Supervised and unsupervised learning algorithms have been applied to discover ontological concepts and relations from the extracted information. During concept learning, words/phrases that only perform grammatical functions and words that are unlikely to carry domain-specific meanings are filtered out using information retrieval techniques. The candidates for domain concepts can be sorted in the descending order of the strength of their relationships with other candidates. Relation learning generally relies on co-occurrence statistics of information. Approaches to relation learning vary in terms of the scope of co-occurrence, the metrics for the significance of co-occurrence, the criteria for selecting candidate concepts, and the thresholds for extracting potential relations. Some learning approaches require the assistance of domain-specific resources such as thesauri.