PreCon – Expressive Context Prediction Using Stochastic Model Checking

Abstract

Ubiquitous systems need to determine the context of humans to deliver the right services at the right time. As the needs of humans are often coupled to their future context, the ability to predict relevant changes in a user’s context is a key factor for providing intelligence and proactivity. Current context prediction systems only allow applications to query for the next user context (e.g. the user’s next location). This severely limits the benefit of context prediction since these approaches cannot answer more expressive time-dependent queries (e.g. will the user enter location X within the next 10 minutes?). Neither can they handle predictions of multi-dimensional context (e.g. activity and location). We propose PreCon, a new approach to predicting multi-dimensional context. PreCon improves query expressiveness, providing clear formal semantics by applying stochastic model checking methods. PreCon is composed of three major parts: a stochastic model to represent context changes, an expressive temporal-logic query language, and stochastic algorithms for predicting context. In our evaluations, we apply PreCon to real context traces from the domain of healthcare and analyse the performance using well-known metrics from information retrieval. We show that PreCon reaches an F-score (combined precision and recall) of about 0.9 which indicates a very good performance.

This research has been supported by FP7 EU-FET project ALLOW (contract number 213339).