GlassMiner

A Structural-Semantic Fusion Framework for Mining Internet-Wide Looking Glass Services

A novel structural-semantic fusion framework that intelligently mines and analyzes Looking Glass (LG) services across the global Internet, combining advanced clustering algorithms, network-aware crawling, and LLM-driven classification to achieve comprehensive discovery and validation of network measurement infrastructure.

描述

About GlassMiner

A breakthrough approach to Internet-wide Looking Glass service discovery using structural-semantic fusion and distributed validation

Template Clustering Analysis

Revolutionary structural-semantic fusion approach that processes seed pages and extracts distinctive features through advanced clustering algorithms, achieving superior accuracy in LG service pattern recognition.

Network-Aware Crawling

Intelligent crawler leveraging Autonomous System (AS) information and clustering insights to efficiently collect web pages with minimal redundancy and maximum coverage across diverse network infrastructures.

LLM-Driven Classification

State-of-the-art large language model integration utilizing few-shot learning techniques to achieve high-precision webpage classification while minimizing manual labeling requirements.

Core Features

Powerful capabilities that make GlassMiner the leading framework for Looking Glass service analysis

Structural-Semantic Fusion

Novel fusion of structural DOM analysis and semantic content understanding to achieve superior Looking Glass service identification and template clustering accuracy.

Modular Architecture

Four-module pipeline design enabling independent execution of seed processing, web crawling, LLM classification, and VP discovery components for maximum flexibility and scalability.

Distributed VP Discovery

Advanced distributed validation system for discovering and confirming Looking Glass vantage points (VPs), with automated template matching and API functionality verification.

System Architecture

Four-module pipeline design enabling comprehensive LG service discovery and validation

Seed Processing

Template clustering and structural-semantic fusion

Web Crawler

AS-informed network-aware data collection

LLM Classifier

Few-shot learning content classification

VP Discovery

Distributed validation and template matching

Research Impact

Measuring the effectiveness and scale of GlassMiner's capabilities

7,556
LG Webpages Discovered
7,125
VPs Discovered
3,237
New Undocumented Sites
97.6%
Classification Precision