Tamanho da fonte: 

Propose an alternative method for to identify Multiwords Expressions extracted from documents through Recurrent Neural Network (RNN) model.Multiword Expressions could be used to represent the meaning of the document in the more semantic way, in many activities of the natural language processing. We found few researches that use RNN as a method to obtain MWE. It was the definition of the training corpus created within lines classified as bigrams or not, extracted from documents through traditional statistics methods; It was to train the model based on the corpus created; It was to validate the results obtained aim to generalize the process through RNN method.As results we obtained an accuracy around 80,54% in test train to identify bigram in the new documents.We propose the use of machine learning algorithm to generalized the extraction of bigrams in the documents in the specific domain.The idea is to do the information retrieval process to obtain searched documents through of a document rather than using keywords."