Semantic Similarity Measures

To enable artificial systems to converse and adapt to users’ needs and desires, which is a fundamental aspect of phenotropic interaction, understanding the semantics of exchanged information is vital. Semantic similarity enables the estimation of the closeness of meaning between various elements. This is key for the processing of information in a human-like manner, allowing to automatically extend knowledge to unknown concepts and perceptions. Semantic similarity measures play a key role in the computing with words pipeline’s precisiation phase and the identification of relationships between concepts in approximate reasoning. Modeling perceived commonalities between elements relies on similarity estimation, which inherently possesses a subjective component. Therefore, similarity cannot be defined crisply but requires a fuzzy nuance to account for the subjectivity and imprecision of natural language. In this chapter, a novel semantic similarity measure taking into account these aspects is defined, covering the specific case of similarity between scalar terms, that is generally not well handled by state-of-the-art methods. In a final evaluation task, the developed measure not only results more accurate than state-of-the-art methods but reaches also a performance similar to that of humans in a semantic sorting task.