Abstract-Papers up to 4 Pages Should Be Submitted Using This Format s5

Improvement of Web-Based Intelligent Tutoring Systems with the use of fuzzy logic

CONSTANTINOS KOUTSOJANNISa,b, IOANNIS HATZILYGEROUDISa aDept of Computer Engineering & Informatics School of Engineering University of Patras bDepartment of Nursing, School of Health Sciences, Technological Educational Institute of Patras 26500 RION Patras, Greece

Abstract In this paper, we improve our Web-based Intelligent Tutoring System (ITS) for distant health professionals in fundamental aspects of health care technology, incorporating fuzzy logic. Our system offers course units covering the needs of users with different knowledge levels and personal characteristics. It tailors the presentation of the educational material to the users’ diverse needs by using AI techniques to specify each user’s model as well as to make pedagogical decisions, in An environment with linguistic or imprecise data. This is achieved via an expert system that uses a hybrid knowledge representation formalism integrating fuzzy rules with neurocomputing. The final model based on fuzzy neurons is more global and efficient, since it requires the least possible training effort, the number of the produced units is kept as small as possible, and it works in an environment of fuzzy as well as crisp data. According to our preliminary results the number of the students that were used the system and passed ITS’s tests is not significantly different from them that were participated the same course in our previous system or in the traditional classroom.

Keywords: Web Based Education, Hybrid Expert System, Biomedical technology Introduction that are equipped with any kind of Internet-connected computer. Thousands of Web-based courses and other The introduction of new technologies in health care can be seen educational applications have been made available on the Web as improved diagnosis and treatment as well as increased within the last years. The problem is that most of them are efficiency and quality of the services involved. At the same nothing more than a network of static hypertext pages. A time, however, the implementation of these new improved challenging research goal is the development of advanced Web- technologies, introduces concerns about trained personnel and based educational applications that can offer some amount of imposes new demands for tools and methods for their overall adaptivity and intelligence. These features are important for management. The support services, for biomedical technology, WBE applications since distance students usually work on their address a variety of technical and administrative issues, own (often from home). An intelligent and personalized concerning the safe and efficient operation of medical assistance that a teacher or a peer student can provide in a equipment over the period of its intended use. Health care normal classroom situation is not easy to get. In addition, being technology is our first line of defense against any disease. adaptive is important for Web-based courseware because it has Health care professionals have to be familiar with the most to be used by a much wider variety of students than any common medical equipment used in health care units. "standalone" educational application. A Web courseware that is Educational programs already include courses with the designed with a particular class of users in mind may not suit fundamental aspects of the proper care and safety of medical other users. Since the early days of the Web, a number of instruments [6],[12]. Thus our nursing department’s research teams have implemented different kinds of adaptive curriculum includes the “Biomed” course teaching fundamental and intelligent systems for on-site and distance WBE. aspects of nursing-medical informatics and the most common Intelligent Tutoring Systems (ITSs) constitute an advanced medical devices including safety and regulations according to generation of CAIs. Their key feature is their ability to provide internationally recognized quality standards and EU Directives. a user-adapted presentation of the teaching material [2],[14], Throughout the last years biomedical education as well as [15]. This is accomplished by using Artificial Intelligence medical practice education depends widely on Computer Aided techniques to represent the pedagogical decisions and the Instruction (CAI) “standalone” systems (i.e., non-Web-based) information regarding each student. ITSs have become [9],[5],[11]. Web-based education (WBE) is currently a hot extremely popular during the last years and have