Abstract:
Sentiment analysis in the world allows us to extract useful information from the opinions and feelings of followers and customers of an organization. It analyzes the text written in natural language about services and products. Then after, classifying sentiments, opinions, feeling as positive, negative or neutral. However, Collectng data from Facebook of the Amharic News page is challenging. In addition to this, the data is huge and not easy to understand customers feeling and opinion. To solve this problem design Aspect level approach to mine the overall sentiment or opinion polarity from the data is needed.
We followed Aspect/Feature level opinion mining in detail to meet customers and organizations need. We employed crafted rules using rule-based for labeling data and supervised approach to training and testing the data. Research flow seven major components of the architecture which includes data collection, preprocessing of data, features extraction, Morphological Analyzer, aspect extraction, aggregate opinions sentiment classification, and result. Support Vector Machine (SVM) and Naive Bayes (NB) classifiers were used for sentiment classification. The collected Amharic opinionated sentences and phrase texts from Amhara Mass Media Facebook page were 1200. Among those, 960 data for training (80%) and 240 data for testing (20%). Experiments indicated that the bag of word module feature extraction methods performs the best through two algorithms (NB and SVM). The result showed as Naïve baye’s precision, recall and F-measure evaluation metrics 84%, 80% and 81 % respectively. For SVM precision, recall and F-measure evaluation metrics are 87%, 82% and 84 % respectively.
Therefore, as to conclude, the SVM revealed the best category of the customers’ sentiment and opinion, which gives a valuable approach for researchers and other users.