Personalized Memory Testing for Names Using Email Archives

We describe a personalized and scalable system for testing of autobiographical memory (specifically, the recall of names) using email archives. We have developed a novel system that creates a personalized test by analyzing the subject’s sent email messages over the past year. The system generates fill-in-the-blank questions from the user’s messages, with the answers being names determined to be significant in the email corpus using text analysis. Such testing can be ecologically valid and deeply personalized compared to existing techniques. We describe technical aspects of question and answer generation and report on a study where 35 participants answered an aggregate of 1,400 questions. We obtained a dataset of about 80 features per question and user response. Our preliminary analysis of this dataset supports several expected memory characteristics, such as that recall dips over time, and answer terms that span many days are better remembered, as are terms that are mentioned often. It also hints at some differences between memories associated with different sentiments. Our technique can be regularly employed by individuals without incurring major expense, in the privacy of their homes. Moreover, one of our findings is that users enjoyed taking such a test (31 of 35 users signed up to take more tests in the future), implying that such tests may be a practical and effective way of testing and studying autobiographical memory on a large scale and would lead to better screening tests for memory disorders.