been shown to research and development area. Benefits of Web-based be quite effective at increasing users’ performance and education are clear: classroom independence and platform motivation [1]. The emergence of the World Wide Web independence. Web courseware installed and supported in one increased the usefulness of such systems [3],[4],[6],[10],[13], in place can be used by thousands of learners all over the world distant learning environments. In this paper, we present the architecture and the first results The knowledge concepts refer to basic pieces of knowledge of the use of a Web-based ITS for teaching the use and safety concerning the domain of health care technology. Every aspects of health care technology. The system is addressed concept has a number of general attributes such as, its name, its health professionals. It includes course units covering the needs level of difficulty etc. Furthermore, it can have links to other of users with different knowledge levels and characteristics. concepts. These links denote its prerequisite concepts. In this The system models the users’ knowledge state and skills. Based way, one or more concept networks are formed representing the on this information, it constructs lesson plans and selects the pedagogical structure of the domain to be taught. appropriate course units for teaching each individual user. The The concepts are organized into concept groups. A concept functionality of the system is controlled by an expert system group contains closely related concepts based on the knowledge based on fuzzy neurules, a type of hybrid rules integrating they refer to. Therefore, the domain space is dissected into fuzzy rules with neurocomputing [6],[7]. subdomains. For instance, a concept group can involve magnetic resonance imaging whereas some of its concepts may System Description concern the structure or the manipulation of the specific equipment. Fig. 1 depicts the basic architecture of the ITS. It consists of The course units constitute the teaching material presented the following components: to the system users as Web pages. The teaching material can  the domain knowledge, containing the structure of the form a variety of courses starting from introductory topics and domain and the educational content. The educational scaling up to more advanced ones. Each course unit is content refers to the following main chapters: magnetic associated with a knowledge concept. The user is required to resonance, computed tomography, nuclear medicine, know the concept’s prerequisite concepts in order to be able to electro-medicine, ultrasound, safety and regulations grasp the knowledge contained in the corresponding course according to internationally recognized quality standards unit. The distinct representation of the domain’s pedagogical and EU Directives. structure (concepts) and the actual teaching content (course  the user modeling component, which records information units) facilitates the updates in the domain knowledge. A course concerning the user, unit may present theory, may be an example or an exercise.  the pedagogical model, which encompasses knowledge Examples assist a user in grasping the theory's key points. regarding various pedagogical decisions, Exercises as questions with multiple choice answers are based  the user interface for user-system interaction on the examples and are used to evaluate a user’s knowledge level. When solving an exercise the user can ask the system for The ITS is based on an expert system aiming to control the help and view related examples. Each exercise is associated teaching process. The expert system employs a hybrid with an explanation assisting the user in case of a wrong knowledge representation formalism, called fuzzy neurules [6], answer. [7]. In the following, we elaborate on the system's key aspects. The pedagogical model based on the user model selects and orders the course units presented to the user. In this way, a user- adapted presentation of the teaching material is achieved. To this end, the explanation variant method is used, implemented by the page variant technique [4]. More specifically, the system keeps variants of the same page (course unit) with different presentations. To facilitate the selection and ordering of the course units, the domain knowledge includes a meta- description of the course units based on their general attributes. Main such attributes for a course unit are its level of difficulty, its pedagogical type (theory, example, exercise), its multimedia type (text, image, animation, interactive simulation), the required Internet connection, etc.

