Distribution transformers are important devices in power distribution networks. The mining of defect texts related to distribution transformers can provide valuable information that impacts the safe and stable operation of these transformers. However, due to the complexity and large volume of textual data in the power grid domain, the application of text mining is still in its early stages in China’s power system. This paper focuses on the research of a semantic recognition-based clustering algorithm for defect texts of distribution transformers. It aims to uncover potential patterns among these defect texts and contribute to the operation and maintenance of distribution networks.First, the semantic recognition of defect texts in distribution transformers is implemented to generate semantic vectors. Then, a clustering algorithm framework based on semantic vectors for defect texts in distribution transformers is proposed. Since the semantic vectors obtained from text representation models contain the main feature information of the texts and can be directly processed by computers, they can be inputted into different clustering algorithms to achieve clustering of defect texts in distribution transformers. Next, clustering experiments on defect texts in distribution transformers are conducted. Four different types of clustering algorithms are used to cluster the text feature vectors obtained from different text representation models. The experimental results are then compared and analyzed based on three clustering evaluation metrics.