This measure calculates relatedness by considering the depths of the two synsets in the WordNet taxonomies, along with the depth of the LCS
WUP(s1, s2) = 2*dLCS.depth / ( min_{dlcs in dLCS}(s1.depth - dlcs.depth)) + min_{dlcs in dLCS}(s2.depth - dlcs.depth) ), where dLCS(s1, s2) = argmax_{lcs in LCS(s1, s2)}(lcs.depth).
Parameters
min score = 0.0
max score = 1.0
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
JCN
Latency / Throughput
Description of JCN
Originally a distance measure which also uses the notion of information content, but in the form of the conditional probability of encountering an instance of a child-synset given an instance of a parent synset.
JCN(s1, s2) = 1 / jcn_distance where jcn_distance(s1, s2) = IC(s1) + IC(s2) - 2*IC(LCS(s1, s2)); when it's 0, jcn_distance(s1, s2) = -Math.log_e( (freq(LCS(s1, s2).root) - 0.01D) / freq(LCS(s1, s2).root) ) so that we can have a non-zero distance which results in infinite similarity.
Parameters
min score = 0.0
max score = Infinity
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
LCH
Latency / Throughput
Description of LCH
LCH relies on the length of the shortest path between two synsets for their measure of similarity.
LCH(s1, s2) = -Math.log_e( LCS(s1, s2).length / ( 2 * max_depth(pos) ) ).
Parameters
min score = 0.0
max score = Infinity
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
max depth N = 20
max depth V = 14
LIN
Latency / Throughput
Description of LIN
Idea is similar to JCN with small modification.
LIN(s1, s2) = 2*IC(LCS(s1, s2) / (IC(s1) + IC(s2)).
Parameters
min score = 0.0
max score = 1.0
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
RES
Latency / Throughput
Description of RES
Resnik defined the similarity between two synsets to be the information content of their lowest super-ordinate (most specific common subsumer)
RES(s1, s2) = IC(LCS(s1, s2)).
Parameters
min score = 0.0
max score = Infinity
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
PATH
Latency / Throughput
Description of PATH
This module computes the semantic relatedness of word senses by counting the number of nodes along the shortest path between the senses in the 'is-a' hierarchies of WordNet.
PATH(s1, s2) = 1 / path_length(s1, s2).
Parameters
min score = 0.0
max score = 1.0
error score = -1.0
acceptable pos pairs = [['n', 'n'], ['v', 'v']]
use all senses = true
use root node = true
LESK
Latency / Throughput
Description of LESK
Lesk (1985) proposed that the relatedness of two words is proportional to to the extent of overlaps of their dictionary definitions. This LESK measure is based on adapted Lesk from Banerjee and Pedersen (2002) which uses WordNet as the dictionary for the word definitions. Computational cost is relatively high due to combinations of linked synsets to explore definitions, and need to process these texts.
LESK(s1, s2) = sum_{s1' in linked(s1), s2' in linked(s2)}(overlap(s1'.definition, s2'.definition)).
This relatedness measure is based on an idea that two lexicalized concepts are semantically close if their WordNet synsets are connected by a path that is not too long and that "does not change direction too often".
Computational cost is relatively high since recursive search is done on subtrees in the horizontal, upward and downward directions.
HSO(s1, s2) = const_C - path_length(s1, s2) - const_k * num_of_changes_of_directions(s1, s2)