B. User Modeling Component Fig. 1. - Architecture of the ITS. User modelling remains at the core of ITS research. What distinguishes ITS from CAI is the goal of being able to respond A. Domain Knowledge to the individual student's learning style to deliver customized The domain knowledge contains knowledge regarding the instruction. Although some authors are questioning the goal of subject being taught as well as the actual teaching material. It student modelling either because of technical limitations or consists of two parts: (a) the knowledge concepts and (b) the larger philosophical grounds, this is still an area of active course units. research. In our system the user-modeling component is used to record information concerning the user which is vital for the system's user-adapted operation. It contains models of the system's users and mechanisms for creating these models (Fig. 2). The user model consists of four types of items: (i) personal data (e.g. name, email), (ii) interaction parameters, (iii) knowledge of the concepts and (iv) student characteristics. The student characteristics and the knowledge of the concepts directly affect the teaching process, whereas the interaction parameters indirectly. The interaction parameters form the basis of the user model and constitute information recorded from the interaction with the system. They represent things like, the type and number of the units accessed, the type and the amount of help asked, the answers to the exercises etc. The student characteristics are mainly the following: Fig. 2. -The User Modeling Component. (a) Multimedia type preferences (e.g. text, images, or animations) regarding the presented course units. In order to maintain user models, only registered users have (b) Knowledge level (novice, beginner, intermediate, access to the system. A registered user identifies himself/herself advanced) of the subdomains and the whole domain. to the system whenever logs on by giving a valid login name (c) Learning ability level. and password. An unregistered user must first submit personal (d) Concentration level. data (e.g. name, email) to the system in order to obtain an (e) Experience concerning the use of computers, hypermedia account and access to the system's functionality. applications and the specific ITS. A new user must give values for the directly obtainable (f) Available Internet connection. student characteristics. The user has the option to change them The student characteristics are represented with the during the teaching process. Furthermore, a new user is stereotype model, that is the user is assigned to predefined required to take some tests in order to initialize the other classes (stereotypes). The stereotypes denote typical users. student characteristics. Based on the way they acquire their values, student characteristics are discerned into two groups. They can be C. Pedagogical Model either directly obtainable or inferable. The directly obtainable The pedagogical model represents the teaching process. It characteristics such as (a) and (f) obtain their values directly provides the knowledge infrastructure in order to tailor the from the user whereas the values of the inferable ones such as presentation of the teaching material according to the characteristics (b)-(e) are inferred by the system based on the information contained in the user model. As shown in Fig. 1, interaction parameters and the knowledge of the concepts. The the pedagogical model consists of three main components: (i) knowledge level of the whole domain is deduced from the concept neurules, (ii) course units’ fuzzy neurules and (iii) knowledge levels of its subdomains. A neurule base containing evaluation module. classification fuzzy neurules is used to derive the values of the The task of the concept fuzzy neurules is to construct a user- inferable characteristics. The user models are updated during adapted lesson plan by selecting and ordering the appropriate the teaching process. concepts. This is based on the user's knowledge of the concepts, The stereotype model cannot sufficiently represent the user’s the user's (sub)domain knowledge level, the concepts' level of knowledge of the domain because the adaptation techniques of difficulty and the links connecting the concepts. the pedagogical model require a more fine-grained model in According to the plan constructed by the concept fuzzy order to be effective. For this reason the user’s knowledge of neurules, the course units’ fuzzy neurules select and order the the domain is represented as a combination of a stereotype and course units that are suitable for presentation. For this purpose an overlay model. The stereotype denotes the (sub)domain the student characteristics of the user model as well as the knowledge level. The overlay model is based on the concepts meta-description of the course units are taken into account. associated with the course