Aspect-oriented sentiment analysis is done in two phases like aspect term identification from review and determining related opinion. To carry out this analysis, features play an important role to determine the accuracy of the model. Feature extraction and feature selection techniques contribute to increase the classification accuracy. Feature selection strategies reduce computation time, improve prediction performance, and provides a higher understanding of the information in machine learning and pattern recognition applications etc. This work specifically focuses on aspect extraction from restaurant review dataset but can also be used for other datasets. In this system, we proposed a multivariate filter strategy of feature selection which works on lemma features. This method helps to select relevant features and avoid redundant ones. Initially, the extracted features undergo preprocessing and then the “term-frequency matrix” is generated which contains the occurrence count of features with respect to aspect category. In the next phase, different feature selection strategies are applied which includes selecting features based on correlation, weighted term frequency and weighted term frequency with the correlation coefficient. The performance of weighted term frequency with correlation coefficient approach is compared with the existing system and shows significant improvement in F1 score