A hybrid approach for identifying sentiments around aspects

In this paper, we have discussed the development of a complete solution for identification of Voice of Customer (VOC) at the mobile domain. We have defined 27 possible aspects and 3 possible sentiments in this domain. The task is to identify the aspects a specific reviewer is talking about and the sentiments he conveys. The words (terms) and the types (tags) of aspects are identified using a statistical sequence leveling technique (based on CRF) trained on 950 manually annotated sentences and the sentiments of a sentence are identified using a statistical bag-of-word technique (based on SVM) trained on 5450 manually annotated sentences. The association of sentiments with aspects is carried out using rules. The system identifies sentiment around aspect with 70.08% accuracy. This would help the customers to get the insight of the product he or she is looking for and the company to get the real VOC.