ASPECT ORIENTED SENTIMENT ANALYSIS OF USER TEXT MESSAGES

Abstract

Aspect-oriented sentiment analysis plays a crucial role in understanding users' opinions and sentiments towards specific product or service features. This study investigates intelligent algorithms for aspect-oriented tone analysis of user text messages, focusing on smartphone reviews as a case study. The study includes data collection and preprocessing, studying the methods and performance of models for aspect extraction and tone analysis. The performance of these models was compared on test datasets containing customer reviews. For aspect extraction, we combine cross-lingual syntactic analysis with topic models to improve accuracy over the combination of Russian-language syntactic analysis and topic models. In particular, the BERT transformer-based model, BER Topic, shows high performance in aspect detection due to its ability to understand the context in sentences. In the sentiment analysis task, the RuBERT-tiny model based on the BERT transformer outperforms the others, showing higher accuracy in sentiment classification in smartphone reviews. This study provides valuable insights into aspect-oriented tonality analysis, emphasizing the importance of selecting appropriate methods and approaches. The results provide researchers and practitioners with recommendations for effective aspect-oriented tonality analysis.