TY - GEN
T1 - Aspect-Based Multilabel Classification and Emotional Polarity Analysis of E-Commerce Reviews for Marketing Strategies
AU - Firdaus, Aji Akbar
AU - Alzamzami, Moch Nafkhan
AU - Putri, Tesa Eranti
AU - Ravana, Sri Devi
AU - Dina, Nasa Zata
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2027.
PY - 2027
Y1 - 2027
N2 - Customer reviews on e-commerce platforms offer significant insights into product quality and consumer satisfaction. Nonetheless, a singular review frequently addresses various prod- cut attributes, making aspect identification a multilabel classification challenge. This study presents an aspect-based multilabel classification and emotion analysis framework to identify product aspects and evaluate customer emotions. Four aspects were examined: finishing, materials, construction, and color. Three embedding techniques (Word2Vec, SBERT, and e5- large-v2) were integrated with two multilabel problem transformations (Classifier Chains and Label Powerset) and three classifiers (Logistic Regression, Random Forest, and Stochastic Gradient Descent). Performance was evaluated with Accuracy, F1-score, and Hamming Loss. The findings indicated that transformer-based embeddings surpassed Word2Vec in performance. The e5-large-v2 utilizing Label Powerset and SGD attained the highest F1-score (0.8228), whilst e5-large-v2 employing Label Powerset and Random Forest reached the highest accuracy (0.8946). Emotional analysis indicated that trust, joy, and anticipation were the predominant emotions. These findings provide valuable insights for product improvement and marketing strategies.
AB - Customer reviews on e-commerce platforms offer significant insights into product quality and consumer satisfaction. Nonetheless, a singular review frequently addresses various prod- cut attributes, making aspect identification a multilabel classification challenge. This study presents an aspect-based multilabel classification and emotion analysis framework to identify product aspects and evaluate customer emotions. Four aspects were examined: finishing, materials, construction, and color. Three embedding techniques (Word2Vec, SBERT, and e5- large-v2) were integrated with two multilabel problem transformations (Classifier Chains and Label Powerset) and three classifiers (Logistic Regression, Random Forest, and Stochastic Gradient Descent). Performance was evaluated with Accuracy, F1-score, and Hamming Loss. The findings indicated that transformer-based embeddings surpassed Word2Vec in performance. The e5-large-v2 utilizing Label Powerset and SGD attained the highest F1-score (0.8228), whilst e5-large-v2 employing Label Powerset and Random Forest reached the highest accuracy (0.8946). Emotional analysis indicated that trust, joy, and anticipation were the predominant emotions. These findings provide valuable insights for product improvement and marketing strategies.
KW - consumer feedback
KW - customer behavior; natural language processing
KW - customer review analysis
KW - e-commerce analytics
KW - multilabel classification
KW - product quality improvement
KW - sustainable consumption
UR - https://www.scopus.com/pages/publications/105048481239
U2 - 10.1007/978-981-92-4305-1_20
DO - 10.1007/978-981-92-4305-1_20
M3 - Conference contribution
AN - SCOPUS:105048481239
SN - 9789819243044
T3 - Communications in Computer and Information Science
SP - 273
EP - 286
BT - Information Retrieval and Knowledge Management - 7th International Conference, CAMP 2026, Proceedings
A2 - Mohd Sharef, Nurfadhlina
A2 - Azman, Azreen
A2 - Herawan, Tutut
PB - Springer Science and Business Media Deutschland GmbH
T2 - 7th International Conference on Information Retrieval and Knowledge Management, CAMP 2026
Y2 - 12 August 2026 through 13 August 2026
ER -