Information security applications of natural language processing techniques

Umut Topkara, Purdue University

Abstract

In this thesis we investigate applications of natural language processing (NLP) techniques to information security problems. We present our results in this direction for two important areas: password authentication, and information hiding in natural language text. We have limited this thesis to the realm of language engineering, i.e., our emphasis is on adapting the existing NLP techniques for our purposes, rather than in developing new NLP techniques. Our password mnemonics system helps users to remember random passwords, hence making it possible to implement organizational policies that mandate strong password choices by users. Moreover, in our system password changes do not necessitate a new mnemonic, thereby further easing the users' task of memorizing their respective mnemonics. Our robust natural language text watermarking system can avoid the removal of the watermark text by an automated adversary, in the same way used by authentication systems to avoid an automated adversary's compromise of the password string hidden within the password mnemonic. We have also laid the groundwork for followup research in this area.

Degree

Ph.D.

Advisors

Atallah, Purdue University.

Subject Area

Computer science

COinS