The invention provides a random walk model-based zero-reference article recommendation method and system. The method comprises the following steps of 1, constructing an academic network model, and obtaining feature values corresponding to a first author, a conference or periodical, a mechanism and a published time of each paper through a random walk method; 2, establishing a sorting model, and selecting paper data processed in the step 1 to construct a training set; 3, sorting the training set through a weak classifier; 4, judging whether the sorting result of the weak classifier is matched with a real sorting result of the training set or not, so as to obtain an optimal sorting model; and 5, recommending zero-reference literatures required by the users through the sorting model. According to the method and system provided by the invention, a brand new paper sorting through is used, so that the newly published papers can be recommended more effectively and then the users can obtain the most relevant new papers.