A text mining approach to Internet abuse detection

As the use of the Internet in organizations continues to grow, so does Internet abuse in the workplace. Internet abuse activities by employees-such as online chatting, gaming, investing, shopping, illegal downloading, pornography, and cybersex-and online crimes are inflicting severe costs to organizations in terms of productivity losses, resource wasting, security risks, and legal liabilities. Organizations have started to fight back via Internet usage policies, management training, and monitoring. Internet filtering software products are finding an increasing number of adoptions in organizations. These products mainly rely on blacklists, whitelists, and keyword/profile matching. In this paper, we propose a text mining approach to Internet abuse detection. We have empirically compared a variety of term weighting, feature selection, and classification techniques for Internet abuse detection in the workplace of software programmers. The experimental results are very promising; they demonstrate that the proposed approach would effectively complement the existing Internet filtering techniques.