Improving Knowledge Extraction from Texts by Generating Possible Relations

Existing research focus on extracting the concepts and relations within a single sentence or in subject-object object pattern. However, a problem arises when either the object or subject of a sentence is "missing" or "uncertain", which will cause the domain texts to be improperly presented as the relationship between concepts is no extracted. This paper proposes a solution for the enrichment of the knowledge of domain text by finding all possible relations. The proposed method suggests the appropriate or the most likely term for an uncertain subject or object of a sentence using the probability theory. In addition, the method can extract the relations between concepts (i.e. subject and object) that appear not only in a single sentence, but also in different sentences by using a synonym of the predicates. The proposed method has been tested and evaluated with a collection of domain texts that describe tourism. Precision, recall, and f-score metrics have been used to evaluate the results of the experiments.