Sentiment analysis has become an important topic in the field of natural language processing. Short text possesses multiple characteristics such as semantic sparsity and polysemy, which bring numerous challenges to short text sentiment analysis. Currently, the sentiment analysis of short text is insufficient and faces many problems such as feature semantics loss that may occur in the training process and the features that are used to classify with excessive dimension. Aiming at the existing problems, we proposed a short text sentiment classification model with a sentence vector enhancement mechanism (ERNIE-BiGRU-SVE-DPCNN, EBSD). This model uses Bi-GRU to extract the word order features of short text and uses DPCNN to reduce the dimension of features that are used for classification. Based on the group-wise enhancement mechanism, we proposed a Sentence Vector Enhancement Mechanism (SVE), which can enhance the features according to the sentence vector generated by ERNIE. The purpose of the SVE is to reduce the loss of semantic meanings. The result of experiments on the short text dataset online_shopping_10_cats has shown that the EBSD has higher accuracy than other baselines.