Adaptive user support in educational environments: A Bayesian Network approach

Adaptive user support in educational environments:
A Bayesian Network approach

A. G. Stoica University «Dunãrea de Jos » of Galaţi Faculty of Economic & Administrative Sciences Str. Nicolae Bãlcescu nr. 59-61, Galaţi 6200, România ace@k.ro N.K Tselios, C. Fidas University of Patras ECE Department – HCI Group GR-26500 Rio Patras, Greece nitse@ee.upatras.gr, fidas@clab.ee.upatras.gr

SUMMARY This paper is concerned with the design and implementa- tion of an innovative user support system in the frame of an open educational environment. The environment adapted is ModelsCreator (MC), an educational system supporting learning through modelling activities. The pupils’ typical interaction with the system was modelled using Bayesian Belief Networks (BBN). This model has been used in ModelsCreator to build an adaptive help system providing the most useful guidelines according to the current state of interaction. A brief description of the system and an overview of application of Bayesian tech- niques to educational systems is presented together with discussion about the process of building of the Bayesian Network derived from actual student interaction data. A preliminary evaluation of the developed prototype indi- cates that the proposed approach produces systems with promising performance. KEYWORDS: Bayesian Belief Networks, educational systems, inference algorithms, on-line adaptation, Mod- elsCreator. INTRODUCTION During the last years a number of open problem-solving environments have been built that are based on the con- structivist approach. Because of the nature of these envi- ronments, user interaction in their context could be very rich and the patterns of use of the tools that are included in them cannot be fully anticipated. The complexity of such environments can often lead to poor usability, which can be an obstacle to obtaining the expected peda- gogical value from their use. Improving usability of these environments is an objective that can be achieved in var- ious ways. For instance various usability evaluation tech- niques that take in consideration both the pedagogical value and the usability of the environments have been proposed [1,13]. Also more complex task modelling ap- proaches have been proposed for the design of such envi- ronments [16]. Also the support at run time to the user through adaptive user support systems is an approach that can improve usability. It should be recognised that
overall artificial intelligence techniques have not suc- ceeded to deliver the expected results in the educational field, through the Intelligent Tutoring Systems. However the premise of adaptive system behaviour through which higher usability and increased system transparency can be obtained, remains a valid scientific objective. In re- cent years the heavy modelling involved in traditional AI approaches has been replaced by implicit models built from rich data sets through techniques proposed by the machine learning and knowledge discovery fields. These techniques have produced during the last years efficient algorithms and found new areas of applications. Among them a technique that has certain advantages and has been extensively used during the last years are the Bayes- ian Belief Networks (BBN’s) [11]. A Bayesian Belief Network [14] is a directed acyclic graph where each node represents a random variable of interest and each edges represents direct correlations between the varia- bles. Because of the underlying probabilistic model that describes the belief on the existence of a specific event, BBNs are considered one of the strongest ways to repre- sent uncertainty. By capturing decisions on accurate cog- nitive models of the users and then modelling uncertainty in human computer interaction, the modelling of the user, the user’s interface behaviour and, therefore, the effi- ciency of the interaction can be improved. Bayesian rea- soning is based in formal probability theory and is used extensively in several current areas of research, including pattern recognition and classification. Assuming a ran- dom sampling of events, Bayesian theory supports the calculation of more complex probabilities from previous- ly known results [10]. The advantages of BBNs include the simple process for constructing probabilistic net- works even from relative a small amount of data, the ef- ficient algorithms to evaluate probabilities for instances of a node and the versatile knowledge representation which such networks provide. During the reported experiment we have attempted to build such an adaptive user support system for an open educational environment using Bayesian Networks. In this paper we first present an overview of Bayesian prob- abilistic theory together with a brief description of BBNs and their implementation in educational environments. Then a short description of ModelsCreator is included, i.e. the system, which was us

…(Full text truncated)…