Sentiment Analysis for Afaan Oromo Sentence Classification using deep learning Approaches

Abstract

Sentiment analysis has recently become one of the growing research fields related to natural language processing and machine learning. An opinion can be positive, negative, or neutral, it depends on each individual's judgment or assessment of a topic. Social media now plays a vital role in influencing people's emotions for or against a government or organization. Therefore, to understand the sentiment of any social media post, an effective method is the ultimate necessity. We analyzed several social media posts to understand social sentiment. Within this broad scope, we performed this research by applying the most advanced sentiment analysis technology on AO using deep learning in the social domain. Subsequent preparation of the dataset on another domain will improve the language. So, in this study, let's try to display data extracted from OBN official page using Social Media Graph Application interface on PR issues and prepare data for the process subsequent preprocessing. Therefore, after crawling from the OBN using the post id, all the pre-processing, tokenization, stop word removal, and phrase stemming steps will be performed. Manual annotation of sentences extracted from data containing both text files and Emojis annotated using language experts into three classes, positive, neutral and negative, see examine the impact of the most popular Emojis. For the classifiers, we used an 80% train and 20% test rule. We used tokenization, stop word removal, normalization, and derivation as preprocessing, and tf-idf was used as feature extraction. Performance of proposed approaches RNN and Transformer models achieve 90% and 91% accuracy on Facebook, Twitter, and YouTube OBN datasets with corpus size of 2051, respectively.