Mining Causality Knowledge from Textual Data

Mining causality knowledge will induce a knowledge of reasoning that is beneficial for our daily use in diagnosis. Then, this framework is for discovering causality existing between causative antecedent and effective consequent discourse units. There are three main problems in the causality or cause-effect extraction; cause-effect identification, causality ordering and cause-effect boundary determination. The cause-effect identification and the causality ordering problems can be solved by learning verb pairs among different elementary discourse units and learning lexico syntactic pattern (i.e., NP1 V NP2) within a single elementary discourse unit from annotated corpus, by using the Naive Bayes classifier. WordNet will be used in this learning for providing the concept for the verb pairs and NP pair of the lexico syntactic pattern after translation of Thai words to English words by using the Thai-English dictionary. The cause-effect boundary determination problem can be solved by using centering theory and interesting cue phrase or causality link, where the interesting cue phrase would include the discourse markers and verb phrases. Our model of causality extraction shows the precision and recall of 86% and 70% respectively, where our evaluation is based on the expert's results.