Chat assistant for criminal group infiltration

点击翻译按钮获取中文摘要

Thales Research & Technology Canada (TRT), based in Québec City, is investigating ways to develop natural language processing (NLP) tools that will assist an infiltration agent trying to participate in a hacker chat room on the Dark Web, with the goal of uncovering ahead of time which software vulnerabilities are being discussed and targeted in upcoming attacks. The company-specific problem is to develop tools that will help the infiltration agent by monitoring chat conversations, understanding the language, and suggesting ideas for replies. Each of these three steps represents a major applied problem. Firstly, monitoring chat conversations requires untangling a web of messages from multiple simultaneous conversations taking place in the same room at the same time between a dynamically-changing set of participants. This we propose to do by computing the similarity between features found in the content and metadata of individual messages. Secondly, understanding the messages requires normalizing the unconventional spelling found online and interpreting the new idioms created by the web community. We have at our disposal a set of state-of-the-art NLP tools that can help in this challenge. Finally, in order to suggest relevant information to the infiltration agent, the system must have access to a database of information it can query, as well as an awareness of the agent's online persona to filter this information and retain useful facts. This part of the system can be based on the virtual assistant technology developed by the applicant's research team. This project will give Thales the ability to learn about and anticipate security attacks before they happen, and will thus give an important boost to software and network security for Canadian companies, banks, government agencies, and individuals.