Jungsoon P. Yoo

Generalized clustering, supervised learning, and data assignment

Clustering algorithms have become increasingly important in handling and analyzing data. Considerable work has been done in devising effective but increasingly specific clustering algorithms. In contrast, we have developed a generalized framework that accommodates diverse clustering algorithms in a systematic way. This framework views clustering as a general process of iterative optimization that includes modules for supervised learning and instance assignment. The framework has also suggested several novel clustering methods. In this paper, we investigate experimentally the efficacy of these algorithms and test some hypotheses about the relation between such unsupervised techniques and the supervised methods embedded in them.

An adaptive stock tracker for personalized trading advice

The Stock Tracker is an adaptive recommendation system for trading stocks that automatically acquires content-based models of user preferences to tailor its buy and sell advice. The system incorporates an efficient algorithm that exploits the fixed structure of user models and relies on unobtrusive data-gathering techniques. In this paper, we describe our approach to personalized recommendation and its implementation in this domain. We also discuss experiments that evaluate the systems behavior on both human subjects and synthetic users. The results suggest that the Stock Tracker can rapidly adapt its advice to different types of users

Modeling student online learning using clustering

This paper discusses and evaluates a modeling approach for student online learning. It is developed as a key component of an adaptive online tutoring system, AToL. At the beginning of the learning process, classification of student learning style is applied based on each students responses to a few learning related questions. Clustering is then used to model student behavior for each learning style using data collected as the students interact with the system. A Bayesian Markov chain based temporal data clustering method is developed for this step. We evaluated the student modeling component of the AToL system using data collected from the CS-I students who participated in the experiments in Spring 05. We compared the quality of the models built using these two approaches. We also compared the models built for the same group of students when learning different concepts.

Student progress monitoring tool using treeview

In this paper, we present an extensible visualization tool that is being used in a web-based closed laboratory system. The goal of this project is to provide a tool for both students and teachers that can help trace deficiencies in a students understanding back to individual concepts. This visualization tool has been developed by using the tree abstract data type (ADT) which is built from the concepts to be covered in a lab, lecture, or course. Once the tree ADT is built, each node can be associated with different entities such as student performance, class performance, or lab development. Using this tool, a teacher can help students by discovering concepts that need additional lecture coverage, and students may discover concepts for which they need to spend additional time working on reinforcement exercises.

Software testing: a machine learning experiment

Testing is a critical part of the somare development process. As the cost of so&are development has escalated, attempts to accurately estimate the cost (and time) of so&are testing have become more important. Research is being done to predict soware development costs and to develop tools to help in cost estimations. In this research, machine learning techniques are applied to determine which sojlware testing attributes are important in predicting soBware testing costs and more specifically soflware testing time. Testing data on 25 sofrware projects were collected. Using this database, a machine karning system identifies the factors that affect testing time by creating a classification tree in which nodes share similar values. By analyzing the classification tree we are able to determine the salient factors which placed a program into its group. 77ze factors that we consider include code complexity measures, measures of programmer and tester experience, measures of the use of sohare engineering principles such as structured programming techniques, andstatistics collected during actual testing of the projects. ‘Ihe programs in a node have an average testing time that is reflective of any program whose attributes would place it into this group. Thus, the tree can be used, among other things, to predict testing time. PariialIy suFFo& by NSF Iu-Grant DUE-9352219. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copyright is by permission of the Association of Computing Machinery. To copy otherwise, or to republish, requires a fee and/or swcific txxmission. 8 1995 ACM O-89791-737-5

Adaptive teaching strategy for online learning

Finding the optimal teaching strategy for an individual student is difficult even for an experienced teacher. Identifying and incorporating multiple optimal teaching strategies for different students in a class is even harder. This paper presents an Adaptive tutor for online Learning, AtoL, for Computer Science laboratories that identifies and applies the appropriate teaching strategies for students on an individual basis. The optimal strategy for a student is identified in two steps. First, a basic strategy for a student is identified using rules learned from a supervised learning system. Then the basic strategy is refined to better fit the student using models learned using an unsupervised learning system that takes into account the temporal nature of the problem solving process. The learning algorithms as well as the initial experimental results are presented.

Concept formation in numeric domains

COBWEB is a concept formation system that organizes observations into a classification tree. However, it can only handles symbolic data. COBWEB/3 handles numeric data assuming a normal distribution of attribute values. However, COBWEB/Q requires a userspecified acuity value that has to be carefully tuned to obtain the optimal concept hierarchy. However, single acuity value may not be appropriate to all attributes which may have various ranges of values. In this paper, we show how important the normalization of each attribute is, and present an algorithm that can properly handle symbolic and/or numeric data. Topics : Artificial Intelligence, Machine Learning

Intelligent tutoring system for CS-I and II laboratory

A Web-based adaptive tutoring system which dynamically adapts to each students needs and gives a student immediate feedback is being developed for our CS-I and CS-II closed laboratories. The system currently contains the question tutor, the program tutor, and the course management components. The tutoring components help students learn programming concepts through hands-on, self-paced exercises. The course management component helps teachers prepare and maintain the lab materials. Experiments have been conducted to evaluate the effectiveness of this new tutoring system and promising preliminary results were obtained.

A hybrid conceptual clustering system

Conceptual clustering is used to organize observations into an abstract hierarchy which can be used to predict classes and/or attribute values. Typically this clustering has been done using either incremental or nonincrementallearning. Incremental learning suffers from the ordering problem, while nonincremental learning cannot handle a dynamic environment efficiently. This paper proposes a hybrid conceptual clustering system which uses two learning algorithms. The first stage is based on Genetic Algorithms and is used as a preprocessor for the second stage. The second stage is an incremental learning system. This hybrid conceptual clustering system overcomes the difficulties encountered by using either a nonincremental or an incremental learning system alone.

Can algotutor change attitudes toward algorithms

The ability to design an algorithm is one of the most important learning outcomes of a computer science program. Unfortunately, not only is learning how to design algorithms a challenging task, but many students believe that algorithm design is not an important part of problem solving. To address this challenge and hopefully change student attitudes, we developed AlgoTutor, a web-based algorithm development tutoring system. AlgoTutors primary components are the algorithm composer and the algorithm tracer. A third component, ProgramPad, was added to show the connection between algorithms and code. This paper presents the results of experiments that assessed AlgoTutors effectiveness in changing student attitudes about algorithm development. The results show that students who used AlgoTutor in CS-I were more likely to realize the importance of algorithm design in problem solving and to have confidence in their own algorithm development abilities.