We present a distributed service to model and control contextual information in mobile and ubiquitous computing environments. We describe the general user model and context ontology GUMO for the uniform interpretation of distributed situational information in intelligent semantic web enriched environments. Furthermore, we present the relation to the user model and context markup language USERML, that is used to exchange partial models between different adaptive applications. Our modeling and retrieval approach bases on semantic web technology and complex conflict resolution concepts.
Ontologies provide a shared and common understanding of a domain that can be communicated between people and heterogeneous and widely spread application systems. Since ontologies have been developed and investigated in artificial intelligence to facilitate knowledge sharing and reuse, they should form the central point of interest for the task of exchanging situation models. The user model & context markup language USERML is defined as an XML application, see [6]. However, XML is purely syntactic and structural in nature. Nonetheless, the web ontology language OWL has more facilities for expressing semantics. OWL can be used to explicitly represent the meaning of terms in vocabularies and the relationships between those terms. Thus, OWL is our choice for the representation of user model and context dimension terms and their interrelationships. This ontology should be available for all user-adaptive and context-aware systems at the same time, which is perfectly possible via internet and wireless technology. The major advantage would be the simplification for exchanging information between different systems. The current problem of syntactical and structural differences between existing adaptive systems could be overcome with such a commonly accepted ontology. GUMO1 collects the user’s dimensions that are modeled within user-adaptive systems like the user’s heart rate, the user’s age, the user’s current position, the user’s birthplace or the user’s ability to swim. Secondly, the contextual dimensions like noise level in the environment, battery status of the mobile device, or the outside weather conditions are modeled. The main conceptual idea in SITUATIONALSTATEMENTS, is the division of user model & context dimensions into the three parts: auxiliary, predicate and range. Apart from these mainpart attributes, there are predefined attributes about the situation, the explanation, the privacy and the administration. Thus, our basic context modeling is more flexible than simple attribute-value pairs or RDF triples. If one wants to say something about the user’s interest in football, one could divide this into the auxiliary=hasInterest, the predicate=football and the range=low-mediumhigh. GUMO is designed according to this USERML approach. Approximately one thousand groups of auxiliaries, predicates and ranges have so far been 1 GUMO homepage: http://www.gumo.org identified and inserted into the ontology. However, it turned out that actually everything can be a predicate for the auxiliary hasInterest or hasKnowledge, what leads to a problem if work is not modularized. The suggested solution is to identify basic user model dimensions on the one hand while leaving the more general world knowledge open for already existing other ontologies on the other hand. Candidates are the general suggested upper merged ontology SUMO, see [7], and the UBISONTOLOGY2, see [8], to model intelligent environments. Identified user model and context auxiliaries are for example hasKnowledge, hasInterest, hasBelief, hasPlan, hasProperty, hasPlan and hasLocation. A class defines a group of individuals that belong together because they share some properties. Classes can be organized in a specialization hierarchy using rdfs:subClassOf. 3
The architectural diagram in figure 2 shows the SMARTSITUATIONRETRIEVAL or
smart context retrieval process. The focus is set on the semantic conflict resolution part.
The oval numbers indicate the reading direction. Item (
1. ON THE SEMANTICAL LEVEL: the systems are not forced to use the same vocabulary, to say the same ontology, to represent the meaning of the concepts, which leads to the user model integration problem number one: ontology merging and semantic web integration. 2. ON THE OBSERVATION AND INFERENCE LEVEL: several sensors can see same things differently and claim to be right, measurement errors can occur, systems may have preferred information sources 3. ON THE TEMPORAL AND SPATIAL LEVEL: information can be out of date or out of spatial range, a degree of expiry can hold. Reasoning on temporal and spatial meta data is necessary 4. ON THE PRIVACY AND TRUST LEVEL: information can be hidden, incomplete, secret or falsified on purpose, a system of trustworthiness could be applied
Conflict Resolvers are a special kind of filters that control the conflict resolution process. An ordered list of these resolvers define the conflict resolution strategy. They are modeled in the query.strategy attribute. These resolvers are needed if the match process and filter process leave several conflicting statements as possible answers. Three kinds of conflict resolvers can be identified: the most(n)-resolvers that use meta data for their decision, the add-resolvers that add expired or replaced statements to the conflict sets, and the return-resolvers that don’t use any data for their selection. mostRecent(n) Especially where sensors send new statements on a frequent basis, values tend to change more quickly as they expire. This leads to conflicting non-expired statements. The mostRecent(n) resolver returns the n newest non-expired statements, where n is a natural number between 1 and the number of remaining statements. mostNamed(n) If there are many statements that claim A and only a few claim B or something else, than n of the ”most named” statements are returned. Of course it is not sure that the majority necessarily tells the truth but it could be a reasonable rule of thumb for some cases. mostConfident(n) If the confidence values of several conflicting statements can be compared with each other, it seems to be an obvious decision to return the n statements with the highest confidence value. mostSpecific(n) If the range or the object of a statement is more specific than in others, the n ”most specific” statements are returned by this resolver. For example if: auxiliary=hasKnowledge, predicate=chess and first range=yesNo while the second range=Novice-Occasional-Professional-Expert-Grandmaster, the statement with the second range contains a more specific information. Another specificity range ordering is for example: yesNo < lowMediumHigh < 0%-100% mostPersonal(n) If the creator of the statement is the same as the statement’s subject (a self-reflecting statement), this statement is preferred by the mostPersonal(n) resolver. Furthermore, if an is-friend-of relation is defined, statements by friends could be preferred to statements by others.
These conflict resolver rules are based on common sense heuristics, however they need not to be true for specific sets of statements. An important issue to keep in mind is the problem that resolvers and strategies imply uncertainty. To contribute to this fact, the confidence value of the resulting statement is appropriately changed, furthermore the conflict situation is added to the evidence attribute.
Summary and Acknowledgements We have introduced an integrated architecture for Situation Modeling and Smart Context Retrieval. We have clarified a model for situated interaction and context-awareness. The context exchange language UserML has been presented as well as the general user model & context ontology GUMO. Our approach bases on semantic web technology and a complex conflict resolution and query concept, in order to be flexible enough to support adaptation in human-computer interaction in ubiquitous computing. This work has been supported by the German Ministry of Education and Research within the project SPECTER at the German Research Center for Artificial Intelligence (DFKI).