learning units. More specifically, The evaluation module evaluates the user's performance and each concept is associated with a value denoting the user updates accordingly the user model. More specifically, it knowledge level of this concept. assigns knowledge values to the concepts based on the The combination of stereotype and overlay modeling for the interaction parameters and updates the inferable student representation of a user’s domain knowledge has given good characteristics based on the classification fuzzy neurules of the results [4]. The overlay model has the problem of initialization user modeling component. When the user gains an acceptable since for a new user it is hard to set the knowledge values for knowledge level of the concepts belonging to the initial lesson all of the concepts. This difficulty can be overcome by plan, a new plan is created. associating a fixed set of concept-value pairs with each The user can intervene in the teaching process. An stereotype [4]. This association will be based on the concepts' experienced user has the ability to set learning goals based on level of difficulty. the concept groups. For this purpose, the user can view a list of the concept groups and choose the concept group he/she is more interested in. The list of concept groups is sorted in D.2 Fuzzy neurules & fuzzy inference decreasing order according to the user's knowledge state of A hybrid neural net, like the one consisted of neurules; is a their concepts. Therefore, the system supports active learning neural net with crisp signals and weights and crisp transfer increasing the user’s performance and motivation and giving function [17, 10]. However, we can: 1) combine inputs and him/her some control on its adaptation. These features increase weights using a t-norm or some other continuous operation, 2) the system’s usability taking into consideration the fact that the aggregate the results using a t-norm or some other continuous users will be adults. function and transfer function can be a continuous function from input to output. We emphasize that all inputs, outputs and D. The Expert System the weights of a hybrid neural net are real numbers taken from D.1 Neurules interval [0,1]. This modification leads to a structure of fuzzy Our ITS operates under the control of a hybrid expert neuron [12], based on fuzzy operators. Using fuzzy logical system. The expert system has an inference engine, to make neurons [11], the output is more or less influenced by the decisions based on the known facts and the rule bases contained values of inputs. This influence depends on both the weights in the user modeling component and the pedagogical model. and the operation of fusion: 1) for a neuron of type AND, the The expert system's knowledge representation formalism is influence of its inputs having a weak weight is most important based on neurules, a type of hybrid rules integrating symbolic (min) 2) for a neuron of type OR the inputs whose weight is rules with neurocomputing. The attractive feature of neurules is significant are rather taken into account (max) This defined the that they improve the performance of symbolic rules [6] and interval of possible values for the output. Our fuzzy inference simultaneously retain their naturalness and modularity [7] in system is composed of five functional blocks (Fig.4). contrast to other hybrid approaches.  A rule base containing a number of fuzzy if-then rules The form of a neurule is depicted in Fig. 3a. Each condition  A database which defines the membership functions Ci is assigned a number sfi, called its significance factor. of the fuzzy sets used in the fuzzy rules Moreover, each rule itself is assigned a number sf0, called its  A decision-making unit which performs the inference bias factor. Internally, each neurule is considered as an adaline operations on the rules unit (Fig. 3b). The inputs Ci (i=1,...,n) of the unit are the  A fuzzification block which transforms crisp inputs conditions of the rule. The weights of the unit are the into degrees of match with linguistic values significance factors of the neurule and its bias is the bias factor  A defuzzyfication block which transform the fuzzy of the neurule. Each input takes a value from the following set results of the inference into a crisp output [13, 15]. of discrete values: [1 (true), 0 (false), 0.5 (unknown)]. The output D, which represents the conclusion (decision) of the rule, is calculated via the formulas:

(1) where a is the activation value and f(x) the activation function, a threshold function (f(a) = 1, if a  0, and f(a) = -1, otherwise). Hence, the output can take one of two values, ‘-1’ and ‘1’, representing failure and success of the rule respectively. Neurules are constructed either from empirical data (training patterns) or symbolic rules. Each neurule is individually trained via the LMS (Least Mean Square) algorithm. In case of inseparability in the training set, special techniques are used [6],[7]. In this way, the neurules contained in the pedagogical model and the user modeling component are constructed. The inference mechanism is based on a backward chaining strategy. Fig.4 Fuzzy inference system

