Modeling Relational Data as Graphs for Mining

The focus of this paper is to develop algorithms and a framework for modeling transactional data stored in relational database into graphs for mining. Most of the real-world transactions (e.g., money withdrawal, travel, phone calls) are recorded as individual transactions which needs to be transformed into a graph based on structural relationships embedded in them. We present a graph representation that not only preserves all information embedded in a database, but also removes ambiguity and information redundancy. We present a suite of space- and time-efficient algorithms for modeling graphs from relational data. Extensive experimental analysis shows the scalability of our approaches. From a pragmatic viewpoint, our framework separates database-specific aspects from modeling aspects to make it applicable for all database systems. Real-world data has been used for generating graphs and mining them for various patterns.

• Type: Conference
• Year: 2009
• Where: COMAD
• Authors: Subhesh Pradhan, Sharma Chakravarthy, Aditya Telang
• Keywords: COMAD 2008, COMAD 2009, Real-world Transactions, Relational Database, Transactional Data