WuPalmer

WuPalmer, or Wu-Palmer similarity, is a semantic similarity measure used in natural language processing to assess how closely related two WordNet synsets are within its hierarchical taxonomy. It computes similarity by considering the depth of the least common subsumer (LCS) of the two synsets and the depths of each synset themselves. The standard formula is similarity(s1, s2) = (2 × depth(LCS(s1, s2))) / (depth(s1) + depth(s2)). Depth is measured as the number of edges from the root to the node. The result ranges from 0 to 1, with higher values indicating greater similarity.

Wu-Palmer emphasizes commonality near the root of the taxonomy and does not explicitly use information content. This makes the measure simple and fast, suitable for large-scale comparisons or as a baseline. It is commonly applied in tasks such as word sense disambiguation, semantic similarity, information retrieval, and text clustering where WordNet-based lexical relations are informative.

Limitations include sensitivity to the uneven depths of branches in WordNet, which can bias scores toward terminology in deeper or shallower parts of the hierarchy. It does not account for the frequency or context of usage, nor does it integrate statistical information beyond the taxonomy structure. As a result, Wu-Palmer is often used in conjunction with other measures to capture different aspects of semantic relatedness.

The measure was introduced by Wu and Palmer in 1994 and remains a widely cited baseline in WordNet-based similarity studies.