Ontology-Based Opinion Mining

Background

There are two fundamental research efforts in opinion mining area that have been given attention, namely sentiment classification and property-level opinion mining. Traditional document classification methods are often employed in sentiment classification; the entire document is classified as positive, negative, or neutral. However, these methods must be considered too vague. In most cases, both positive and negative opinions can appear in the same document, for example, “Overall, the accommodation is super, the glacier-hike is thrilling, but the food is tasteless.” There are some works focused on sentiment classification. They are based on manually or half-manually constructed a priori knowledge dictionary which contains polarity words. Some classic machine learning methods (Naive Bayes, Maximum Entropy, and SVM) have been tested by Pang et al. (2002) who point out that machine learning methods are more effective than manually labelling. Hearst (1992) proposed a model-based approach trying to analyze the structure of the document more deeply, but the method is so far not supported by experimental data.