基于Bert-BiLSTM的商品评论情感分析研究

Commodity review sentiment analysis based on Bert-BiLSTM model

商品评论为商家的选品和用户的购买提供了重要的决策帮助。为了获得商品评论的情感特征,并在评论中捕捉更多的情感信息,本研究提出一种Bert-BiLSTM的情感分类模型,运用斯坦福情感分析数据集、亚马逊商品评论数据集做一个情感分类模型,该模型利用Bert嵌入层对句子进行分割并将其转换为词向量,然后将其传递到BiLSTM模型中,以获取评论文本中的属性和情感词;训练后的模型使用混淆矩阵作为评价指标,相比其他深度学习模型在最终结果上表现出明显优势。将训练好的模型对结果进行分类和预测,从而分析结果的情感,这就为用户与商户在购买商品或选品时提供了建议方向和引导情绪。

Commodity reviews provide important decision-making assistance for merchants′ selection and users′ purchase. In order to obtain the emotional characteristics of commodity reviews and capture more emotional information in sentences, this study proposes an emotion classification model of commodity reviews. By using Standford sentiment analysis data set and Amazon products review dat set, the research uses the Bert embedding layer to segment sentences and convert them into word vectors, and then passes them into the BiLSTM model to obtain the attributes and emotional words in the review text. The trained model uses confusion matrix as the evaluation index, and has a better performance compared with other deep learning models. The trained model is used to classify and predict the results, so as to analyze the emotion of the results. The results could provide users and merchants with suggested directions and guided emtions during the process of purchasing goods.