TupleRank: Ranking Relational Databases using Random Walks on Extended K-partite Graphs

  • Author(s) / Creator(s)
  • Technical report TR09-02. The significant increase in open access digital information has created incredible opportunities for modern database research, especially in exploiting significant computational resources to determine complex relationships within those data. In this paper, we consider the problem of analyzing relational databases and explaining relationships between entities in order to rank tuples based on a notion of relevance. For this purpose, we propose a solution of a class of link analysis algorithms known as the random walk, which can be deployed to discover interesting relationships amongst partial tuples of relational databases that would otherwise be hard to expose. We focus on a shortcoming of the absence of a special kind of relationship, which we call \"returning relationship\". We demonstrate our ideas on the DBLP database, where we exploit structural variations on relationships between authors, conferences, topics, and co-authorships. We show how a distinction between normal relations and returning relations on objects within that database provides the basis for structuring a random walk algorithm to determine interesting relevance measures. We also show how structural changes in the organization of the random walk can produce novel results that are not attainable with previous database ranking methods. | TRID-ID TR09-02

  • Date created
  • Subjects / Keywords
  • Type of Item
  • DOI
  • License
    Attribution 3.0 International