Activity Context Representation — Techniques and Languages: Papers from the 2011 AAAI Workshop (WS-11-04) CARe: An Ontology for Representing Context of Activity-Aware Healthcare Environments Marcela D. Rodríguez1, Mónica Tentori1, Jesús Favela2, Diana Saldaña1, Juan-Pablo García1 1 Master and Doctoral Program of Computer Science, Autonomous University of Baja California 2 Computer Science Department, CICESE Research Center {marcerod,mtentori}@uabc.edu.mx, favela@cicese.mx, {pablo.garcia,dianasj}@uabc.edu.mx design approach which enabled us to identify the concepts and relationships that characterize nurses’ activities in a hospital [Tentori,11], and elders activities of daily living carried at their homes [García-Vázquez, 10]. The CARe ontology is a knowledge base consisting of a history of users activities and information that shapes the way an activity-aware application will behave [García-Vázquez, 10]. To facilitate the use of the CARe ontology we have specialized the SALSA agent [Rodríguez, 05] framework with customizable activity-aware mechanisms that enable autonomous agents to infer and represent activities of hospital workers and activities of daily living of elders [Tentori, 11]. Abstract Representing computational activities is still an open problem in the field of Activity-Aware Computing. In this paper, drawn from our experiences in developing activityaware applications in support of two populations: nurses working in hospitals and elders living independently; we defined the Context Aware Representational (CARe) model. CARe is an ontology that enables the representation and management of computational activities. We illustrate, through application scenarios, that the CARe ontology is flexible enough to enable developers to customize the computational representation of healthcare activities. Introduction Activity-aware computing allows smart environments to provide continuous activity awareness and opportunistically offer assistance to support users’ current activities [Tentori, 08]. One of the biggest challenges for developing activity-aware applications includes the representation of context in the form of activities. Although, some projects have extensively researched how to manage computational activities [Bardram, 07] and how to infer activities [Tentori, 08]; little has been said regarding their representation. Indeed, there are still open challenges including the identification of the characteristics of human activities, how these characteristics could be matched to its computational representation and how this representation could be tailored for different applications domains. Building from in-depth fieldwork and the development of several activity-aware applications, we developed the CARe ontology as an approach to cope with the abovementioned problems. To reach this end, we carried out two qualitative studies following an activity-centered Related Work Several research works held that the foundations of Activity Theory could be used to produce context classification systems. For instance, [Cassens, 06] stated that contextual knowledge can be categorized on personal context (information about the subjects), task context (the objects towards which activities are directed and the rules), spatio-temporal context (the community), environmental context (mediating artifacts) and social context (division of labour). Similarly, [Kaenampornpan,04] identified from the Activity Theory foundations, the key elements and their relationships that should be modeled; additionally, they argument that time is a crucial element that should be modeled to enable the creation of activity histories in order to enable context-aware applications to predict future user’s actions. Other works have focus on providing development architectures that incorporate efficient mechanisms to model context by introducing two-layered ontologies to enable context-aware applications to access and interpret context information [Gu, 05] [Bochini,07] [Soldatos, 07]. Compared to these ontologies that were designed to Copyright © 2011, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. 59 represent the general concepts that characterize users, network, devices and applications of a ubiquitous computing environment, the CARe ontology was designed to represent the context of the activity in which the user is engaged, which is supported by a computational unit designated as e-Activity. providing supervisors with awareness of nurses’ activities: “The nurse Letty begins her shift by following the scheduled tasks in her activity-aware assistant (Figure 1a). After a few hours, while Letty is finishing administering medicines, the oncologist, Dr. Guillen, arrives to examine the patient in room 226. Letty’s activity-aware assistant suggests her to join Dr. Guillen. Letty suspends her current activity and moves to room 226 to help Dr. Guillen with the medulla extraction. Once they start the surgical procedure, the activity-aware assistant in Letty’s phone automatically reschedules her remaining tasks (Figure 1b). After a while, the system notifies the head nurse, Carmen, about Letty’s suspended activity. Carmen consults Letty’s activity timeline (Figure 1c) and decides to assign the suspended task to an available nurse”. The smart application in this scenario reacts and adapts its behavior based on the computational representation of Letty and Dr. Guillen’s activities. Qualitative Studies In the next sections we present the results of two qualitative studies that led to the design of activity-aware application for healthcare. The Routines of Daily Hospital Care We analyzed data captured during a nine-months