With the large-scale use of social media applications, the value of social relationship data has drawn wide attention. The structure of social networks can be abstracted as a graph. To divide communities from the relational structure is equivalent to graph segmentation. When we use NJW multiple spectral clustering algorithm to process graph segmentation issue, we need to determine the clustering number manually. In order to solve this problem, this paper tried to introduce the concept of Eigengap to predict the number of clusters after spectrum analysis of input sample's Laplacian matrix. The effectiveness of the proposed algorithm was verified with on experimental data and got the desired results.%随着社交应用软件的广泛普及，社交关系数据中存在的价值得到人们的广泛关注，社交关系网络可以抽象成一种图结构，将用户从关系结构上进行划分等价于对图进行分割。针对NJW多路谱聚类算法在处理图分割时需要人为确定聚类数目的问题，引入本征间隙的方法，通过对输入样本数据的拉普拉斯矩阵进行谱分析，得出样本的聚类数目。实验证明，改进后的NJW算法，在实验数据集上可以自动获取聚类数目并具有较好的聚类效果。
展开▼
