Authentication is an important aspect of day-to-day life. This paper shows the accuracy of activity-based authentication (ABA) for 1000 users accessing Yahoo! Japan websites. The authors claim this is the first evaluation of ABA under a large user population. This type of authentication is important, as the most common form of authentication, password, is known to be hard to use.

The authors also introduce the notion of bags with multiple actions and time periods for each bag. Machine learning is then used to predict the user. The accuracy obtained is about 80 percent.

Although the accuracy is pretty high, the authors do not discuss whether this accuracy is practical. Some experimental and evaluation details are also missing, such as the machine learning algorithm used, the number of false positives, and the time taken to perform the ABA. Finally, not all the elements in table 1 are included in table 3.

The intended audience for this paper is researchers in the field.