Sn-gram and its Application in Sentiment Analysis

CNN-based deep learning method has been largely used in sentiment analysis of film review, and this provides insight into the appreciation, production and marketing of film literature. In CNN model, the key is to provide high-quality word vectors. At present the word vector is obtained through word pairs formed by a word and its neighbors. The word pairs may not necessarily bear any semantic relations, thus posing potential drawbacks to the quality of word vectors. In this research sn-gram (syntactic n-gram) model is used to produce word vectors. Word pairs for the training of word vectors are obtained by a word and another one which is dependent on it syntactically. Word pairs thus produced are syntactic 2-gram, and there must be some kind of semantic relation between them. Word vectors obtained thus are used in sentiment analysis of film review. Two methods are used: neighbour-based skip-gram and dependency-based sn-gram. As train data, there are 10000 sentences with 5000 positive and 5000 negative; as testing data, there are 10000 sentences with 5000 positive and 5000 negative. Training and testing are carried out under CNN framework. It is found that sn-gram achieved a higher value of accuracy and recall as compared to skip-gram.