The invention is applicable to the technical field of information mining, and provides a hot event extraction method based on network media. The hot event extraction method comprises the steps of: firstly, partitioning a time interval for releasing news and comment texts into a plurality of time slices at equal intervals; secondly, extracting keywords and weights of the news texts in each time slice, carrying out aggregation classification on the news texts by an optimal incremental clustering method to obtain each topic and a spacial vector model thereof in the time slice, calculating topic heat according to each model, and according to the heat, retaining active topics; thirdly, carrying out aggregation classification on the topics of all the time slices by the optimal incremental clustering method to obtain different events, and calculating event heat; and finally, determining whether the events are hot, and by the heat, carrying out ranking on the hot events. According to the invention, a common incremental clustering algorithm is improved, multiple stages of optimal incremental clustering models are established, and a calculating speed is improved; and the heat of the events is calculated according to attention conditions of the media and public opinion conditions of net citizens, whether the events are hot is determined, the hot events are ranked, the method is more objective, and an event ranking error is reduced.