The Aspect-Based Sentiment Analysis (ABSA) task is designed to judge the sentiment polarity of a particular aspect in a review. Recent studies have proved that GCN can capture syntactic and semantic features from dependency graphs generated by dependency trees and semantic graphs generated by Multi-headed self-attention (MHSA). However, these approaches do not highlight the sentiment information associated with aspect in the syntactic and semantic graphs. We propose the Aspect-Guided Multi-Graph Convolutional Networks (AGGCN) for Aspect-Based Sentiment Classification. Specifically, we reconstruct two kinds of graphs, changing the weight of the dependency graph by distance from aspect and improving the semantic graph by Aspect-guided MHSA. For interactive learning of syntax and semantics, we dynamically fuse syntactic and semantic diagrams to generate syntactic-semantic graphs to learn emotional features jointly. In addition, Multi-dropout is added to solve the overfitting of AGGCN in training. The experimental results on extensive datasets show that our model AGGCN achieves particularly advanced results and validates the effectiveness of the model.