Scholarlink

<p>Integrating customer feedback into product refinement is critical for businesses aiming to improve product offerings, enhance customer experiences, and boost revenue. While user-generated reviews on e-commerce platforms provide timely and unbiased feedback, the unstructured nature and high volume of such textual data pose significant challenges for extracting actionable insights. This study investigates the application of natural language processing (NLP) techniques, specifically topic modeling and text classification, to address this issue. In the first phase of the study, we employed Latent Dirichlet Allocation (LDA) and BERTopic to identify latent topics within customer reviews, providing a thematic overview of customer discussions. Due to the bad performance of these models, manual labeling was introduced in the second phase based on the topics identified in the initial step. The final classification model was developed using PhoBERT embeddings combined with Logistic Regression. The experiments were conducted on a dataset of 17,002 Vietnamese customer reviews from Shopee, one of Vietnam’s largest e-commerce platforms. The model successfully categorized reviews into five primary topics: Product Quality, Customer Service, Price, Shipping, and Packaging, achieving an F1-score of 0.96 and a Hamming Loss of 0.022. This result helps e-commerce managers identify customer issues, allowing for effective improvements in the buying experience.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Topic classification of vietnamese product reviews in e-commerce using PhoBERT

  • 期刊论文

  • 发表日期：

    2025年4月5日

查看全文
  • Tuan Duy Nguyen,
  • Duc Minh Nguyen,
  • Huu Manh Nguyen,
  • Thi Quynh Giang Nguyen

摘要

Integrating customer feedback into product refinement is critical for businesses aiming to improve product offerings, enhance customer experiences, and boost revenue. While user-generated reviews on e-commerce platforms provide timely and unbiased feedback, the unstructured nature and high volume of such textual data pose significant challenges for extracting actionable insights. This study investigates the application of natural language processing (NLP) techniques, specifically topic modeling and text classification, to address this issue. In the first phase of the study, we employed Latent Dirichlet Allocation (LDA) and BERTopic to identify latent topics within customer reviews, providing a thematic overview of customer discussions. Due to the bad performance of these models, manual labeling was introduced in the second phase based on the topics identified in the initial step. The final classification model was developed using PhoBERT embeddings combined with Logistic Regression. The experiments were conducted on a dataset of 17,002 Vietnamese customer reviews from Shopee, one of Vietnam’s largest e-commerce platforms. The model successfully categorized reviews into five primary topics: Product Quality, Customer Service, Price, Shipping, and Packaging, achieving an F1-score of 0.96 and a Hamming Loss of 0.022. This result helps e-commerce managers identify customer issues, allowing for effective improvements in the buying experience.