Week3-2Thesaurus-based Word Similarity Methods

Remember word net

Given 2 words, we could calculate the number of the links between these 2 words in the word net forest(tree). The great the distance, the smaller the similarity.

Path similarity

• Version 1
  
  • Sim(v, w) = -pathlength(v, w)
• Version 2
  
  • Sim(v, w) = -log( pathlength(v, w) )

Problems of the approach

• There may be no tree for the specific domain or language
• A specific word(a term) may be not in any tree
• IS-A(hypernym) edges are not equally apart in similarity space

Advanced version

• Version 3
  $Sim(v,w) = -log P(LCS(v,w))$
  LCS = lowest common subsumer
  e.g. ungulate for deer and horse
  deer for deer and elk

• Version 4