A comprehension-based framework for measuring semantic similarity

We present a comprehension-based framework for measuring semantic similarity between documents of text. In various situations, vector-based similarity measures fail to capture deep semantic relations between terms. Our computational comprehension model processes textual content in a way that resembles human readers, paying attention to context, location, and acquisition time of semantic concepts. The model extracts key semantic structures that are representative of the document. These semantic structures are compared using the WordNet WUP measure giving a Semantic-similarity score of the processed documents. Three experiments are illustrated comparing our results with three popular vector-based similarity measures and human readers. Our framework provided correct results in cases where vector-based methods fail. These results highlight the importance of using computational cognitive methods, such as comprehension models, in semantic analysis and text mining.