基于BERT和LDA模型的酒店评论文本挖掘

HOTEL REVIEW TEXT MINING BASED ON BERT AND LDA MODEL

通过主题模型对酒店评论文本进行文本挖掘,有利于引导酒店管理者和客户对评论信息做出合适的鉴别和判断。提出一种基于预训练的BERT语言模型与LDA主题聚类相结合的方法;利用中文维基百科语料库训练BERT模型并从中获取文本向量,基于深度学习算法对评论文本进行情感分类;通过LDA模型对分类后的文本进行主题聚类,分别获取不同情感极性文本的特征主题词,进而挖掘出酒店客户最为关注的问题,并对酒店管理者提出具有参考价值的建议。实验结果表明,通过BERT模型获取的文本向量在情感分类任务中表现较好,且BERT-LDA文本挖掘方法能使酒店评论文本的主题更具表达性。

Text mining of hotel review texts through topic models helps guide hotel managers and customers to make appropriate identification and judgments on review information.This paper proposes a method based on the combination of pre-trained BERT language model and LDA topic clustering.We used Chinese Wikipedia corpus to train the BERT model,obtained the text vectors from the BERT structure,and made a sentiment classification of review text by using deep learning algorithms.We established a LDA model to cluster the classified texts to obtain the feature topic words of the texts with different sentiment polarity,digged out the most concerned issues of hotel customers,and put forward suggestions with reference value for hotel managers.The experimental results show that the vectors obtained by the BERT model performs better in sentiment classification task and the proposed BERT-LDA text mining method makes the topic of hotel review texts more expressive.