BERT-Based Sentiment Analysis of Turkish e-Commerce Reviews: Star Ratings Versus Text
View Full Paper
Ask AI
Bookmark
Share
Key Points
Star ratings often misrepresent customer sentiment, as many high ratings correspond to negative reviews based on textual analysis.
Model 1's reliance on numerical ratings led to a tendency for overclassification of negative sentiments, while Model 2 utilized direct sentiment labels more effectively.
Using BERT for sentiment analysis on Turkish e-commerce reviews highlights the advantages of deep learning in handling complex language structures over traditional methods.
The significant difference in model predictions from chi-square tests underscores the importance of the approach used in sentiment classification.
Discussion
No takes yet. Share an insight, caveat, or question.
Implication
This analysis compares sentiment classification methods in Turkish e-commerce reviews, indicating that text analysis is more reliable than star ratings.