Abstarct: Over the last few years, researchers in the field of natural language processing have been making great progress through the use of deep learning models, and Q & A systems as a subset of the science of natural language processing have not been exempted from this. In this thesis, we intend to provide a deep learning-baxsed model for automatic answer to the question. With this in mind, we can divide the question-answer system into three parts: categorizing questions and answers, retrieval information, and choosing an answer, first we classify the questions and answers using a neural-probabilistic modeling and transform into a vector, then We teach a deep neural network similarity to find out the similarity between questions and answers, and finally we select the best answer. Our proposed system, by proposing a method baxsed on Neural probabilistic modeling, transforms the exxpression into a vector, so that the vector of each statement become very similar to the word vector or its keywords so that the likelihood of the similarity of the question with its corresponding answers could raise and finally SIAMESE neural network, with changes, learns in a way that it can choose the right answer for the question. In this model, we managed to reduce the error by about 25% compared to the current one.