Keystroke dynamics-based user identification via n-graph paths for shorter characteristic texts

摘要

Needless to say, computer systems should be more secure for our safe life and society nowadays. A keystroke dynamics-based user authentication is known as one of potentially valid biometrics. We have so far dealt with a middle-size input, which is neither short text such as password nor long text such as free text. In our previous research, some middle-size input text composed of two hundred keystrokes provides less than 1% EER for fifty-one subjects. To revise data input situation, this report deals with shorter texts. To this end, we have asked totally twenty-nine subjects to type nine shorter texts of Japanese old poems, proverbs and palindromes individually. Although we have proposed digraph model for user authentication or identification, we have not made use of it sufficiently. This report proposes to use size and kind of directed paths composed of n consecutive keys on that model to find different subjects beforehand. Since those paths depend on data filtering, our proposal requires tuning. As identification experiments for those nine characteristic texts, we have a little revised around 65% correctness from around 60% on average.