In order to solve the problem of scarcity of labeled collusion data in the spot market, a collusion recognition model based on semi-supervised support vector machine is designed. Firstly, a set of collusion identification index system is designed and its detailed calculation method is introduced. Secondly, the Topsis model modified by Delphi method is used to make a preliminary judgment of the unit, and the collusion recognition training set is constructed. The possibilities of collusion of the unit are divided into “high”, “medium” and “low”. Then, a collusion recognition model based on semi-supervised support vector machine is proposed. The collusion recognition model is trained by using labeled collusion recognition samples, and the unlabeled samples are labeled. The group of labels with the largest classification interval is selected as the final label. The original label samples and the new labeled samples are mixed to retrain the collusion recognition model. Finally, the effectiveness of semi-supervised support vector machine algorithm is verified by an example of spot market data in a certain region.