一种融合Text-Rank的旅游文本情感分析方法

A Sentiment Analysis Method for Tourism Text Integrated with Text-Rank

摘要: 对于在线平台和论坛中的旅游评论文本进行情感分析一方面有助于景区理解游客的需求，提高景区服务品质，另一方面为游客提供出游参考信息，以达到更好的旅游满意度，具有较高的应用价值。本文针对旅游评论文本普遍较长的特征，提出了融合Text-Rank的情感分类方法。在进行情感分类之前先使用Text-Rank方法对较长的文本进行自动摘要，使用摘要压缩后的内容作为输入进行情感分类。使用RNN、LSTM、Text-CNN、BERT等深度学习模型进行实验，结果显示融合后的方法在准确率等各项指标上均取得了一定程度的提升。该方法对于原始文本较长、包含多方面内容的输入信息具有较大价值，能够提高文本处理效率，提高分析精度。

Abstract: Sentiment analysis of tourism comment text on online platforms and forums helps scenic spots to understand tourist needs and improve their service quality. It also provides valuable reference information for tourist satisfaction, with high value for both parties. Dominated by long text in tourism comment, a sentiment classification model integrated with Text-Rank is proposed. Before feeding original text into the classifier, the Text-Rank method generates a summary of the long text, and the compressed summary is used as the input for sentiment classification. RNN, LSTM, Text-CNN and Bert are used in the experiment. Its result shows that the fused method yields significant performance over the original model. The proposed method improves text processing efficiency and accuracy, especially when the original text is over certain length or contains multiple aspects.