Abstract: In response to the problem of insufficient utilization of rich context information in sentiment classification tasks using graph convolutional networks, a knowledge-driven aspect based sentiment analysis model for multi-semantic syntactic interaction is proposed.Multiple semantic relationships between aspects and contexts are used to construct a multi-semantic relationship diagram, and external knowledge is used to construct an emotional dependency diagram.In addition, a more advanced context representation is obtained by optimizing the attention mechanism. Then, different representations were fused.The experimental results on five public datasets show that the proposed model, KSSGCN＿GloVe, has achieved an average improvement of 1.59% in accuracy and 3.17% in macro F1 compared to the baseline model ASGCN.