Artificial Bee Colony Optimization for Feature Selection in Opinion Mining

Opinion mining, a sub-discipline of information retrieval and computational linguistics concerns not with what a document is about, but with its expressed opinion. Feature selection is an important step in opinion mining, as customers express product opinions separately according to individual features. Earlier research on feature-based opinion mining had many drawbacks like selecting a feature considering only grammatical information or treating features with same meanings as different. However this led to a large corpus which subsequently affected the classification accuracy. Statistical techniques like Correlation Based Feature (CFS) have been extensively used for feature selection to reduce the corpus size. The selected features are sub optimal due to the Non Polynomial (NP) hard nature of the technique used. In this work, we propose Artificial Bee Colony (ABC) algorithm for optimization of feature subset. Naïve Bayes, Fuzzy Unordered Rule Induction Algorithm (FURIA) and Ripple Down Rule Learner (RIDOR) classifiers are used for classification. The proposed method is compared with features extracted based on Inverse Document Frequency (IDF). Hence, this method is useful for reducing feature subset size and computational complexity thereby increasing the classification accuracy.