DOC PREVIEW
Preface to UMUAI Special Issue on Machine Learning

This preview shows page 1 out of 3 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 3 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 3 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

User Modeling and User-Adapted Interaction 8: 1–3, 1998.© 1998 Kluwer Academic Publishers. Printed in the Netherlands.1Preface to UMUAI Special Issue on MachineLearning for User ModelingIt can be argued that every interactive software system utilizes a user model, albeit,in many cases, an implicit model of the user’s objectives, and capabilities. Ratherthan such implicit models, research on user modeling has concentrated on explicitmodels that provide some form of assessment of specific attributes of the user.There are three main ways in which the content of such a model might be generatedand maintained.– it might be specified by an external source, either through pre-session config-uration, or by externally specified update;– it might be specified by the user; or– it might be specified by the software, usually on the basis of observation ofthe user’s performance.The preferences settings to be found in many modern software packages serveto illustrate the first two methods. Such preferences are usually pre-set to reflectthe manufacturer’s assumptions about the software’s community of users. The usermay then modify them to suit his or her individual requirements. Much of theresearch in user modeling, however, has concentrated on the third method by whichthe software forms the user model.Formation of a user model by observation of the user’s actions usually involvesa process of induction. The system infers a model of whatever aspects of the userare of interest – such as preferences, objectives, skills and aptitudes – from itsobservations of the user. Automated induction, such as this, has been extensivelystudied under the name ‘machine learning’. This special issue of User Modelingand User-Adapted Interaction brings together a collection of papers presenting awide variety of machine learning techniques and their use in a diverse range ofuser modeling applications.Three of the papers use approaches in which the user model has a predefinedstructure. The task of the modeling system is to infer appropriate values for thevarious variables within the model.For both Albrecht et al and Gymtrasiewicz et al these variables represent prob-abilities. Both use forms of Bayesian update to infer appropriate values for thesevariables. However, the types of models to which these Bayesian approaches areapplied differ greatly. Albrecht et al use a Dynamic Belief Network, which supportsthe combination of evidence from multiple observed independent variables in orderto assign appropriate levels probability to the potential values of the dependent167428.tex; 29/05/1998; 15:12; p.1(Disc) INTERPRINT: J.N.B. USERPRE7 (userkap:mathfam) v.1.152 PREFACE TO UMUAI SPECIAL ISSUE ON MACHINE LEARNING FOR USER MODELINGvariables of interest. In their case the independent variables are a user’s individualactions and the dependent variable is the ultimate objective of the sequence ofactions being observed. Gymtrasiewicz et al develop models of multiple interact-ing agents. Each agent models each other agent’s beliefs, desires, intentions andcapabilities. These models are further complicated by the ability to include in agentA’s model of agent B, a model of agent B’s model of agent A. This may in turn,recursively, include agent B’s model of agent A’s model of agent B, and so on.Balabanovic uses gradient descent parameter tuning to infer suitable values forthe parameters to a model of a user’s preference rankings in the context of thevery topical subject of text recommendation on the world wide web. This paperexamines the important issue of how a system that employs a user model to selectappropriate system actions should manage the conflicting objectives of selectingactions that best satisfy the existing model, and selecting actions that best supportrefinement of the existing model.The remaining papers explore machine learning techniques that infer both theappropriate structure and parameters for a model.Sison et al use conceptual clustering to form bug descriptions when modelingstudent programming errors. From analysis of incorrect Prolog programs, theirsystem generates a set of error classes, where each class represents a specificcombination of underlying misconceptions and other knowledge errors.Each class is represented by a characterization of discrepancies between theideal solution to a task and the solution that will be generated in the presence ofthe error class.Chiu and Webb use decision tree learning for modeling subtraction skills. Theypresent and compare a series of techniques for increasing the numbers of predic-tions made by the FBM-C4.5 modeling system. The models of the initial systemtake the form of a set of decision trees, where each tree makes predictions abouta specific aspect of a future actions from a description of the context in which theaction will be performed. Most of the techniques examined vary in terms of howthe disparate predictions are combined to create a specific single action prediction,although techniques are also considered for forming a single tree and for generatingand combining predictions from multiple models each of the initial form.The issue includes papers that variously present the subjects of the modelsformed as ‘agents’, ‘users’, and ‘students’. It is worth clarifying why papers onagent and student modeling are included in an issue on machine learning for usermodeling. Agents, users and students form part of a generalization hierarchy. Stu-dents are a type of user and users are a type of agent. Irrespective of the applicationdomains utilized in the various papers, all of the techniques presented have thepotential for broad applicability in a variety of user modeling contexts.There is a long history of distinguished applications of machine learning foruser modeling. The contributions to this issue show that this tradition has reacheda stage of considerable maturity and sophistication.167428.tex; 29/05/1998; 15:12; p.2PREFACE TO UMUAI SPECIAL ISSUE ON MACHINE LEARNING FOR USER MODELING 3Author’s VitaDr G. WebbDeakin University, School of Computing and Mathematics, Geelong, Vic., Aus-tralia, 3217. Home page: http://www.cm.deakin.edu.au/ webbDr Geoff Webb is a Reader in Computing at Deakin University where he has es-tablished and leads the Deakin University Knowledge Acquisition and Processingresearch group. He received his B.A. and Ph.D. degrees in Computer Science fromLa Trobe University. The author of more than sixty


Preface to UMUAI Special Issue on Machine Learning

Download Preface to UMUAI Special Issue on Machine Learning
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Preface to UMUAI Special Issue on Machine Learning and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Preface to UMUAI Special Issue on Machine Learning 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?