Nowadays, high volumes of data are generated and published at a very high velocity, producing heterogeneous data streams. This has led researchers to propose new systems named RDF Stream Processors (RSP), to deal with this new kind of streams. Unfortunately, these systems are fallible when their maximum supported speed is reached especially in a limited system resources environment. To overcome these problems, recent efforts have been made in the field. Some of them decrease the volume of RDF data streams using compression or load-shedding techniques, mostly according to a probabilistic approach. In this paper we propose POL: a Pattern Oriented approach to Load-shed data from RDF streams based on a deterministic approach. As a pre-processing task through a unique pass, the approach extracts the exact needed semantic data from the stream. The conducted experiments on public available datasets have demonstrated the effectiveness of our approach.