Abstract: The huge size and complex structure of the social network make it impossible to explore the community structure of the whole network. Therefore, the community search, which focuses on the local community structure, has important research significance. Previous community-search algorithms usually combine irrelevant structures into the target community. This paper uses the Skip-gram model to learn the latent representations of networks, and uses the new goodness metric of community which combines the similarity and degree of the nodes together to find the target community. The experiments on real networks demonstrate the accuracy and consistency of the new algorithm.

Key words: community search, local community detection, Skip-gram, similarity of nodes

摘要： 社会网络的巨大规模和复杂结构使得探索整个网络的社区结构的代价变得高昂。因此，着眼于网络局部结构特征的社区查询有着重要的应用意义。常见的社区查询算法易将与查询无关的子结构合并到目标社区中。利用Skip-gram模型将序列化后的社会网络映射到连续的向量空间以求解节点之间的相似度，并结合节点的度这个属性特征修正了原有的社区尺度，以此作为标准进行节点聚类，从而得到查询节点所属的社区结构。经过在真实数据集上的实验，改进的社区查询算法的准确性和查询一致性较已有算法有了较大提高。

关键词: 社区查询, 局部社区发现, Skip-gram模型, 节点相似度