The steps of fuzzy reasoning are: 1) compare the input variables with the membership functions on the premise part to obtain the membership values (fuzzyfication) 2) combine (trough a specific t-norm operator, usually multiplication or min), the membership values on the premise part to get firing strength (weight) of each Fig. 3. - (a) Form of a neurule (b) corresponding adaline unit rule 3) generate the qualified consequent (fuzzy or crisp) of For the development of the system, the Microsoft Internet each rule depending on the firing strength Information Server and the MS SQL Server for Windows NT 4) aggregate the qualified consequences to produce a are used. Active Server Pages scripts manipulate the crisp output (defuzzification). information stored in the databases and dynamically produce the contents of the applications presented to the users. E. The Fuzzy hybrid architecture The term ‘fuzzy neural networks’ denotes neural networks that Results are fuzzy and deal with fuzzy signals and fuzzy weights. By employing a hybrid neural learning procedure, this architecture A number of nursing students (18 – 20 years old) that were refines fuzzy if-then rules obtained from human experts, to participated the project classified in three different groups: A. describe the input-output behavior of a complex system [16,17]. 20 students that were used the system as “standalone”; B. 20 Even if human expert is not available, it can still set up students that were used the previous system as “standalone” intuitively reasonable initial membership functions and start the and finally C. 10 students as a control group; that were in a learning process to generate a set of if-then fuzzy rules to regular classroom with the teacher and the appropriate approximate a desired data set. In Fig. 4 the functional presentation equipment, undergoing the same tests and architecture of our hybrid system (except the fuzzyfication and activities, that were used from the system. Table 1 presents the defuzzyfication modules) to incorporate the integrated preliminary results of their progress after a 3 month intensive formalism is presented [18,19]. The run-time system consists of course including 4 hours/week in the classroom or in the lab, three modules, functionally similar to those of a conventional using Fisher Exact Test for difference estimations. fuzzy rule-based system: hybrid rule base (HRB), hybrid inference mechanism (HIM) and working memory (WM). Table 1 - Results of the “Biomed” course under ITS (with or without a tutor) as well as in traditional classroom

Student Group Passed Rejected P A (fuzzy ITS) 13 7 0,7 B (non fuzzy ITS) 12 8 1,0 C (control) 7 3 Total 32 18

According to our results there were no significant differences between the number of the students that were users of the system and the students of the traditional “Biomed” course in the school of nursing. The system will be soon available at the Fig.5 - The fuzzy neurule-based architecture site of the nursing school: http://www.teipat.gr/pages/kedd/seyp2/biomed2/. HRB (corresponding to the Knowledge Base in Fig. 3) contains fuzzy neurules produced by conversion from the fuzzy Conclusion rule base (FRB) via the conversion and training mechanism The ever-increasing specialization of nursing care may (CTM). FRB contains only fuzzy rules. So, production of the require nursing professionals to provide nursing care outside of fuzzy neurules takes place before run-time. their specialty. Nurses will have to familiarize themselves with HIM (corresponding to the Inference Engine in Fig.3) is a new specialization area at short time [12], [9]. From another responsible for making inferences by taking into account the side it should be clear to anyone who is familiar with the needs initial conditions in the WM and the fuzzy neurules in the HRB. of Web-based education, that adaptive and intelligent WM contains facts. A fact has the same format as a technologies can enhance different sides of Web-based condition/conclusion of a rule [6]. educational systems. Adaptive presentation can improve the F. User Interface usability of course material presentation. In this paper, we The user interface is responsible for the system’s interaction describe the design of a Web-based Intelligent Tutoring System with the user. Due to the fact that it is a layer of the system that (ITS) for teaching nursing students fundamental aspects of communicates directly with the user it should be carefully medical informatics and the most common medical equipment. designed [8]. The primary objective is to design a user interface It will also support training of hospital personnel, safe and that