I need to measure semantic similarity between CSO ontology concepts, depending on Ontology structure (concept path, depth, least common subsumer (LCS) ...).
CSO (Computer Science Ontology) is a large-scale ontology of research areas.
A concepts in CSO may have multiple parents/super concepts (i.e. a concept may be a child of many other concepts)
e.g. :
(world wide web) is parent of (semantic web)
(semantics) is parent of (semantic web)
i found some measures that meet my needs, but the papers proposing these measures are not cited, so i got hesitated.
I also found a measure that depends on weighted edges, but multiple inheritance (super concepts) is not considered..
Anyway, I am not sure if it is better to consider all super concepts
Any help would be appreciated. Thanks in advance
Peter Ludemann :
i will be using Ontology to measure how much a text is similar to a topic (Ontological concept)
If your training set is something like the Reuters Corpora (https://trec.nist.gov/data/reuters/reuters.html), I don't think you need a notion of ontological similarity; just measure document similarity with the labeled set (I got good results using kNN).
Whether you use a labeled set or not, it's a good way to validate your metric. But the metric might be domain-dependent ... I suspect that a good metric for medical terminology wouldn't work so well for the Reuters ontology. Speaking of which, try searching for "SNOMED similarity" -- there are some papers (I haven't read any of them).
The Ontology that i want to use is CSO (Computer Science Ontology).
It is an Ontology for computer science topics.
I read about a measure that considers LCS , hypernyms and hyponyms of the compared Ontology concepts. This measure works when there is multiple inheritance, but i don't know how to validate results, i mean how to estimate whether returned similarity values are reasonable or not.
https://ieeexplore.ieee.org/abstract/document/8367176
You might also consider the techniques used in "fuzzy unification". See, for example the "unification degree" in https://swi-prolog.discourse.group/t/bpl-3-2-released/519
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators