A group of methods evaluate synset relation quality by using implicit relations between synset. For example, Nadig et al. (2008) proposed an approach that receives two synsets as input and states and judge whether they have a hypernym-hyponym relationship. In their method, firstly ”prefix forms as an indicator of hypernymy”, secondly ”using web search to validate hypernymy”, and lastly “using coordinate terms to validate hypernymy”.

Implicit Relations

Special relations add more semantic properties on the synset lattice in lexicons. For example,  Hayashi (2016) proposed an approach that analyzed evocation relation that might hold between a pair of lexicalized concepts evocation dataset. They first used neural network model to classify evocation relations into 4 categories, and then used forest regression to predict evocation strength.

Special Relations

There are two types of synset relations corresponding evaluation methods:

1)	Implicit Relations

2)	Special Relations


University of Notre Dame

A good LSR depends on synset correctness and completeness, and also depends on the connectivity degree of the synset with other synsets in the semantic network. There are different evaluation methods for each part of the LSR:

1)	Synset Lemmas Evaluation Methods

2)	Synset Gloss Evaluation Methods

3)	Synset Relations Evaluation Methods


Quality Evaluation Methods for LSR

Khalilia, H., Freihat, A. A., & Giunchiglia, F. (2021). The Quality of Lexical Semantic Resources: A Survey. In Proceedings of The Fourth International Conference on Natural Language and Speech Processing (ICNLSP 2021) (pp. 117-129).

The Quality of Lexical Semantic Resources: A Survey

Based on how the synset lemmas are assessed, synset lemmas evaluation methods can be categorized as:

1)	Lemmas Validation Methods

2)	Lemmas Clustering Methods.



Synset Lemmas Evaluation Methods

Synset gloss contains background information about the synset. To evaluate it, Zhang et al. (2011) introduced a model that can measure semantic relatedness between gloss and synset, where they used the WordNet gloss and semantic relations as features that builds concept vectors. 

Learn Before

Related

Learn After