The invention provides a new keyword extraction technology. According to the technology, vocabulary position weights and word class weights are determined according to a Chinese word segmentation preprocessing process, the relevancy between two vocabularies is calculated with reference to a core vocabulary with the highest text vocabulary contribution, a multi-subject network model is constructed, an objective function is constructed to extract connection words, a cross function is utilized to fuse the connection words into the multi-subject network model, a new model graph is obtained, and then an anteposition vocabulary, namely a text keyword is extracted. The technology is high in accuracy and has higher application value, the contributions of different vocabularies to text ideology can be precisely calculated, multi-subject performance is considered, different characteristics are distinguished, and a good theoretical basis is provided for subsequent text similarity analysis and text clustering.