Abstract: Man-machine dialogue is a realistic application scenario which derived from the field of natural language processing. Based on the large amount of short text knowledge obtained from reality, the man-made short dialogue intelligent response chatterbot is constructed. Based on the traditional information retrieval typed chatterbot, the Recurrent Neural Network (RNN) is introduced to characterize the semantic vector of short text in the interactive knowledge base and reconstructs the expression semantic space. Experiments show that the effect of the method of using the coding vector is improved compared with the traditional method of using TF-IDF vector.

Key words: chatterbot, RNN sentence auto-encoder, TF-IDF