workplace study to understand activities’ hierarchy, characteristics, and execution model [Tentori,11]. Specifically, we looked into nursing activities, actions, and operations. We identified six actions that nurses executed on a daily basis as activities of daily care (ADCs) (Figure 1). For instance, a nurse begins her shift at 7:00 a.m. by taking her patients’ vital signs. Then, she administers medicines and monitors patients’ diets. Around 10:00 a.m., the nurse calculates the patients’ liquid balance by weighting each patient’s urine bag and measuring the liquid output. She also cleans and bathes patients. Meanwhile, the physicians start their rounds in the ward, and the nurse makes sure the patients’ information is available for clinical case assessment. When the rounds conclude, the nurse completes the ADCs by carrying out the physicians’ new instructions. This scenario illustrates how the ADCs are a central part of the patients’ clinical care and are executed as a routine. Supervisors closely monitor these routines, that we called the routines of daily care (RDC) for managing the work rhythms in hospitals. Understanding the Elders’ Medication Routines Ambient assited living environments aimed at allowing elders to “age in place”. These environments help elders to carry out the activities of daily living, such as managing their medication –a critical activity for independent living. To identify the kind of support that the elderly may need for medicating in their homes, we carried out a qualitative study following the activity-centered design approach, which allowed us to understand and model how elders perform these activities [García-Vázquez et al, 10]. It consisted of 40-minute semi-structured and contextual interviews based on the MedMaIDE index, which is an instrument to determine the deficiencies of elders to adhere to their medication prescription [Orwing, 96]. The participants were 17 elders ranging in age from 63 to 86 years old, who were able to live independently in their homes. The study results show that elders have created their own strategies for adhere to their medication prescription, such as having a specific place to medicate, maintaining notes for indicating the purpose of taking the medicines and visiting periodically their doctor for refilling their medicines. Modeling elders’ medication management routines enabled us to identify and represent how elders carry out actions using supporting resources that help them to accomplish their medication routines, including the community that helps them to refill their medications, and the objects or tools used to take their medicines. Thus, we identified the actions sequence followed by elders when it is time for medicating. For instance, elders move to the place where they usually medicate, such as the kitchen; then they start to perform different actions by interacting with the objects or tools they need to medicate such as medicine container. We also identified that elders need help from their children to periodically visit their doctor Figure 1 The RDC Compass System (a) A nurse’s list of pending tasks (b) The timeline view of nurses’ executed activities and (c) A nurse’s routing visualization plot Our results helped us to envision scenarios that show how activity-aware computing could augment smart hospital environments. For instance, in the following scenario, we illustrate how scheduling and reminder services aware of nurses’ activities can help them to perform their activities of daily care (ADC) while 60 and get their medicines. The understanding we gained from the elders’ medication routines, helped us to envision scenarios which illustrates how activity-aware computing could augment the elders’ home environments to assist them: “Mrs. Maria is 72 years old. It is noon and while she is preparing her meals, her wearable notification display generates an audio notification to make her aware that it is time to take her medicines (figure 2a). When Maria approaches her medicine dispenser the e-Activity in her wearable display moves to the dispenser to emphasize the two containers with the medicines that she need to take (figure 2b)”. then decide to interrupt his activity, which will cause the eActivity to move to a suspended state, such as the nurse Letty suspends her current activity to help Dr. Guillen. As a consequence, a user may also resume an e-Activity when going back and forth from one activity to another [Bardram, 07]. The main elements of these models are: the activity executed by the user, its attributes and execution context. The CARe Ontology The Context Aware Representation of e-Activities (Figure 3), CARe was implemented as an ontology that represents: Executed Activity. As described in scenario 2, an Activity is composed of Actions or sub-Activities; for example, when the elder is going to perform her medication routine, she first goes to her medicine dispenser, selects the appropriate doses, and then takes a glass of water to take her pills. Additionally, an activity can be linked to other activities, either sequentially or given a pre-specified order ―i.e., when a nurse gives medication to a patient following a physician’s diagnostic plan. Execution context. The target device that allocates an eActivity determines its execution context. These devices may be objects available in the environment which could be categorized as: Basic Objects, if they are indispensable to execute the activity, i.e. the pills containers and the medical card used to schedule the next visit to the doctor, and as Complementary Objects which enhance the execution of an activity, such as the elder using a