Data Mining in Heterogeneous Information Networks with Attributes

点击翻译按钮获取中文摘要

Most research in the area of data mining has concentrated on attribute data, i.e. on data where entities are represented by a set or vector of attributes. In the last decade, more and more network data has become available, which has driven the development of data mining methods for network data. However, in many scenarios, data can be represented as networks with node and/or edge attributes, with complex interactions between the topology of the network and the attributes. Nevertheless, data mining methods for attributed networks have received relatively little attention in the literature. Another limitation of existing methods is their focus on homogeneous networks, i.e. networks with a single type of objects and links, while real-life network data is often heterogeneous, consisting of multiple object and link types. Many systems can be modeled as heterogeneous information networks, for instance social networks among authors, papers and conferences, or biological networks integrating protein-protein interactions and gene regulations. Heterogeneous information networks with object and/or link attributes are characterized by complex interplays between the topology and the attributes of the network. In a social network, for instance, actors tend to connect to actors with similar attributes, while friends tend to become more similar to each other in the course of time. The long-term objective of our research is to explore data mining methods that can model and analyze the interplay of the topology and the attributes of AHINs. The proposed research program will address the limitations of the state-of-the-art and investigate several fundamental issues of data mining methods that can model and analyze the interplay of the topology and the attributes in heterogeneous information networks. The methods to be developed will have significant applications in many domains, and we will consider the analysis of social networks and biological networks as driving applications. For social network analysis, we will apply the proposed methods to the task of recommendation in social networks. For biological network analysis, we will apply our methods to the problem of network-based patient stratification and to the problem of detecting genetic causes of adverse drug reactions in gene and disease networks associated with patient records.