Abstract: The supervised approaches suffer from low precision when detecting user profile attacks. Aiming at this problem, an ensemble detection approach is proposed by introducing bionic pattern recognition theory and ensemble learning technology. Firstly, through calculating the distance between a covered line segment and its nearest genuine profile, an adaptive calculational algorithm is proposed to adaptively assign a proper radius of hypersphere for each neuron. Secondly, the assigned radius is used to redesign an existing constructive neural network to make it more reasonable to cover attack profiles so as to improve its classification capability. Finally, an ensemble framework is proposed to detect user profile attacks. To create diverse base training sets, a base training set generation algorithm is proposed by combining various attack types. These base training sets are used to train the redesigned neural network in order to generate base classifiers. Based on ITS algorithm, a selective ensemble detection algorithm is proposed to select parts of base classifiers and the majority voting strategy is used to integrate the outputs of these base classifiers. The experimental results on two different scale of real datasets, MovieLens and Netflix, show that the proposed approach can effectively improve the precision under the condition of holding a high recall.