Customer Feedback Analysis in Afan Oromo Texts

Title: Customer Feedback Analysis in Afan Oromo Texts: A Lexicon-Based Sentiment Analysis ModelThis study introduces a novel hybrid sentiment analysis framework tailored for Afan Oromo an under-resourced yet widely spoken Ethiopian language using a custom-built subjectivity lexicon and a rule-based approach. The model comprises nine core components: text preprocessing, morphological analysis to handle rich inflection, grammar checking, sentiment term detection, ambiguity resolution, polarity propagation, polarity weight calculation, polarity classification, and a manually constructed lexicon of over 1,200 polarity-tagged terms (positive, negative, and intensifiers). It features context-aware polarity propagation to manage modifiers like negations and intensifiers, marking it as the first sentiment analysis system for Afan Oromo. Evaluated on 375 domain-diverse reviews (house rentals, transportation services, and online platforms), it achieved F1-scores between 0.716 and 0.718, with 89% accuracy in binary sentiment classification. The framework supports applications in business intelligence, public policy, and low-resource NLP research, and is based on over 1,500 annotated sentences in the Qubee script.