Abstract: This paper proposes a method to identify potential dissatisfaction of service hotline based on RoBERTa model,which extracts negative emotional information from massive voice data of service hotline to solve potential dissatisfaction problems. Based on the traditional text sentiment analysis model,by adding the RoBERTa statement vector module,and introducing the attention mechanism,the text sentiment analysis technology can obtain better results in long text learning. The system has demonstrated excellent application effect in the customer service work of Tianjin Mobile and significantly improved the customer satisfaction.

Key words: emotional computing, text emotion recognition, attention mechanism, deep learning