The invention relates to a transfer learning and deep learning-based citizen mediation case classification system and method. The system comprises a data acquisition module, a feature extraction module, a feature transfer module and a network training module, and is simple in structure and wide in application range; The method comprises the steps of constructing a character vector matrix, performing auxiliary data vectorization processing, performing vectorization processing on the citizen mediation data, inputting the vectorized auxiliary data into a neural network, extracting auxiliary datafeatures, migrating the extracted auxiliary data features into the vectorized citizen mediation data, and training a classification model. According to the method, all texts can be effectively converted, low-frequency words cannot be ignored, dimension reduction is obvious, the training speed is high, and follow-up online iterative optimization is facilitated; And meanwhile, the difference betweenthe people's mediation field and the auxiliary field is solved, and the individual requirements of the specific field are met.