A social message monitoring method is implemented by receiving, from a social network server, a social message, and obtaining a theme probability vector of the social message, comparing the theme probability vector of the social message with a theme probability vector of each representative message to obtain a theme similarity, and acquiring a similarity between the social message and each representative message according to the theme similarity, and saving the social message in a message class that contains a representative message most similar to the social message, and outputting the message class to a social network client when a quantity of social messages in the message class reaches a first threshold or themes of social messages in the message class are consistent.