Transcript sigmod02
Cube Explorer: Online Exploration of Data Cubes Jiawei Han, Jianyong Wang, Guozhu Dong, Jian Pei, Ke Wang Mining Guided Cube Explorer Novel Algorithms and Methods: Faster Creation of Iceberg Cube Predictive Gradient Analysis Multi-dimensional Gradient Mining Association and Sequence Cube Analysis Integrated with commercial software such as Microsoft OLE DB for DM, OLAP, RDMS, and DBMiner 2 Cube Explorer Benefits For Users: Superior performance and scalability Saving analysis cost and time –reusable mining queries, work directly on OLAP and relational data, Easy to use – SQL like mining, integrated with data sources Leverage OLAP & data warehouse engines– versatile functionality and strong synergy 3 Iceberg Cube Exploration Demo Novel H-Tree Iceberg Cube Creation Cube Computation with Complex Measures Dataset: large retail POS transaction data 4 Iceberg Cube Exploration Results 3D Visualization (Scatter Plot) 5 Gradient Mining Issues 1: “What products sold with ‘TV’ will significantly change profits of ‘TV’ ?” Answer: -TV profit is up 10% when sold with DVD -TV profit is down 5% when sold with VCR 2: “What are changes of housing price in Big City in 2001 comparing against 2000?” Answer: -downtown apartments go up 15% while houses in suburb go down 5% 6 How to Mine Meaningful Changes? 1 Naïve and manual method Compute two sub-cubes Big City housing in 2000 Big City housing in 2001 Tremendous costs Space Time 2 Innovation Only interesting changes wanted “gradient constraint” to capture and predict significant changes automatically 7 Gradient Mining Prediction Demo What products sold with ‘Muffins’ will change Sales of ‘Muffins’? Select ‘Muffins’ as promotion Itemset, Sales average as Measure: 8 Gradient Mining: Results 1 Most profitable patterns (Ratio >1) Rule #1: cereal increases ‘muffins” avg. sales by 8% 9 Gradient Mining: Results 2 Least profitable patterns (Ratio <1) Rule #1: Ice Cream reduces ‘muffins” avg. sales by 4% 10 Gradient Mining: Visualization Results plotted using 3D bar graph 11