Multiple-scale community of complex networks has attracted much attention. For the problem, previous methods can not investigate multiple-scale property of community. To address this, we propose a novel algorithm (h_LPA) to detect multiple-scale structure of community. The algorithm is a heuristic label propagation algorithm associated with spectral analysis of complex networks. Label updating strategy of h_LPA is combined with heuristic function from the perspective of networks dynamics. The heuristic function further improves the dynamic efficiency of h_LPA. Extensive tests on artificial networks and real world networks give excellent results.

Dimensionality reduction is useful for improving the performance of Bayesian networks. In this paper we suggest an effective method of modeling categorical and numerical variables of the mixed data with different Bayesian classifiers. Such an approach reduces output sensitivity to input changes by applying feature extraction and selection, and empirical studies on UCI benchmarking data show that our approach has clear advantages with respect to the classification accuracy.

According to the coordinates of the particled graph, the paper presents a secondary pretreatment method which is invariant to translation, scaling and rotation. Having the advantages of efficient and accurate, the new method is significant to image processing. As the core of the pretreatment, orientation pretreatment is described in detail. In combination with the semi-structured data storage model, the new pretreatment method can help to achieve image content retrieval and data mining of a deeper level.