As Internet develops rapidly huge amounts of texts need to be processed in a short time. This entails the necessity of fast, scalable methods for text processing. In this paper a method for pairwise text similarity on massive data-sets, using the Cosine Similarity metric and the tf-idf (Term Frequency-Inverse Document Frequency) normalization method is proposed. The research approach is mainly focused on the MapReduce paradigm, a model for processing large data-sets in parallel manner, with a distributed algorithm on computer clusters. Through MapReduce model application on each step of the proposed method, text processing speed and scalability is enhanced in reference to other traditional methods. The CSMR (Cosine Similarity with MapReduce) method’s implementation is currently at the implementation stage. Precise and analytical conclusions concerning the efficiency of the proposed method are to be reached upon completion and review of the overall project phases.