bracelet to receive timely reminders for taking her medicines. The CARe ontology represents the objects users employ when executing an activity. This enables activity-aware applications to determine the activity being executed and decide what e-Activity would support and where it should run (i.e. medicines containers execute e-Activity to aid elders to take their pills in order to prevent them from taking inappropriate doses). Thus, an e-Activity, allocated in an object, has an execution State indicating if it is active, suspended, resumed, or has moved. Activity Attributes and Properties. From our studies we identified that activities are characterized by the following Attributes: the Person who executes the activity, the StartTime and FinishTime of the activity and the Location where the activity is being executed. An activity may have properties, such as the Frequency of executing it, the Schedule followed, and Community, which indicates that other persons participated in the activity. Developers can customize the CARe ontology by adding domain-centric properties, which are particular characteristics of a specific application domain. Attributes and Properties are used to determine if a CARe Rule has been met. a) b) Figure 2. Activity-aware system providing aids for medicating on different objects: a) Mobile medication reminder; b) Medicine container providing ambient aids. e-Activity: a Computational Representation of Human Activities In the aforementioned application scenarios, an e-Activity is the computational representation of the user’s activity, and it is defined by its attributes, its execution model and behavior [Tentori, 11]. The attributes of an e-Activity are essential characteristics that determine its uniqueness and depict its execution context. Such attributes include user and environmental contextual information and the rules that define how the environment expects users to act when some contextual conditions are met. The e-Activity attributes include contextual information regarding who is performing the action, the location in which it is being executed, and the used objects or artifacts. Thus, an eactivity follows an execution model according to the way users execute their activities in real life. For example, to enable an elder to independently perform their medication activity, an e-Activity is aware of her context to autonomously activate an action plan composed by a set of actions or operations that illustrate how such activity must be executed, i.e the e-Activity jumps to the dispenser to provide ambient aids that indicate what medicines to take. An e-Activity may have different states according to their execution model [Bardram, 07]. For instance, a user perceives interruptions or changes in the environment, and 61 Figura 3. CARe-Ontology Agent-based Architecture for Developing Activity-Aware Systems for Assisting Elderly. J. UCS 16(12):1500-1520. Gu, T.; Keng-Punga, K.; and Da-Qing, Z.; 2005. A serviceoriented middleware for building context-aware services. J. of Network and Computer Applications 28: 1–18. Rodríguez, M. D.; Favela, J.; Preciado, A.; and Vizcaino, A., 2005. Agent-based ambient intelligence for healthcare. AI Communications 18(3):201-216. Tentori, M.; and Favela, J. 2008. Activity-aware computing in Healthcare. IEEE Intelligent Systems 7 (2): 51-57 Tentori, M.; Rodríguez, M .D.; and Favela, J. (in press). An Agent-based middleware for the design of activity-aware applications. IEEE Intelligent Systems Cassens, J.; and Kofod-Petersen, A. 2006. Using Activity Theory to Model Context Awareness: a Qualitative Case Study. Proc. of the 19th International Florida Artificial Intelligence Research Society Conference, 619-624: AAAI Press, Kaenampornpan, M., O'Neill, E., 2004. Modelling Context: An Activity Theory Approach. Proceedings of Ambient Intelligence: Second European Simposium, EUSAI, Ed. Springer, pp. 367-374. Orwing, D.; Brandt, N.; and Gruber-Baldini, A. 2006. Medication management assessment for older adults in the community, The Gerontologist, 2: 661-668. Soldatos, J.; Dimakis, N.; Stamatis, K.; and Polymenakos, L. 2007. A breadboard architecture for pervasive context-aware services in smart spaces: middleware components and prototype applications. Personal & Ubiquitous Computing (1):193–212. The CARe ontology was implemented with the OWL language enabling applications that were not developed on the top of SALSA to access it. For instance, an Electronic Health Record System may be updated based on the CARe ontology content. Conclusions and Future Work Developing representational models for domain specific environments may facilitate implementing activity-aware applications that not only provide opportunistic services and information, but that learn from and adapt to users behaviors changes. Considering this, we plan to extend the proposed activity-aware mechanisms to determine how the CARe ontology facilitates the identification of users behavior patterns from their activities histories. This will facilitate the development of activity-aware applications that infer risks associated with an activity (i.e an elder taking a long nap) or to adapt them to new adopted behaviors and routines changed by the doctor. References Bardram, J.; and H.B. Christensen, 2007. Pervasive Computing Support for Hospitals: An Overview of the Activity-Based Computing Project. IEEE Pervasive Computing 6(1): 44–51. Bochini, C.; Curino, C.; and Quintarelli, E. 2007. A data-oriented survey of context models. ACM SIGMOD Record 36(4):19-26. García-Vázquez, J.; Rodríguez, M.D.; Tentori, M.; SaldañaJimenez, D.; Andrade, A.G.; and Espinoza, A.N., 2010. An 62