In collaboration with PayPal
Reuth Mirsky, Kobi Gal and David Tolpin
Interfaces and Scheduling and Planning (UISP) ICAPS (2017)
This paper presents preliminary results of our workwith a major financial company, where we try to usemethods of plan recognition in order to investigate theinteractions of a costumer with the company’s onlineinterface. In this paper, we present the first steps ofintegrating a plan recognition algorithm in a real-worldapplication for detecting and analyzing the interactionsof a costumer. It uses a novel approach for plan recognitionfrom bare-bone UI data, which reasons aboutthe plan library at the lowest recognition level in orderto define the relevancy of actions in our domain, andthen uses it to perform plan recognition.We present preliminary results of inference on three differentuse-cases modeled by domain experts from thecompany, and show that this approach manages to decreasethe overload of information required from an analystto evaluate a costumer’s session — whether thisis a malicious or benign session, whether the intendedtasks were completed, and if not — what actions areexpected next.