Enhanced Intention Recognition Model Based on Text Classification

Internet technology has been deeply integrated into human daily life, with a wide array of life-oriented apps significantly enhancing the convenience of daily activities. The customer service system, acting as the central communication hub between enterprises and customers, is under substantial business pressure. Intelligent customer service systems can efficiently assist human customer service representatives in addressing relatively simple and common issues based on user descriptions. As a vital branch of artificial intelligence, intention recognition can give all kinds of intelligent systems the ability to understand natural language. To address the limitation that traditional pre-trained model-based intent classification fails to introduce key text information in target intention classification, this paper proposes a novel model incorporating multi-granularity matching and inconsistent regularization. This model utilizes encoders to model key text information within the target intent and introduces multi-granularity matching to ensure enhanced model performance. Various experiments were conducted on a public dataset called KUAKE_QIC, and the results show that the proposed model has a good performance in terms of accuracy, in addition to the effectiveness of tail category recognition and the performance of small sample scenarios. Specifically, compared with the existing models, the accuracy rate of our model increased by 10.1%–0.96%, the recall rate increased by 4.85%–0.81%, the F1 score increased by 7.46%–0.88%, and the recall-3 metric improved by 7.14%–1.63%. The model proposed in this paper represents a significant advancement over existing models without pre-trained on dialogue corpora.