Article Preview
TopThe semantic similarity approaches are developed to measure the extent to which two concepts are similar using the structural information collected from concept taxonomy or information content. The input to the semantic similarity approach is a concept pair and output is a numerical value showing semantic similarity between the concepts. There are many applications using the semantic similarity value to rank the similarity between different concept pairs.
The knowledge-based approaches (Rada et al., 1989),(Leacock & Chodorow, 1998),(Wu & Palmer, 1994),(Li et al., 2003), compute the semantic similarity between the two ontology concepts using semantic information contained in an ontology. The knowledge-based approaches can be further categorized based on how the semantic similarity between concepts is assessed, as edge-counting approaches, information content-based approaches, feature-based approaches, and hybrid knowledge-based approaches.
The edge counting approaches count the edges in the path connecting two concepts in ontology to compute the similarity between them. If the distance between the two concepts is large then the semantic similarity between them is less. In contrast, if the distance between the two concepts is small then the semantic similarity between them is more.
Rada et al. (1989) stated path-based approach that calculates the semantic similarity between two concepts Conc1 and Conc2 by using the shortest path length represented as SPL (Conc1, Conc2) as (Rada et al., 1989)
(1)Leacock, Claudia, and Martin Chodorow (1998) stated lch approach that computes the semantic similarity between two concepts Conc1and Conc2 using a non-linear function illustrated as (Leacock & Chodorow, 1998)
(2) where SPL (Conc1, Conc2) is the shortest path length and MAX_DEPTH is the maximum depth of the ontology.
The disadvantage of edge-counting approaches is that all concept pairs having same path length gives the same semantic similarity value.
Figure 1.
WordNet “is-a” hierarchical taxonomy fragment