Please download to get full document.

View again

of 9

Using Rules Discovery for the Continuous Improvement of e-Learning Courses

Using Rules Discovery for the Continuous Improvement of e-Learning Courses
6 views9 pages
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Documenttranscript
  E. Corchado et al. (Eds.): IDEAL 2006, LNCS 4224, pp. 887   –   895, 2006. © Springer-Verlag Berlin Heidelberg 2006 Using Rules Discovery for the Continuous Improvement of e-Learning Courses Enrique García, Cristóbal Romero, Sebastián Ventura, and Carlos de Castro Escuela Politécnica Superior. Universidad de Córdoba 14071 Córdoba, Spain {egsalcines, cromero, sventura, cdecastro}@uco.es Abstract.  This paper presents a cyclical methodology for the continuous improvement of e-learning courses using data mining techniques applied to education. For this purpose, a specific data mining tool has been developed, which discovers relevant relationships between data about how students use a course. Unlike others data mining approaches applied to education, which focus on the student, this method is aimed professors and how to help them improve the structure and contents of an e-learning course by making recommendations. We also use a rule discovery algorithm without parameters in order to be easily used by non-expert users in data mining. The results of experimental tests performed on an online course are also presented, demonstrating the usefulness of the proposed methodology and algorithm. Keywords:  association rule, e-learning, authoring tool. 1 Introduction The huge increase in Internet access means that online education or e-learning is now a reality. Increasingly more private and public teaching institutions provide their students with web-based learning management systems (LMS). WebCT, Virtual-U and TopClass are examples of commercial LMS, although freely distributed learning management systems, such as Moodle, ATutor, ILIAS [1], and educational adaptive hypermedia courses as Interbook, ELM-ART and AHA [2] are also gaining importance. These systems accumulate a vast amount of information which is very valuable in analyzing students’ behavior and to assist authors in detecting possible errors, shortcomings and improvements. However, due to the vast quantities of data these systems can generate daily, it is very difficult to manage manually, and authors demand tools which assist them in this task, preferably on a continuous basis. In order to solve this problem, some specific educational data mining tools have been developed to help educators in analyzing different aspect of the learning process: personalization of learning systems [3]; recommendation systems [4] that classify students and contents in order to recommend optimum resources and routes; and systems that detect irregularities [5], discovering irregular browsing patterns. These systems can be classified according to the field of application or focus [6]: 1) aimed at students [7], to suggest good learning experiences depending on the students’ preferences, needs and level of knowledge; and 2) aimed at professors [8] , so they can know more about the students that learn on the net, assess students according to their  888 E. García et al. browsing patterns, classify students into groups or restructure website contents in order to customize the course. This paper looks at the use of data mining techniques applied to e-learning but from the point of view of the course professor or creator. The main aim of the proposed system is to detect possible problems in the course structure or contents, based on information about how students use the course. This is an increasingly important area of research, which enables the enormous amount of information generated when students interact with the system to be put to efficient use. It also introduces a feedback stage so that the course designed can be continuously improved. 2 Methodology for the Improvement of e-Learning Courses We propose to use a continuous improvement of e-learning courses methodology to detect possible problems in the design and contents of e-learning courses [9]. This cyclical methodology includes a feedback or maintenance stage based on how the students use the course, and consists of the following stages: •   Course creation.  This is the first stage, when the course is created. The professor usually creates the adaptive course, providing all the contents and structural information required. A generic or specific authoring tool [10] is normally used to make this task easier. At the end of this stage, the course should be uploaded onto a web server so that students can use it remotely. •   Course completion.  In order to complete the course, students must use a web browser to connect to the Web Server where the course is stored. As the students are completing the course, usage data are collected and stored on the server, depending on the data model used, as well as the different log files. •   Course improvement. The data generated as students complete the course is used as input. The methodology applies a data mining algorithm to this data to detect any possible problems. The results of this process are displayed to the professor in the form of recommendations about how to improve the course structure or contents. Our aim, therefore, is to discover relevant information from a teaching point of view and about the effectiveness of the course in the form of rules based on the data stored about all the students who complete the course. The sub-sections below describe each of the modules that make up the Course Improvement stage, which is the nucleus of the proposed methodology (see Figure 1). The data mining modules used in the course improvement are: •   Data mining module without parameters.  This module aims to find association rules about a specific data set once the data have been pre-processed and converted into a single summary table that guarantees the quickest possible management of this information. The output of this module is then analyzed by the subjective analysis module. A comparative study between the main algorithms that are currently used to discover association rules can be found in [11]: APriori [12], FP-Growth [13], MagnumOpus [14], Closet [15]. The most widely used algorithm is Apriori, to which many improvements have been made. The Apriori algorithm uses two parameters, the minimum support and the minimum confidence, to find all the   Using Rules Discovery for the Continuous Improvement of e-Learning Courses 889 rules that exceed the thresholds specified by the user. However, the user must possess a certain amount of expertise in order to find the right balance between support and confidence that gives the best N rules. Weka [16] package implements an Apriori-type algorithm that solves this problem partially. This algorithm, reduces iteratively the minimum support, by the factor delta support ( Δ s) introduced by user, until minimum support is reached or required number of rules (NR) has been generated. Another improved version of the Apriori algorithm is the Predictive Apriori algorithm [17], which automatically resolves the problem of balance between these two parameters, maximizing the probability of making an accurate prediction for the data set. In order to achieve this, a parameter called the exact   expected predictive accuracy (acc) is defined and calculated using the Bayesian method, which provides information about the accuracy of the rule found. Fig. 1.  