Assisted Pattern Mining for Discovering Interactive Behaviours on the Web

TL;DR: This work assists user researchers in customising event sets, filtering the noisy outputs of the algorithms and providing tools for analysing such outputs in an exploratory fashion, which facilitates the agile testing and refinement of the formulated hypotheses of use.

Abstract: When the hypotheses about users’ behaviour on interactive systems are unknown or weak, mining user interaction logs in a data-driven fashion can provide valuable insights. Yet, this process is full of challenges that prevent broader adoption of data-driven methods. We address these pitfalls by assisting user researchers in customising event sets, filtering the noisy outputs of the algorithms and providing tools for analysing such outputs in an exploratory fashion. This tooling facilitates the agile testing and refinement of the formulated hypotheses of use. A user study with twenty participants indicates that compared to the baseline approach, assisted pattern mining is perceived to be more useful and produces more actionable insights, despite being more difficult to learn.