Resumen
The issue of aquatic product quality and safety has gradually become a focal point of societal concern. Analyzing textual comments from people about aquatic products aids in promptly understanding the current sentiment landscape regarding the quality and safety of aquatic products. To address the challenge of the polysemy of modern network buzzwords in word vector representation, we construct a custom sentiment lexicon and employ the Roberta-wwm-ext model to extract semantic feature representations from comment texts. Subsequently, the obtained semantic features of words are put into a bidirectional LSTM model for sentiment classification. This paper validates the effectiveness of the proposed model in the sentiment analysis of aquatic product quality and safety texts by constructing two datasets, one for salmon and one for shrimp, sourced from comments on JD.com. Multiple comparative experiments were conducted to assess the performance of the model on these datasets. The experimental results demonstrate significant achievements using the proposed model, achieving a classification accuracy of 95.49%. This represents a notable improvement of 6.42 percentage points compared to using Word2Vec and a 2.06 percentage point improvement compared to using BERT as the word embedding model. Furthermore, it outperforms LSTM by 2.22 percentage points and textCNN by 2.86 percentage points in terms of semantic extraction models. The outstanding effectiveness of the proposed method is strongly validated by these results. It provides more accurate technical support for calculating the concentration of negative emotions using a risk assessment system in public opinion related to quality and safety.