An empirical study of the Algerian dialect of Social network

In this paper, we present analysis on the use of Algerian dialect in Youtube. To do so, we harvested a corpus of 17M of words. This latter was exploited to extract a comparable Algerian corpus, named CALYOU by aligning pairs of sentences written in Latin and Arabic. This one was built by using a multilingual word embeddings approach. Several experiments have been conducted to fix the parameters of the Continuous Bag of Words approach that will be discussed in this article. The method we proposed achieved a performance of 41% in terms of Recall. In the following, we present several figures on the collected data that led to several unexpected results. In fact, 51% of the vocabulary words are written in Latin script and 82% of the total comments are subject to the phenomenon of code-switching.