Recently, semantic data streams were proposed as a solution to cope with the heterogeneity of the original streams. However, nowadays, huge volumes of data are produced on the web, at very high velocity. This may provoke bottleneck effect and decrease efficiency of RDF stream processing engines. One approach to address this issue is to compress the data in the stream to decrease the delays and costs of the RDF exchange on the network. In this paper, we propose Patorc: a PATern ORiented Compression approach, a lossless method for compressing semantic data stream. Our approach takes advantage of the RDF data streams key features, which are the regularity of their graph structure and the redundancy of part of data. Experiments carried on publicly available datasets have demonstrated the effectiveness of our approach