Aspect sentiment triplet extraction (ASTE) is an advanced and challenging task in aspect-level sentiment analysis, which aims to explain target opinions by extracting triplets consisting of aspects, opinions, and sentiments. However, existing models struggle with complex relationships, such as many-to-one and one-to-many relationships corresponding to aspects and opinions. To solve this problem, we propose a complex relationship modeling based method, called CRMM. First, the two-channel graph convolutional network is presented to add semantic and syntactic dependency information for capturing the complex relationships between aspects and opinions. Second, word pair tagging is proposed to identify the position of aspects and opinions. Then, the sentiment of aspects and opinions are designed to construct the aspect sentiment triplets. Experiments on four benchmark datasets show that our proposed model has a noticeable improvement in dealing with various relationships comparing the other models.