Recognizing Intentions in Free Text Messages: Studies with Portuguese Language

Recent literature indicates that user intention analysis brings benefits for several computational tasks including information retrieval and communication. However, intentions are expressed implicitly in natural language texts. Domain related specificities and cultural language aspects hamper their machine representation and interpretation. This requires thorough investigations of intention recognition methods in free text to permit further exploring them. In this paper, we propose a technique based on the matching with representative key phrases and semantic extension of terms to detect instances of intention classes in natural language sentences. We explore a multidimensional framework of illocution categorization to structure the distinct intention classes. The conducted experiments with Portuguese language datasets of different characteristics reveal the potentialities of our method when analyzing the outcome of state-ofthe- art machine-learning based text-mining techniques.