Learning multilinguistic knowledge for opinion analysis

Most existing opinion analysis techniques used word-level sentiment knowledge but lack the learning capacity on the behaviors of context-dependent opinion words. Meanwhile, the use of collocation-level sentiment knowledge is not well studied. This paper presents an opinion analysis system, namely OA, which incorporates the word-level and collocation-level sentiment knowledge. Based on the observation on the NTCIR-6 opinion training corpus, some word-level and collocation-level linguistic clues for opinion analysis are discovered. Learning techniques are developed to learn the features corresponding to these discovered clues. These features are in turn incorporated into a classifier based on support vector machine to identify opinionated sentences and determine their polarities from running text. Evaluations on NTCIR-6 opinion testing dataset show that OA achieved promising overall performance.