FEATURE LEVEL OPINION MINING FOR AFAAN OROMOO MEDIA PROGRAM SERVICE REVIEWS TEXT ANALYSIS

MAJOR ADVISOR: - GETACHEW MAMO (Ph.D) Nowadays, opinions are found on the Internet everywhere and anytime. Automatically mining and organizing opinions from heterogeneous information sources are very useful for individuals, organizations and even governments. Opinion mining is the computational study of people’s opinion expressed in written language or text towards their feature. With the explosive growth of social media on the Web, organizations are increasingly relying on opinion mining methods to analyze the content of these media for decision making. As Afaan Oromo opinionated text on media program service increasing rapidly, there is no mechanism for people or organization to analyze all opinions at a time to make good decisions. To overcome this challenge, in our work we have proposed feature level opinion mining for Afaan Oromo reviews texts in OBN entertainments program service domain. The proposed model consists of four major modules that can perform tasks sequentially; identify and extract the feature of service, identifying opinions along with the extracted feature, determine their orientation and finally summarize the reviews by grouping multiple opinions along features. Once these task are accomplished by applying different opinion mining techniques. The task of identifying and extracting features in the review sentences are done by using two knowledge- poor based approach technique .The first one is by using POS, finding noun and noun phrase in the given sentences review. The second one is by using sequentially pattern rule (SPR) extracting feature. All features obtained by both techniques are set as candidate feature of service program. However, not all candidate feature results generated are relevant features. Therefore, feature Pruning was done by using frequency based approach to extract relevant feature and to detect unlikely features. For determining opinion and orientation at each feature we used opinion lexicon we have developed for media program service reviews text. After all process, finally feature-opinion-orientation pairs are summarized and visualized by table and bar chart form. To evaluate the performances of the systems; we have collected 420 reviews sentences from OBN media entertainment program service domain. The experiment is conducted using two methods, to evaluate in three perspectives (feature, opinion and orientation). Accordingly, 90 % average precision, 75% average recall for the features extraction, 85% average precision, 89% average recall for opinion words determination and 88% average precision, 75% average recall for opinion orientation at each feature performances were obtained by the first method. The second method achieves an average precision of 89% and an average recall of 67% performances for features extraction, 91% average precision and 81% average recall for opinion words determination, 87% average precision and 67% average recall for opinion orientation at each feature performances were obtained. The results show that this study is promising. Key Words: Opinion Mining, Feature Level Opinion Mining, Media Program Service Reviews, Knowledge-Poor Approach, Frequency Based Approach, Sequentially Pattern Rule (SPR), Feature, Opinion, Orientation, and Summarization.