Contrastive Learning for Aspect-Based Sentiment Analysis (ABSA) of E-Commerce Logistics and Delivery Complaints on Social Media

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

The exponential growth of e-commerce has been accompanied by a corresponding surge in customer complaints regarding logistics and delivery failures, frequently expressed on social media platforms. Aspect-Based Sentiment Analysis (ABSA) offers a granular approach to understanding these complaints by identifying sentiment toward specific service aspects such as delivery timeliness, package condition, and customer support responsiveness. However, existing ABSA models struggle with the linguistic complexity of social media text, particularly in distinguishing overlapping sentiment cues and disambiguating aspect-specific polarities. This study proposes a novel contrastive learning-enhanced framework for ABSA that systematically addresses these limitations. The proposed framework integrates a dual contrastive learning mechanism with a hypergraph convolutional network to model both intra-sentence and inter-sentence relational structures. Using a curated dataset of 6,500 e-commerce logistics-related complaint conversations from Twitter, the model achieves an accuracy of 89.4% on aspect-level sentiment classification, significantly outperforming baseline machine learning models, which achieved a maximum accuracy of 82.1% using Bagging with RepTree. The contrastive learning approach demonstrates a 50% improvement in sensitivity to implicit sentiment intensity compared to conventional embedding methods. These findings establish contrastive learning as a viable paradigm for enhancing ABSA in the logistics domain, offering practitioners a replicable framework for real-time customer sentiment monitoring and providing researchers with a foundation for future work on multimodal and cross-lingual ABSA applications.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.