The goal of event detection is to locate trigger words and determine their corresponding event types. Given the proven effectiveness of syntactic dependency trees, graph convolution-based event detection methods have gained increasing popularity. However, these methods face two key challenges. First, the graph structure disrupts the inherent sequential structure of the sentence. Second, the semantic information contained in the trigger word’s label is often overlooked. To address these issues, we propose a Trigger Word Similarity Enhanced Sequence injected Heterogeneous Graph Convolutional Network (TSE-SIHGCN). TSE-SIHGCN reconstructs the syntactic graph while preserving the sentence’s sequential structure within the graph convolutional network. Furthermore, to strengthen the model’s capacity for comprehending the semantics associated with trigger word labels, TSE-SIHGCN introduces an encoder module based on trigger word similarity enhancement, enriching the semantic representation of word embeddings. We conduct extensive experiments and compare our approach with existing methods. The results demonstrate the effectiveness of the proposed model, highlighting its ability to improve event detection performance.