A Rule-Based Holistic Approach for Turkish Aspect-Based Sentiment Analysis
Kivanc Bayraktar, Uraz Yavanoğlu, Alper Ozbilen · 2019
In this study, a holistic method which uses statistical, linguistic and rule-based approaches for Turkish aspect-based sentiment analysis is proposed. The proposed method has been tested on the Turkish restaurant dataset created within the scope of SemEval Aspect Based Sentiment Analysis (ABSA) 2016. Firstly, candidate aspect terms were acquired employing LDA, C-value and WSBFE. Afterwards, aspect terms were found by rule-based approach and aspect-sentiment pairs were determined. In aspect term extraction 56,28% f-score was obtained while in aspect-sentiment matching phase 52,05% accuracy was achieved.