Information retrieval can benefit from contextual information to adapt the results to a user’s current situation and personal preferences. In this respect, semantics-based information retrieval is especially challenging because a change in context may require modifications to the knowledge base at hand, such as updates to or reclassifications of individuals. This thesis introduces a novel approach for context-aware semantics-based information retrieval that covers two aspects.
First, context-aware system design requires an identification of relevant contextual information. For information retrieval, the impact of a contextual aspect on the query results determines its relevance. Performing the same query in different contexts often leads to different result rankings. The comparison of such rankings can provide insights into the effects of context changes on the information retrieval results. While numerous methods exist for assessing the result relevance with respect to a query, the question how different two result rankings are has not been tackled yet. The first part of this thesis is therefore concerned with the definition of a cognitively plausible dissimilarity measure for information retrieval results (DIR). It is based solely on the results and thus applicable independent of the retrieval method. The DIR measure supports cognitive engineering tasks, such as work flow and user interface design: Using DIR, developers can identify which contextual aspects strongly influence the outcome of the retrieval task and should therefore be in the user’s focus. DIR’s purpose is to reflect how human users quantify the changes in information retrieval result rankings. Its cognitive plausibility has been evaluated in two human participants tests, which show a strong correlation with user judgments.
Second, the relevant contextual aspects have to be modeled in a way that supports interaction with semantics-based knowledge bases. The Semantic Web is based on nominal data and it is therefore inherently difficult to integrate information from the Sensor Web, which is an increasingly important source of contextual information. The second part of this thesis introduces an approach based on semantic rules that bridge these two worlds to enable context-aware information retrieval from the Semantic Web. It demonstrates how user preferences can be modeled in the Semantic Web Rule Language (SWRL). SWRL’s support for rules with free variables allows for reasoning on the individuals in an ontology – in the running scenario, the current conditions at surf spots in California are compared against a user model and ranked on the basis of their deviation from a user’s preferences. Moreover, novel SWRL built-ins are introduced to dynamically read observations from the Sensor Web during rule execution, and to perform queries by example on individuals’ data type values. This approach allows for a strict separation of static knowledge about individuals in an ontology and any dynamic information through an explicit link to sensors.