Putting Similarity Assessments into Context: Matching Functions with the User's Intended Operations

This paper presents a practical application of context for the evaluation of semantic similarity. The work is based on a new model for the assessment of semantic similarity among entity classes that satisfies cognitive properties of similarity and integrates contextual information. The semantic similarity model represents entity classes by their semantic relations (is-a and part-whole) and their distinguishing features (parts, functions, and attributes). Context describes the domain of an application that is determined by the user’s intended operations. Contextual information is specified by a set of tuples over operations associated with their respective entity-class arguments. Based on the contextual information, a partial word-sense disambiguation can be achieved and the relevance of distinguishing features for the similarity assessment is calculated in terms of the features’ contribution to the characterization of the application domain.