Aspect-level sentiment classification (ALSC) aims to extract fine-grained sentiment polarities from unstructured textual data. While some studies can combine syntactic structure and semantic information in sentences to improve the accuracy of sentiment classification, they ignore the limitations of using external sentiment knowledge information to enhance sentiment representation. Therefore, this paper proposes a Sentiment Knowledge-Enhanced Text Mining Model (SKETMM), which encodes external sentiment knowledge into textual statement vectors and syntactic dependency trees with the aspect as the root node, to obtain richer sentiment representations, consisting of two specific modules. The statement knowledge enhancement module utilizes the sentiment vectors from the widely-used sentiment lexicon SenticNet to enhance textual representations and redistributes attention weights through a multi-head attention mechanism. The syntactic dependency knowledge enhancement module encodes the pruned dependency tree into a matrix, fuses the sentiment knowledge, and processes it via a graph convolutional network to capture structured sentiment patterns in text. Experiments on public review datasets prove that SKETMM improves the accuracy and Macro-F1 of sentiment classification.