| In most machine learning and data mining tasks, data is typically assumed to be independent and identically distributed. In real applications, this assumption is not always correct. Data points are related to each other. Taking these relationships into account is a challenge but if done properly can provide new insights into the data. Data represented with its inner relationships is called information networks. A Social Network or Information Network is a structure made up of nodes representing entities, and edges representing the relationships among nodes. Understanding the behaviour of social networks is known as Social Network Analysis (SNA). SNA is used to study organizational relations, to analyse citation or computer mediated communications, etc. One of the most important applications of SNA is to find the similarity/relevance among entities in the network for a specific query. Finding the relevance between different entities, we are able to rank them based on each other. Ranking a set of entities with respect to one instance is required in many application domains. For example, in E-Advertisement, the goal is to show the most related advertisement to each user. This essentially means to rank the advertisements based on each user and to show the high ranked ones to the user. A researcher has a new idea in a particular topic. Wanting to publish the idea in the right place, an ordered list of conferences with respect to that particular topic is required. A person is eager to know the similar games to a favourite game, to prevent buying many irrelevant games. Taking a query, a search engine wants to rank all the web pages with respect to it. Consequently the user explore the most relevant pages rather than reviewing all the pages which only match with the keywords in the query.;In this study we focus on ranking the entities in heterogeneous multiple relation social networks, networks for which nodes belong to different classes and relationships have different types. We investigate social networks from bibliographic databases with authors, conferences and topics. Analysing such networks is a non-trivial task dealing with large k-partite graphs. We propose an algorithm to find the most related entities for each instance. We develop a tool called DB-Connect which applies our method on the academic social network. |
