DOMAIN ORIENTED SYNTAX BASED ASPECT DETECION FOR STUDENT FEEDBACK SYSTEM OF UCS Taungoo

Abstract

Opinion Mining becomes popular in seeking the information on online review or feedback system. This technique can be usually used in recommender system that supposes the customers for making trust upon the products based on other user’s opinion. Moreover, this technique can also help the development or maintenance of different kinds of products or activities by evaluating the users’ opinion. In conventional opinion mining techniques, it can examine the people feeling from their reviews or comments such as positive or negative only. The process of examining such positive and negative score is also known as sentiment analysis. Sentiment analysis can be applied at different levels of scope such as sentence level, document level and aspect level. So, in the current trend, the goal of sentiment analysis is to dig the aspect word that is the fine grained sentiment information based on the reviews or comments of various domains. So, the proposed system aims to analyze the aspect level sentiment analysis on student feedback system. The required feedback data are collected from the University of Computer Studies, Taungoo(UCST). First step of sentiment analysis is part-of-speech tagging (POS tagging) that can identify the form of each word in the sentence. For POS tagging, this system uses OpenNLP parser which parses the sentence as adjectives, verbs and nouns, respectively. For defining the sentiment score of each word, this system uses sentiWordNet lexical resources by applying the SWN3 algorithm which finds the score of each word in lexical resource and attaching with this word. In order to dig the aspect word for feedback statement, the Domain Specific Ontology relating to UCST is created in the preprocessing stage of this system which composed with the main aspect words of the domain. Finally, the proposed algorithm Onto-to-List can definitely find the matching aspect word from the feedback statement by confirming the domain specific ontology. This system is evaluated by using confusion matrix and the accuracy measurement based on the prediction of user’s opinion. The accuracy of this system is 94% that is evaluated over 100 history records of this system. This system will assist the administrator of UCST to evaluate the performance of the University.