Similarity in Context: Improving Semantic Similarity Rankings with Contextual Information (SimCat II)

Basic data for this project

Description

The overall goal of this research is the development of a generic formal framework for similarity measurement in information retrieval from semantically annotated data. The foundations for this framework have been successfully laid out with the development of the SIM-DL theory in phase I. Building on research in psychology which has shown the strong context-dependence of human similarity judgments, we want to develop SIM-DL into a framework for the computation of cognitively plausible similarity rankings with respect to a contextual situation. We thus propose a second phase that focuses on an extension for context awareness. This requires the development of a formal theory of context for similarity measurement, which is to be integrated with the existing SIM-DL specification and implementation. The proposed phase II focuses on the qualitative and quantitative variations context changes cause. The qualitative aspect focuses on what changes when the context changes. We propose an approach based on context rules to adapt the knowledge base to the current situation. The quantification of the variations in similarity rankings caused by context changes requires a measurement method that reflects human assessments of these variations. Moreover, we want to investigate the implications of context dependencies in the concept definition process and how they eventually affect similarity among these concepts.