Towards a System for Recommending Tasks to Users based on User Intentions and Environment Capabilities Chuong Vo, Seng W. Loke and Torab Torabi Department of Computer Science and Computer Engineering, La Trobe University, Australia ccvo@students.latrobe.edu.au , s.loke@latrobe.edu.au , t.torabi@latrobe.edu.au ABSTRACT The burden of managing and utilizing Pervasive Computing Environments (PCEs) comprising a collection of devices on users and devices embedded in the environment (e.g. user’s comprehension of PCE capabilities, system configurations, service compositions, and management of system failures) currently falls on users. Users might be overwhelmed by information and capabilities that an environment might offer – for example, there could be thirty to a hundred computers within a living room of the future, each providing various functionalities which the user should then learn how to use (some functionalities perhaps provided by a combination of such devices). Even if the user has a mobile device through which task requests can be issued to this collection of devices, the user would still need to know what are feasible/possible tasks s/he could perform here. Our research aims to minimise this overhead by proposing a task recommender system named TASKREC that manages the PCEs (e.g. services, resources, and context information) where the user is located and suggests to users (via his/her mobile device) tasks that are relevant to them and which can be accomplished by the PCE in which the user is located. Via reasoning with knowledge about current environment capabilities, context information, and user intention, the system recommends to the user possible and relevant tasks when he/she asks for it. In this position paper, we present a formal problem of task recommendation. We propose a conceptual architecture for TASKREC and then provide a scenario for the vision of TASKREC. Categories and Subject Descriptors D.2.11 [Software Architectures]: Domain-specific architectures. H.3.3 [Information Search and Retrieval]: Selection process. H.3.4 [Systems and Software]: User profiles and alert services. General Terms Measurement, Design, Human Factors Keywords Task Computing, Recommendation Systems, Context-Awareness, Pervasive Computing. 1. INTRODUCTION The world is moving towards universally connected information spaces where ‘technologies weave themselves into the fabric of everyday life’ [11]. Spaces in such a world are called pervasive computing environments (PCEs) [9]. They are collaborative spaces consisting of mobile users, software services, interconnected electronic devices (e.g. set-top boxes, smart-phones, PDAs, laptops, displays, cameras, speakers), and pervasive networks (e.g. HomeRF, Bluetooth, cellular network, Wi-Fi, WiMAX, mobile broadband). Although PCEs enable us to access information and services anytime and anywhere, they are overwhelming us with an overload of information, services, and complex configurations [2, 4, 10, 12]. This is leading towards a heavy burden to users when they want to accomplish a computing task. Moreover, because of the increasingly sophisticated and feature-rich software applications and computing devices, controlling even one of these devices or applications is increasingly difficult. As a result, controlling and exploiting a PCE which consists of a dynamic range of high-tech devices, feature-rich applications is even more difficult. To exploit a given PCE, the users must (1) recognise feasible tasks in the environment in order to issue feasible tasks; (2) map their high-level goals of tasks to the low-level operational vocabularies of capabilities of devices and functionalities of applications embedded in the environment; and (3) properly specify the constraints for their tasks subject to the context information of their surroundings. These requirements may be beyond ordinary users as the complexity, diversity and sheer number of devices (as well as well different combinations of ways they might work together for the user) continually rises and the amount of information in their surroundings increases exponentially – the user is surrounded by a proliferation of devices in the immediate environment but might not know what tasks the user can perform through these devices in the environment. Also, there is a need for solutions which free users from system managements and configurations so that they just focus on their intended tasks. To address the problem of information overload, there have been research efforts that aim to recommend information and services. However, users are still required to manually integrate the recommended information and services in order to achieve their goals. This calls for task recommender systems. Few efforts (e.g. [1, 6, 7, 8]) have focused on task recommendation. However, the existing systems do not take the capabilities of a PCE as a whole into their recommendation processes. This is critical because accomplishing a task in a PCE often involves a range of services, devices, and context information; the environment capabilities should be utilised in recommending feasible tasks. We present a context-aware task recommender system (TASKREC) which aims to recommend or even automatically accomplish ‘relevant’ tasks. Our approach proposes to use three similarity measurements in the task ranking process for recommendation: the similarity of user intention to task objective, the feasibility of tasks, and the autonomy of task performance. TASKREC can help a user make decisions on the question: given the current situation (including the current environment, the current context, and the current user’s intentions) what tasks can and should be performed? And how much distraction/attention/obtrusiveness do the selected tasks effect on the user? Further, our proposed system can be integrated into context- aware applications which can adapt their behaviours to changes in the environment and user intention.