Research on Sentiment Analysis of Ceramic Review Texts Based on BERT-BiGRU-Attention

Sentiment analysis of product reviews can help both e-commerce shoppers and merchants make better decisions and promote products more effectively. However, existing sentiment analysis models that rely on word embeddings often overlook contextual relationships within sentences, and they face challenges such as metaphorical expressions and semantic variation. To address these issues, this paper proposes a feature fusion framework based on Bidirectional Encoder Representations from Transformers (BERT). The framework first uses BERT to pre-train the text and generate rich, dynamic word embeddings. A Bidirectional Gated Recurrent Unit (BiGRU) is then applied to capture contextual features, while an attention mechanism is introduced to highlight key features. Experiments on a ceramic dataset demonstrate that the BERT-BiGRU-Attention-based sentiment analysis model significantly outperforms existing methods in terms of accuracy, F1-score, and other evaluation metrics.