RuleSelector: Selecting Conditional Action Rules from User Behavior Patterns.

Modern smartphones and ubiquitous computing systems collect a wealth of context data from users. Conditional action rules, as popularized by the IFTTT (If-This-Then-That) platform, are a popular way for users to automate frequently repeated tasks or receive smart reminders, due to the intelligibility and control that rules provide to users. A key drawback of IFTTT systems is that they place the burden of manually specifying action rules on the user. While multiple rule mining algorithms have been proposed in existing work to automatically discover action rules, they generate hundreds of action rules, and the problem of how to present a small subset of rules to smartphone users and allow them to interactively select action rules remains unsolved. In this work, we take the first step towards solving this problem by designing and implementing RuleSelector, the first interactive rule selection tool to allow smartphone users to browse, modify, and select action rules from a small set of summarized rules presented to the user. We propose novel rule selection metrics to address the needs of smartphone users, and analyze the performance of RuleSelector using data from 200 users. We also perform a qualitative user study in order to evaluate how users use the RuleSelector tool and perceive the selected rules, and present the insights gained and design recommendations for future rule selection systems. Our users rated the selected rules from useful to very useful, and an important finding of our study is that users prefer an interactive rule selection system such as RuleSelector that automatically suggests rules, but allows users to select and modify the suggested rules. Finally, we examine the promise of RuleSelector in other ubiquitous computing systems such as smart homes and smart TVs by applying our tool to public context datasets from these domains.