The invention discloses a party building text representation method based on a neural network and subject strengthening. A relevant text of party building work is one of the most important forms of party building work data, and is a kind of domain text with distinct themes. Text vector representation is an important basis for text analysis. According to the relevant text data of the party buildingwork, firstly, a word set and a subject term set of the party building text are extracted, and the party building text is expressed by using the word vector technology; secondly, based on the two-dimensional convolution neural network, the semantic features of the party building text are extracted; thirdly, the subject term set of the party building text and the semantic features of the party building text are fused for strengthening the theme features of the party building text, so that an obtained text vector can more reflect the field features of the party building text. The method is usedfor quantitative calculation and intelligent analysis of the party building text, and especially provides a more effective support for classification, retrieval, translation and other kinds of work of the party building text.