This paper deals with the problem of the lack of resources in the opinion and emotion analysis related to north african dialects in general and the algerian dialect in particular. A collaborative platform "TWIFIL" for the annotation of multilingual public data is proposed. The result is a human generated corpus of extracted tweets. The purpose of this action is two-fold. The first, it addresses the shortage of relevant data for algerian dialect's opinion and emotion analysis. Second, it provides a more reliable (the appreciation of not just one person) annotated corpus. We also report on a number of evaluations, we have performed to test the generated corpus.