Methodology for the continuous improvement of e-learning courses •   Subjective analysis module.  The Predictive Apriori algorithm finds the best N rules without the intervention of the user. However, this method does not guarantee that the rules obtained will be relevant or useful to the professor to detect problems in the e-learning course. Therefore, they need to be assessed in order to find the most relevant ones. For this purpose, objective relevance measurements can be used such as the support and confidence parameters, as well as purely statistical measurements [18] such as Chi-Squared, correlation coefficients, profit or entropy functions, to measure the dependency inference between data variables. However, the use of subjective measurements is becoming increasingly important, measurements based on subjective factors guided by the users. In most approaches to finding relevant rules subjectively, the user has to express, in accordance with his/her previous knowledge, which rules he/she finds relevant. In [19] it is  890 E. García et al. describes a system that compares the rules discovered with the user’s knowledge about the field in question. Using their own specification language, the users indicate their knowledge base about a certain subject, through relationships between fields or items in the database. Our analysis module applies this algorithm, adapting it to our data format and types of rules in order to classify them as expected rules, if they coincide with the knowledge base we have about the domain, or unexpected rules if they do not. The knowledge base is a rules repository of contents that is made up of rules discovered by other LMS users in previous experiences or courses, as well as rules proposed by experts in this area. Expected rules are used as a basis for recommendations to improve the course and unexpected rules should be analyzed by the professor to determine if they are relevant, in which case they could be included in the repository. •   Recommendations module.  The output of the previous module becomes the input for the recommendations module [20], which is made up of two fundamental blocks: -   Recommendations block.  In this block, the rules discovered are displayed to the professor in two different formats, depending on the type of rule found. If the rule is expected, the problem detected is displayed along with the recommended action to resolve it. If the rule is unexpected, it is also shown to the professor, who should then determine whether it is relevant, in which case it could be included in the rules repository to be taken into account in future analyses; if it is not relevant, it can be discarded. There are two types of recommendation: active and passive. The active recommendation implies a direct modification of the course content or structure and they can be linked to: modifications in the formulation of the questions or the practical tasks/exercises assigned to the students; changes in previously assigned parameters such as course duration or the level of difficulty of a lesson; the elimination of resources such as forums or chat rooms, etc. The passive   recommendation   detects a more general problem and the professor is advised to refer to other more specific recommendations. An example is: IF U_FINAL_SCORE(  N  ) = LOW THEN C_SCORE = HIGH, which detects a problem in unit n  and advises the professor to check with other problems related to this unit. -   Rules repository block.  This is the knowledge base from which the recommendations are made. The success of the modifications made to the course depends on the content and structure of this module. The repository can initially be empty, if the professor has not yet discovered any rules, or it can contain an initial set of rules, which the user considers to be reasonably precise knowledge [19] about the domain. In addition to the rule itself, two fundamental fields are included: the problem detected by the proposed rule and a possible recommendation for its solution. Each time a rule is included in the repository, additional identification data are also included, such as author, date and the type of course where said rule was discovered. Based on teaching and our experience of e-learning courses, we have proposed an initial set of rules and their respective recommendations to be included in the repository.   Using Rules Discovery for the Continuous Improvement of e-Learning Courses 891 3 Implementation In order to implement the proposed architecture and to make it easier for the course professor or author to perform the data mining process, a tool called CIECoF (C ontinuous improvement of e-learning courses framework  ) has been developed in the Java programming language (see Figure 2). The main feature is its specialization in education, using specific attributes, filters and restrictions for course usage data; hence it is better suited for use in educational contexts than general purpose tools. To make the rules discovered more comprehensible and to reduce significantly the run time of the search algorithm, these attributes must be discretized. The transformation to discreet variables can be seen as a categorization of the attributes that takes a small set of values. The basic idea involves partitioning the values of the continuous attributes within a small list of intervals. Our discretization process used three possible nominal values: LOW, MEDIUM and HIGH and three discreet transformation methods were implemented: equal width method, equal frequency method, manual method [21], where the user sets the limits of the categories manually. In the case of discretization of times, an additional option is included to eliminate noisy values that exceed a specific threshold in order to avoid erroneous data, for example if a concept or exercise remains on the screen for a long time owing to the fact that the student left without finishing that section. Once the application’s parameters have been configured or using the default values, the professor must select specific data and attributes in order to restrict the search domain. Another panel displays the results obtained in a table with the Fig. 2.  CIECoF interface
Advertisement
MostRelated
View more
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x