Open Source Data Collection Techniques for Weapons Transfer Information

Abstract

Monitoring weapons proliferation requires detecting and tracking weapons transfers. Many public sources of weapons-transfer data come from incomplete and manually collected and maintained government and commercial records. We propose a technique to gather weapons-transfer information by mining publicly available Web pages for features, which we categorize as arms, actions, actors, or money. We design a retrieval system and parser, and develop techniques for extracting currency values from text, measuring precision without available training data, and measuring recall with a parallel but different corpus. Results show that, of the sentences matching four feature categories, 70% of relevant features were found, and sentences that only matched three categories introduced more false positives. We conclude that such a technique can improve the speed at which transfer information is compiled.