Satellite communications transmit data and information by satellites to achieve global communication connectivity. We propose a semantic Non-orthogonal multiple access (NOMA) satellite communication transceiver based on knowledge graph, which uses triples as semantic symbols to extract semantic information. The pre-trained model is fine-tuned to reconstruct the text. In addition, a successive interference cancellation method MeanSIC based on subtractive mean interference is proposed to reduce inter user interference and achieve user signal separation. The experiment result shows that the proposed system can not only reduce the communication resources required to transmit the same data, but also effectively reduce interference between users. Our channel reconstruction scheme ensures high reliability and facilitates signal recovery.