In order to solve the problem that the traditional truth discovery algorithm cannot be applied to text data directly, a truth discovery algorithm(NN_Truth) for text data based on deep neural network was proposed. For the features of multifactorial property of text answers, the diversity of word usages, and the sparsity of the text data, the “source-answer” vector was used as the network input, and the truth vector was recognized as the network output. The relationship between answers from each source could be unsupervised learned according to general hypothesis of truth discovery, and finally obtained the truth. The experiment results show that the proposed algorithm is suitable for text data truth discovery, and it is better than the retrieval methods and traditional truth discovery algorithm.