3

I am working on a project which involves computation of relatedness between different concepts.

Examples: "landing" - related to: "to arrive; to come" and "land", "telephone" - related to: "electronic; electricity" and "to talk; to communicate", "movie" - related to: "to move; to change" and "picture; image" , and so on.

I have checked the WordNet framework and WordNet::Similarity Perl module, but in some cases they are not usable, especially in cases where I need to connect nouns with verbs (for example a noun "dog" and a verb "to bark"). Also, WordNet organizes all synsets in hierarchical structure (ex: computer -> machine -> device -> instrumentation), and this is not that I really need in my project (of course, I may be wrong).

The question is - which framework or database, what approach should I use to solve the problem? Where do I start?

It would be really helpful if you could give me some advice.

user655136
  • 161
  • 2
  • 10
  • You need to organize similarity ,relations and semantics of your need. I think relation between "landing" and "land" and relation between "landing" and "to arrive" are relations of different semantics. You may need framework which can handle memantics and relations among words. – Fumu 7 Sep 25 '14 at 04:27
  • You should take a shot at exploiting the "derivationally related forms" in WordNet for relating nouns to verbs. But ... the *real* (i.e. research'y) way to understand concept relatedness is by modeling words using their distributional similarity over a large dataset. – Chthonic Project Sep 30 '14 at 18:39

1 Answers1

3

word2vec embeddings might do a good job on your task. It gives a much softer notion of similarity than wordnet. You can get some pre-trained embeddings linked from here https://code.google.com/p/word2vec/ .

Rob Neuhaus
  • 9,190
  • 3
  • 28
  • 37