Abstract
To analyze and explore large textual corpus, we are generally limited by the available main memory. This may lead to a proliferation of processor load due to greedy computing. We propose to deal with this problem to compute document similarity. We propose upstream and downstream parallel computing related to data and processing according to three abstraction levels: document, sentences and words. In order to consider these levels in hybrid computing, we opt for the dveloppement of a deployment platform able to deal with large scale applications. This choice was based essentially on the nature and charge of the treatment. Many experiments have been conducted according to different computing scenarios in order to ensure an accuracy of the obtained results through local networks.