The success of Natural Language Processing tasks generally depends on data representation. Representational learning incorporated in predictive models is the trend of deep learning models. Learning sentence representation with full semantics of document is a challenge in natural language processing problems. Because if the semantic representation vector of the sentence is good, it will increase the performance of finding similar question problems. In this paper, we propose to implement a series of LSTM models with different ways of extracting sentence representations and apply them to question retrieval for the purpose of exploiting hidden semantics of sentences. These methods give sentence representation from hidden layers of the LSTM model. The techniques consist using the last hidden layer of LSTM, Max pooling and Mean pooling. The results show that the technique using a combination of both Maxpooling and Meanpooling gives the highest results on the 2017 semeval dataset for the problem of finding similarity questions.