From Data to Knowledge WebBased Knowledge Engineering System

From Data to Knowledge: Web-Based Knowledge Engineering System C. -C. Chan Department of Computer Science University of Akron, OH 44325 -4003 USA chan@uakron. edu UA Faculty Forum 2008 by C. -C. Chan 1

Outline Overview of Data Mining Software Tools A Rule-Based System for Data Mining Concluding Remarks UA Faculty Forum 2008 by C. -C. Chan 2

Data Mining (KDD) From Data to Knowledge Process of KDD (Knowledge Discovery in Databases) Related Technologies Comparisons UA Faculty Forum 2008 by C. -C. Chan 3

Why KDD? We are drowning in information, but starving for knowledge John Naisbett Growing Gap between Data Generation and Data Understanding: Automation of business activities: Telephone calls, credit card charges, medical tests, etc. Earth observation satellites: Estimated will generate one terabyte (1015 bytes) of data per day. At a rate of one picture per second. Biology: Human Genome database project has collected over gigabytes of data on the human genetic code [Fasman, Cuticchia, Kingsbury, 1994. ] US Census data: NASA databases: … World Wide Web: UA Faculty Forum 2008 by C. -C. Chan 4

Process of KDD [1] Fayyad, U. , Editorial, Int. J. of Data Mining and Knowledge Discovery , Vol. 1, Issue 1, 1997. [2] Fayyad, U. , G. Piatetsky-Shapiro, and P. Smyth, "From data mining to knowledge discovery: an overview, " in Advances in Knowledge Discovery and Data Mining , Fayyad et al (Eds. ), MIT Press, 1996. UA Faculty Forum 2008 by C. -C. Chan 5

Process of KDD 1. Selection Ø Ø 2. Pre-Processing Ø 3. Ø Choosing the functions and algorithms of data mining Association rules, classification rules, clustering rules Interpretation and Evaluation Ø 6. Data reduction and projection Data Mining Ø 5. Data cleaning and preprocessing Transformation Ø 4. Learning the application domain Creating a target dataset Validate and verify discovered patterns Using discovered knowledge UA Faculty Forum 2008 by C. -C. Chan 6

Typical Data Mining Tasks Finding Association Rules [Rakesh Agrawal et al, 1993] Each transaction is a set of items. Given a set of transactions, an association rule is of the form X Y where X and Y are sets of items. e. g. : 30% of transactions that contain beer also contain diapers; 2% of all transactions contain both of these items. Applications: Market basket analysis and cross-marketing Catalog design Store layout Buying patterns UA Faculty Forum 2008 by C. -C. Chan 7

Finding Sequential Patterns Each data sequence is a list of transactions. Find all sequential patterns with a user-specified minimum support. e. g. : Consider a book-club database A sequential pattern might be 5% of customers bought “Harry Potter I”, then “Harry Potter II”, and then “Harry Potter III”. Applications: Add-on sales Customer satisfaction Identify symptoms/diseases that precede certain diseases UA Faculty Forum 2008 by C. -C. Chan 8

Finding Classification Rules Finding discriminant rules for objects of different classes. Approaches: Finding Decision Trees Finding Production Rules Applications: Process loans and credit cards applications Model identification UA Faculty Forum 2008 by C. -C. Chan 9

Text Mining Web Usage Mining Etc. UA Faculty Forum 2008 by C. -C. Chan 10

Related Technologies Database Systems MS SQL server Transaction databases OLAP (Data Cubes) Data Mining Decision Trees Clustering Tools Machine Learning/Data Mining Systems CART (Classification And Regression Trees) C 5. x (Decision Trees) WEKA (Waikato Environment for Knowledge Analysis) LERS ROSE 2 Rule-Based Expert System Development Environments CLIPS, JESS EXSYS Web-based Platforms Java MS. Net UA Faculty Forum 2008 by C. -C. Chan 11

Comparisons Pre. Processing Learning Data Mining Inference Engine End-User Interface Web-Based Access Reasoning with Uncertainties MS SQL Server N/A Decision Trees Clustering N/A N/A CART C 5. x N/A Decision Trees Built-in Embedded N/A WEKA Yes Trees, Rules, Clustering, Association N/A Embedded Need Programming N/A CLIPS JESS N/A Built-in Embedded Need Programming 3 rd parties Extensions UA Faculty Forum 2008 by C. -C. Chan 12

Rule-Based Data Mining System Objectives Develop an integrated rule-based data mining system provides Synergy of database systems, machine learning, and expert systems Dealing with uncertain rules Delivery of web-based user interface UA Faculty Forum 2008 by C. -C. Chan 13

Structure of Rule-Based Systems UA Faculty Forum 2008 by C. -C. Chan 14

System Workflow Input Data Set Data Preprocessing Rule Generator UA Faculty Forum 2008 by C. -C. Chan User Interface Generator 15

Input Data Set: Text file with comma separated values (CSV) It is assumed that there are N columns of values corresponding to N variables or parameters, which may be real or symbolic values. The first N – 1 variables are considered as inputs and the last one is the output variable. Data Preprocessing: Discretize domains of real variables into a finite number of intervals Discretized data file is then used to generate an attribute information file and a training data file. Rule Generator: A symbolic learning program called BLEM 2 is used to generate rules with uncertainty User Interface Generator: Generate a web-based rule-based system from a rule file and corresponding attribute file UA Faculty Forum 2008 by C. -C. Chan 16

Architecture of RBC generator Requests Client SQL DB server Middle Tier Responses Workflow of RBC generator Rule set File Metadata File RBC Generator Rule Table Definition UA Faculty Forum 2008 by C. -C. Chan 17

Concluding Remarks A system for generating rule-based classifier from data with the following benefits: No need of end user programming Automatic rule-based system creation Delivery system is web-based provides easy access UA Faculty Forum 2008 by C. -C. Chan 18

Project Status The current version 1. 4 of our system provides fundamental features for data mining from data including: Data Preprocessing Management of preprocessed data files Machine Learning tool to generate rules from data Rule-Based Classifier system supporting uncertain rules Web-Based access UA Faculty Forum 2008 by C. -C. Chan 19

Future Work More advanced features in Data Preprocessing such as data cleansing, data transformation, and data statistics Learning from multi-criteria inputs with preferential rankings to support Multiple Criteria Decision Making processes Concept-Oriented information retrieval and search UA Faculty Forum 2008 by C. -C. Chan 20

Thank You! UA Faculty Forum 2008 by C. -C. Chan 21