The event sequence in telecommunications networks covers important data such as communication records and system alarms, which is crucial for fault detection and network management. However, existing causality discovery algorithms have challenges on casual discovery in telecom networks. To this end, we design the Graph Transformer Hawkes Process (GTHP) model to reveal the causal relationships behind event sequences in telecom networks. Our proposed model is contributed to allow the use of transformer's self-attention mechanism with Hawkes process for casual discovery. To begin with, the proposed GTHP injects graph structure information into Transformer's self-attention by graph convolution networks to construct a graph transformer block, which effectively captures the long-range casual dependencies between event types. We then combine the graph transformer block with Hawkes process, in which the intensities of multiple event types evolve based on a continuous-time transformer mechanism. The extensive experiments are implemented for measuring the performance of GTHP in real-world telecom data. It demonstrates that GTHP has an obvious improvement in comparison with the baselines.