Affiliations: [a] Key Laboratory of China’s Ethnic Languages and Information Technology of Ministry of Education, Northwest Minzu University, Lanzhou, Gansu, China | [b] Key Laboratory of China’s Ethnic Languages and Intelligent Processing of Gansu Province, Northwest Minzu University, Lanzhou, Gansu, China
Correspondence:
[*]
Corresponding author: Fucheng Wan, Key Laboratory of China’s Ethnic Languages and Intelligent Processing of Gansu Province, Northwest Minzu University, Lanzhou, Gansu 730030, China. E-mail: [email protected].
Abstract: Word2vec is often used in text sentiment analysis to generate word vector, which maps the same word into the same vector. Although Word2vec plays a very good effect in the initial model training task, it still cannot solve the problems of polysemy and new use of old words, which leads to inaccurate extracted features and affects the final classification results. In this paper, BERT model was used to vectorize the review text of tourist attractions, and fusion attention mechanism and long and short-term memory model were used to extract the emotional features of the text for classification at the feature extraction layer. The emotional accuracy of the model proposed in this paper reached 95.79% in the review text of tourist attractions.
Keywords: Sentiment analysis, deep learning, BERT model, attention mechanism
