基于RoBerta-BiGRU-Attention的景区评论情感分析研究——以沈阳市为例

Research on Sentiment Analysis of Scenic Area Comments Based on RoBerta-BiGRU-Attention—Taking Shenyang as an Example

摘要: 游客在评论中所表达的意见和感受，能够直观地反映他们对旅游目的地的评价，同时语气鲜明、情感强烈。由此带来的巨大而动态的信息空间需要消费者和产品/服务提供者共同理解和导航。针对目前景区在线评论文本情感分类准确性不高的问题，提出一种基于RoBerta词向量和双向门控循环单元(BiGRU)的改进模型，使用能够表征文本丰富语义特征的Roberta模型进行词向量表示，结合能够长期保留文本上下文关联信息的BiGRU神经网络提高模型的分类效果，并在此基础上引入注意力(Attention)机制，突出文本中更能表达分类结果的情感词权重，提高情感分类的准确率。将上述模型分别在沈阳市3个景区评论数据上进行情感极性分类和预测，实验结果表明，该模型在各数据集上都获得了良好的性能。同时结合LDA主题模型分析，得到游客评论的期望和诉求，为沈阳市旅游业发展提供技术支撑以及未来发展意见。

Abstract: The opinions and feelings expressed by tourists in the comments can intuitively reflect their evaluation of the tourist destination, and at the same time, the tone is clear and the emotion is strong. The resulting huge and dynamic information space needs to be understood and navigated jointly by consumers and product/service providers. Aiming at the problem of low accuracy of emotion classification in online comment texts of scenic spots, an improved model based on RoBerta word vectors and bidirectional gated recurrent units (BiGRU) is proposed. The Roberta model, which can represent the rich semantic features of texts, is used to represent word vectors. The BiGRU neural network, which can retain text context information for a long time, improves the classification effect of the model. On this basis, an attention mechanism is introduced to highlight the weight of emotional words in the text that can better express the classification results, and improve the accuracy of emotion classification. The above models were used to classify and predict emotional polarity on the comment data of three scenic spots in Shenyang City. The experimental results show that the model has achieved good performance on each data set. At the same time, combined with the LDA topic model analysis, the expectations and appeals of tourists’ comments are obtained, and technical support and future development opinions are provided for the development of Shen- yang’s tourism industry.