions to 1.0 for highly valuable data.

Key Metrics by Category

• Retail Cart Items: Number of items (40%), unique products (30%), date range (20%), and active items ratio (10%).

• Digital Items: Number of items (30%), unique products (30%), and total amount (40%).

• Order History: Total orders (20%), monetary value (20%), unique products (20%), and additional aspects such as date range, websites used, and payment diversity.

• Audible History: Number of purchases (25%), unique audiobooks (25%), total amount (30%), and recency.

Prime Video Activity: Viewing sessions (25%), total hours (25%), unique titles (25%), devices used (10%), and date range (15%).

Future Enhancements

PoQ will continue to evolve with plans to integrate machine learning for advanced validation, expand support for additional data types, and implement dynamic threshold adjustments. Enhanced fraud detection mechanisms are also in development to further protect the DLP.

This system ensures contributors are rewarded fairly for high-quality, unique data submissions, fostering a robust and trustworthy data ecosystem.