Abstract: With the increasing of information on heterogeneous network, how to help users get the needed information accurately has become a hot issue. Similarity search has been widely used in web search. The similarity search based on meta-path can express the semantics contained in heterogeneous network better. However the number of selected paths is large when use such kind of methods, which leads to large scale of computation. To address above problems, an important research area, similarity search that uses user-guide to predict the meta-paths, grows up. This kinds of approaches require the user to provide a sample of results as a guide while querying, which will be used to predict the meta-paths associated with user queries. By now, research for predict related meta-path which use the user-guide information mainly aimed at the same type of node in heterogeneous network. In this paper, the structural information of the heterogeneous network graph is used to establish the semantic prediction model, calculate the probability of the candidate paths that match user’s search, and then select the path with the highest probability. Experimental results show that the proposed algorithm can predict the relationship between the same type and different types of nodes in heterogeneous network. Meanwhile, it has good performance and effectiveness. This paper provides a concrete implemental method for how to obtain similarity results in multi-semantic environment.

Key words: similarity search, user-guided prediction, semantic prediction, heterogeneous network

摘要： 随着异构网络的信息量与日俱增，如何准确地帮助用户获取所需要的信息已成为一个热门问题。相似性搜索在网络搜索中得到了广泛的应用。基于元路径的相似性搜索能更好地表达异构网络所包含的语义。但是现有的大多数该类方法存在路径选择数多导致计算量大的问题。根据用户导向去预测元路径以缓解上述问题成为一个重要的研究方向。这类方法需要用户在搜索的同时提供结果样例作为导向，据此预测与用户搜索相关的元路径。目前，相关研究主要是针对异构网络中的同类型结点。利用图的结构信息建立语义预测算法，计算各候选路径与用户搜索的匹配概率，然后选择概率最大的路径。实验表明，提出的算法能够实现对同类型和不同类型结点间语义的预测，具有较好的性能和有效性。并为如何在多语义环境下获得相似性结果提供了具体的实现方法。

关键词: 相似性搜索, 用户导向预测, 语义预测, 异构网络