Intention Classification Based on Transfer Learning: A Case Study on Insurance Data

With the rapid development of Artificial Intelligence and Big Data technology, intelligent chatbot in insurance industry has become the major technical means to reduce labor costs and improve the quality of service. The core technology of this application is to understand and classify the users' intentions accurately. However, insurance as a product with complex knowledge system and long service cycle, users' intentions and the corresponding corpus is rather scattered. The initial corpus is especially scarce at the early stage of new business. So it is very important to classify the customers' intentions accurately based on the rare corpus. This paper offers an empirical case study on intention classification of insurance data by using transfer learning model BERT. The experimental comparative analysis result shows that method based on BERT model can better reduce the error rate than other existing model methods (TextCNN, HAN, ELMo).