OSim: An OLAP-Based Similarity Search Service Solver for Dynamic Information Networks

Abstract

Similarity search service has always been one of the most popular topics in data mining. In recent years similarity search has been embedded in a more comprehensive framework and the semantic meanings behind meta paths play a crucial role in measuring similarity in heterogeneous information networks. PathSim has been considered one of the state-of-art models to find peer objects in the network. However, it only conducts similarity search in a global setting and the object attributes are not taken into consideration. In this paper, we propose OSim, a novel OLAP-based similarity search service solver. OSim is an attribute-enriched meta path-based measure to capture similarity based on object connectivity, visibility and features. A set of common attribute dimensions are defined across different types of objects and each dimension forms a hierarchical attribute tree. A path on the tree is represented by a node vector, pointing from the highest to a lowest level node. An object therefore can be described by a set of such node vectors. Online Analytical Processing techniques are further utilized in this framework to provide analysis in multiple resolutions and to improve search efficiency. Experiments show that our approaches improve search efficiency without compromising effectiveness.