If we Consider two Named Entity Relation systems, one based on the use of word embedding and the other using both word and character embedding jointly. How can we intuitively conclude which model is better for NER task? Is any illustration possible for above case?
If I understand your question correctly, you may train (if they require training) your models on some annotated dataset like CONL-2003 https://www.clips.uantwerpen.be/conll2003/ner/ and compare their accuracies using the test-part of the dataset.