An Efficient and Asymmetrical Feature Mapping Model for Measuring Semantic Similarity

The semantic similarity estimation is an important basic research topic in conceptual modeling. This paper proposes a high-efficiency and asymmetrical feature mapping model for computing semantic similarity. Firstly, to improve efficiency, we take the edge as the main information source of mapped features for concepts. Then, to avoid the computed deviation with the human judgment in the low score range is more significant than that in the high score range, we map the commonalities between concepts only into the single information source of the weighted depth of their Least Common Subsumer (LCS), and map their differences into three types of information sources, including their weighted path, density and encoding distance, to improve the ability to distinguish between different concepts. To further improve efficiency, we consider only the direct hyponyms of LCS as its density, rather than all of its hyponyms. Moreover, the edge weight is introduced into the calculation of depths and paths to reduce the influence of high-layer edges. The experimental results show that the proposed model is an excellent semantic similarity method with high computational efficiency and high measurement accuracy on both the common ontology WordNet and the biomedical ontology SNOMED CT.