SENTIMENT MINING AND ASPECT BASED SUMMARIZATION OF OPINIONATED AFAAN OROMOO NEWS TEXT

Abstract

The inventions of web 2.0 have paved the way for the rapid growth of user-generated content. Now we are in the world in which masses of user-generated content (i.e. word of mouth) are easily unrestricted online on the web in different domain. This massive number of user-generated content can be used as source to mine the sentiment orientation of the opinion holder as positive or negative. This massive number of user-generated content can be difficult manually to collect, understand, summarize, analyses for decision making. For this reason, studying on this particular topic of opinion mining has attracted an attention of many researchers to tackle this obstacle. Opinion mining can be done in one of the three different levels: sentence, document and feature/aspect level. In the middle of these three levels, feature level opinion mining is the detail and complex one but has a better benefit to meet customers and organizations need. While there are many feature level opinion mining models that have been undertaken for international and local languages, to the level of the researcher knowledge opinion mining and summarization at aspect level in Afaan Oromo language is never done until now. Therefore, this study investigates and aims to develop sentiment mining and aspect based opinion summarization of service review in Afaan Oromo language for Oromia Radio and Television Organization (ORTO). A total of 400 reviews collected from Oromia Radio and Television Organization in news domain is used for the study. Developed feature level opinion mining model mainly consists of five components: ORTO document review, pre-processing, aspect extraction, polarity detection, and aspect based sentiment summarization and draw bar chart to represent aspect based opinion polarity graphically. Experimental result shows that for positive class precision of 90% and recall of 87%, whereas for negative class precision 87% and 89.7% recall is achieved. The major challenge observed in this study is polarity of opinions. This is because user mostly give their opinion in context based or/and indirect manner. They may use positive words to provide their negative feeling or vice versa. Thus it needs further research to make the system consider context based or semantic opinion mining