@article{LIU2026, 
author = {Kan LIU and Naying ZHI and Xinyi GAO},
title = {Opinion Target Extraction Based on Dual-Supervised Contrastive Learning},
year = {2026},
journal = {Journal of South China University of Technology (Natural Science Edition)},
volume = {54},
number = {8},
pages = {14-25},
keywords = {opinion mining, target extraction, contrastive learning},
url = {https://www.sciopen.com/article/10.12141/j.issn.1000-565X.250295},
doi = {10.12141/j.issn.1000-565X.250295},
abstract = {With the rapid development of social media and e-commerce platforms, massive user reviews have become a vital source of feedback on products and services. Opinion target extraction is a key task in opinion mining, aiming to identify the specific objects that users evaluate in review texts. The major challenge of this task lies in the fact that users often express their opinions using both explicit and implicit manners, while existing methods achieve limited effectiveness in recognizing implicit opinion targets. To address this problem, this paper proposes a dualsupervised contrastive learning-based opinion target extraction model (DCLWS), which enhances the model’s discriminative ability by leveraging the semantic correlation between explicit and implicit targets. This model integrates intra-sentence contextual information and cross-sentence semantic information to construct a contrastive learning framework: using review sentences containing explicit opinion targets as anchor samples, review sentences corresponding to implicit targets of the same category as positive samples, and those corresponding to implicit targets of different categories as negative samples, thereby guiding the model to learn more discriminative target word representations and sentence-level semantic representations. Experimental results on four benchmark datasets from the SemEval ABSA 2014, 2015 and 2016 challenges demonstrate that the proposed model achieves significant performance advantages over existing state-of-the-art models, with an F1 score reaching up to 89.20%; in the implicit opinion target recognition subtask, the precision is improved to 98.32%. The results validate the effectiveness and robustness of the model in complex linguistic scenarios, providing a reliable approach for enhancing the performance of opinion mining systems in practical applications.}
}