User Modeling for the Internet of Things

Abstract

The Internet-of-Things (IoT) consists of a large number of interconnected, low cost devices and the framework for managing them. While this provides the means for rich and ubiquitous personalized interaction, a key gap is the lack of support for user modeling to harness and manage personal data gathered from IoT. Our work fills this gap by creating the IoTum user modeling framework for Internet-of-Things applications. Our design goals were to make it easy for IoT application developers to use, tackling the difficulty of building ubicomp systems. At the same time, we aimed to achieve light-weight, flexible, powerful, reactive user modeling that is accountable, transparent and scrutable. Applications interact with IoTum via three primitives, 'tell' to add evidence to a user model, 'ask' to retrieve interpreted evidenced from the model and 'listen' which establishes a monitoring process which can trigger the application. The first two of these have been part of many previous user modelling frameworks, including Personis, the server by Kobsa and Fink, CUMULATE and info-bead. However, IoTum is the first user modelling framework we are aware of that supports the last of these, the 'listen', which is key to making effective use of the many existing and emerging low cost sensors of IoT. Another important aspect is that IoTum maintains provenance information in the user model, to support accountability, scrutability and explanations. There are also mechanisms to define the user model and do introspection on existing models. To demonstrate that it meets its design goals, we describe its use to build a nutrition chatbot.