will be usable by users with diverse abilities, needs, efficient use of medical equipment and quality assurance of requirements and preferences. services offered by the clinical engineering department, according to internationally recognized quality standards. The Implementation aspects focus has been placed on the effective management of medical equipment with significant expected benefits relating to safety [8] P.A.M. Kommers, S. Grabinger, J.C. Dunlap, Eds., and proper use. Hypermedia learning environments: instructional design and The system tailors the presentation of the teaching material integration, Lawrence Erlbaum Associates, 1996. to the diverse needs of its users. To some degree, the users can [9] J.M. Lappe, B. Dixon, L. Lazure, P. Nilsson, J. Thielen, J. intervene in the teaching process. The system’s function is Norris, “Nursing education application of a computerized controlled by a hybrid expert system. Fuzzy neurules can nursing expert system”, J. Nurs. Educ., vol. 29(6), pp. 244-248, handle imprecise data and uncertainty as well as numerical 1990. data. The final model based on fuzzy neurons is more global [10] A. Mitrovic, K. Hausler, “Porting SQL-Tutor to the Web”, and efficient, since it requires the least possible training effort, ITS'2000 workshop on Adaptive and Intelligent Web-based the number of the produced units is kept as small as possible, Education Systems, 2000, pp. 37-44. and it works in an environment of either crisp or fuzzy data. [11] J.G. Ozbolt, S. Shcultz, M.A. Swain, I.L. Abraham, “A According to our preliminary results there was no statistically proposed expert system for nursing practice. A springboard to significant difference between the users of the system (with or nursing science”, J. Med. Syst., vol. 9(1-2), pp. 57-68, 1985. without a tutor) and the students of the traditional “Biomed” [12] N. Saleem, B. Moses, “Expert systems as computer course in the school of nursing. We need a large number of assisted instruction systems for nursing education and training”, students to estimate the usefulness of the system. It will be soon Comput Nurs, vol. 121, pp. 35-45, 1994. available on Web. [13] M. Stern, and B. Woolf, “Curriculum sequencing in a The Web’s universality will enable many users to gain access Web-based tutor”, Fourth International Conference on to the system’s operations and consequently, its functionality Intelligent Tutoring Systems, Lecture Notes in Computer will be tested with numerous and diverse cases. Significant Science, vol. 1452, B.P. Goettl, H.M. Halff, C.L. Redfield, V.J. conclusions regarding the system's efficiency will thus be Shute, Eds. Berlin: Springer Verlag, 1998. drawn. [14] J. Vassileva, “Dynamic courseware generation”, Journal of Computing and Information Technology, vol. 5(2), pp. 87- Acknowledgements 102, 1997. [15] B. Woolf, “AI in education”, in Encyclopedia of Artificial This work was supported by the Research Committee of the Intelligence, S. Shapiro, Ed. New York: John Wiley & Sons, Technological Educational Institute of Patras, Greece, Program: 1992, pp. 434-444. EPEAEK II, project No: 2853/2003. [16] Jang JR. “ANFIS: Adaptive-Network-based Fuzzy References Inference System”, IEEE Trans on Systems. Man and Cybernetics vol. 23(3), 1993, 665-685. [1] J. Anderson, Rules of the mind, New Jersey: Lawrence [17] Keller J. M. and H. Tahani, “Implementation of conjuctive Erlbaum Associates, 1993. and disjunctive fuzzy logic rules with neural networks”, [2] J. Beck, M. Stern, and E. Haugsjaa, “Applications of AI in International Journal of Approximate Reasoning, 6(2), 1992, education”, ACM Crossroads, 1996. 221-240. [3] P. Brusilovsky, E. Schwarz, and G. Weber, “ELM-ART: an [18] Rutowska D. Neuro-fuzzy architectures and hybrid intelligent tutoring system on World Wide Web”, Third learning Physica-Verlag Heidelberg, New York, 2002. International Conference on Intelligent Tutoring Systems, [19] Machado RJ., Rocha AF., “Inference, Inquiry, Evidence Lecture Notes in Computer Science, vol. 1086, C. Frasson, G. Censorship and Explanation in Connectionist Expert Systems”, Gauthier and A. Lesgold, Eds. Berlin: Springer Verlag, 1996, IEEE Tranactions on Fuzzy Sytems, 5(3), 1996, 443-459. pp. 261-269. [4] P. Brusilovsky, A. Kobsa and J. Vassileva, Eds., Adaptive hypertext and hypermedia, Dordrecht: Kluwer Academic Publishers, 1998. [5] A.C. Hanson, S.M. Foster, B. Nasseh, K.E. Hodson, N. Dillard, “Design and development of an expert system for student use in a school of nursing”, Comput Nurs, vol. 12(1), pp. 29-34, 1994. [6] I. Hatzilygeroudis and C. Koutsojannis: “Fuzzy Neurules: Edging over Neurules”,3rd WSEAS Int. Conf. On Information Science and Applications (ISA 2003), pp49-50. [7] I. Hatzilygeroudis and J. Prentzas, “Constructing modular hybrid knowledge bases for expert systems”, International Journal on Artificial Intelligence Tools, World Scientific, vol. 10(1-2), pp. 87-105, 200.