THE 3 rd INTERNATIONAL SCIENTIFIC CONFERENCES OF STUDENTS AND YOUNG RESEARCHERS dedicated to the 99
th
anniversary of the National Leader of Azerbaijan Heydar Aliyev
120
𝐹1 − 𝑆𝑐𝑜𝑟𝑒 = 2 ∙
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∙ 𝑅𝑒𝑐𝑎𝑙𝑙
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑅𝑒𝑐𝑎𝑙𝑙
The important point is to predict whether the product is recommended or
not and the star rating from 1 to 5. As it is multi-class classification for the
star rating part, on some level, it is satisfactory because of the data
conditions. The below-mentioned table presents the comparison of the most
widely used two techniques in Natural Language Processing. As it could be
seen, the F1-score is quite high in comparison with the previous methods.
Comp ariso n Recommendation Prediction with RNN and XLNet Recommended or not
Precision
Recall
F1-score
0
0.72
0.88 0.77
0.85 0.75
0.87 1
0.95
0.97 0.94
0.98 0.94
0.97 Rating Prediction with RNN and XLNet Star Rating
Precision
Recall
F1-score
1
0.30
0.79 0.37
0.69 0.33
0.74 2
0.47
0.75 0.30
0.77 0.37
0.76 3
0.40
0.68 0.65
0.61 0.50
0.64 4
0.49
0.82 0.32
0.70 0.39
0.76 5
0.78
0.84 0.85
0.89 0.81
0.86 The sentiment analysis become very essential part of the development
process of e-commerce. By using our improved model for analyzing of the
reviews, the service quality can be enhanced via reviewing and responding
the feedbacks in time. Of course, there are still some obstacles for the Deep
Learning techniques to fully understand the tone from the series of words.
However, continuous increase of the dataset and enrichment of lexicon can
result in a higher model performance.
References :
[1]. Muhammad Marong, “Sentiment Analysis in E-Commerce: A Review on The Techniques
and Algorithms”, Journal of Applied Technology and Innovation, Vol. 4, 2020.
[2]. Sneha Jadav, “Sentiment Analysis: A Review”, International Journal of Advance
Engineering and Research Development, Vol.4, 2017