Technological convergence integrates multiple technologies to provide novel technological solutions. Technology opportunity discovery (TOD) based on technological convergence serves as a powerful approach for industries to gain competitive advantages. Current research on technology classification code co-occurrence network mainly focuses on predicting the convergence of two technologies, with little attention to multi-technology convergence. Additionally, these studies overlook the dynamic semantics of technology nodes in different technological contexts. To address these limitations, this paper proposes a graph-based and graph neural network-based framework for TOD. First, we construct a co-occurrence network of Cooperative Patent Classification (CPC) of the patents. We then extract network node embeddings and, following the principle of maximizing node relevance, construct technology subgraph (TS) that incorporate multiple CPCs. Then, we develop a graph neural network, Tecformer, which evaluates the feasibility of TS, featuring Edge-Guided Attention (EGA) and Graph-Text Fusion (GTF) modules. Finally, we generate candidate TSs in the target domain based on the co-occurrence network, using Tecformer to extract their technological representations, and recommend technology opportunities to the target domain based on cosine similarity. Experimental results demonstrate that Tecformer achieves a 93.03% accuracy in assessing the feasibility of TSs, and approach’s effectiveness is validated through a case study.