Research on Sentimental Evaluation of E-commerce Product Reviews Based on the BiLSTM-Attention Mechanism

Abstract

With the booming development of e-commerce, massive amounts of product review text contain valuable information about user emotions and experiences, providing valuable insights for both merchant decision-making and consumer shopping. However, traditional sentiment analysis methods rely on manual feature extraction and have significant limitations when dealing with complex semantics and contextual dependencies. This study aims to construct a highly accurate and robust sentiment classification model to address the inaccurate identification of sentiment in e-commerce review text. First, multi-platform and multi-category e-commerce review data is collected and preprocessed through cleaning, word segmentation, and word embedding to construct a standardized dataset. Subsequently, a deep learning architecture integrating BiLSTM and an attention mechanism is designed. BiLSTM captures bidirectional temporal dependencies in text, while the attention mechanism dynamically assigns feature weights to enhance the extraction of key sentiment information. Finally, a multi-dimensional evaluation is conducted using accuracy, precision, recall, and F1 score. Experimental results show that the proposed BiLSTM-attention model achieves an accuracy of 91.2%, an 8.9% and 3.6% improvement over the traditional LSTM and single BiLSTM models, respectively. The further optimized CNN-BiLSTM-attention model achieves an accuracy of 93.4%, achieving the best overall performance. This research provides an efficient technical path for sentiment analysis of e-commerce reviews, enriches the theoretical application of natural language processing in the commercial field, and provides data support for e-commerce platforms to optimize services, merchants to formulate marketing strategies, and consumers to make rational purchases.