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, which aims to identify the specific entities evaluated by users in review texts, represents a key task in opinion mining. A major challenge lies in the fact that users often employ both explicit and implicit expressions to convey opinions, while existing methods show limited effectiveness in recognizing implicit opinion targets. To address this issue, this paper proposes a dual-supervised contrastive learning approach for opinion target extraction (DCLWS), which enhances the model’s discriminative power by leveraging semantic relationships between explicit and implicit targets. The method integrates both intra-sentential contextual information and cross-sentential semantic cues. It constructs a contrastive learning framework that treats sentences containing explicit opinion targets as anchors, sentences corresponding to implicit targets of the same category as positive samples, and sentences of different categories as negative samples. This enables the model to learn more discriminative target word representations and sentence-level semantic embeddings. Experimental results on four benchmark datasets from the SemEval ABSA challenges (2014, 2015, and 2016) demonstrate that the proposed method achieves significant performance advantages over existing mainstream models, with the highest F1-score reaching 89.20, and improves accuracy in implicit target recognition to 98.32%. These findings confirm the effectiveness and robustness of the method in complex linguistic environments, offering a reliable pathway to enhance the performance of opinion mining systems in real-world applications.

Key words: opinion mining, target extraction, contrastive learning