Chinese text summarization technology aims to extract the most important and essential information from lengthy texts and present it concisely, enabling readers to efficiently access key information and enhance reading efficiency. In generative Chinese text summarization, key information loss and semantic deviations in the generated content frequently occur. To address this issue, this paper proposes integrating the T5 model with contrastive learning. By expanding the summary section of the dataset to construct a set of positive samples, the model-generated summaries can be better aligned semantically with the reference summaries. Clauses from the source text are used as a set of negative samples to mitigate the loss of key information during the summarization process. We validate our approach through experiments on the CSL dataset and the NLPCC 2017 single-document summarization dataset. Compared to the existing T5 model, our proposed ConT5Sum framework achieves significant improvements across multiple evaluation metrics and demonstrates superior performance.