Abstarct: With the rapid growth of accessible stored data, the process of information retrieval and extraction has become more important than ever. Question answering systems are a form of information retrieval systems that can give users summarized and effective answers. Interactive question answering systems can reduce ambiguities in the question and increase accuracy of the answer by having a two-way interaction with the user. After receiving the question, one of the steps in question analysis is question classification. Multiple machine learning methods for feature extraction and question classification have been used in interactive question answering systems. On the other hand deep learning has improved machine learning in many areas. In this research by using deep learning and the features that it extracts we provide a new method for extracting more relevant features and classifying questions. In our method we use a pre-trained word2vec network for word representation. Additionally feature extraction and question classification is done by combining a convolutional neural network and an LSTM neural network. The designed network can use previous sentences of a dialog in case its accuracy is below a given threshold. Tests of our network show a 4.2 percent improvement compared to other machine learning methods for our collected dataset.