Sentence classification task plays a crucial role in various NLP tasks. Recent studies have shown that contrastive learning can enhance the representational capability of Pre-trained Language Models (PLMs) and that different methods for constructing positive and negative samples can be applied to various downstream application scenarios. Therefore, in this study, we propose W2CL, a novel multi-task learning framework based on Word Classification and Contrastive Learning, aimed at integrating domain knowledge extracted by ChatGPT from raw corporas into PLMs and improving the performance of PLMs in domain-specific sentence classification tasks. Contrastive learning assists the model in gradually learning the semantic similarity and contextual relevance between words during the training process to enhance its ability to understand text. Word classification provides additional contextual understanding, thereby improving the model’s ability to differentiate between different classes within the specific domain. Experiments demonstrate that our multi-task approach significantly outperforms other methods, leading to substantial improvements in domain-specific sentence classification performance. This framework offers a robust solution for adapting general-purpose language models to specialized domains, ensuring better performance and generalization in various domain applications.