Building an E-learning Recommender System Using ... - CiteSeerX

6 downloads 0 Views 244KB Size Report
shows that most of the researchers use data mining approach and information ... profile builder will query the rating repository for the ratings. Both the document ...
2009 Ninth IEEE International Conference on Advanced Learning Technologies

Building an E-Learning Recommender System using Vector Space Model and Good Learners Average Rating Nor Aniza Abdullah University of Malaya [email protected]

Khairil Imran Bin Ghauth Multimedia University [email protected]

that can increase their performance. Furthermore, to know which learning materials that are highly rated by good learners in the real world is difficult. In this paper, we propose an e-learning recommender system with two types of recommendations: i) content-based recommendation and ii) recommendation based on good learners’ average ratings. The objective of the first recommendation type is to recommend additional learning resources that are similar to those of the viewing item (i.e. current learning context). The second recommendation aims to guide learners in selecting good learning resources in order to improve their learning process. Our proposed method also ensures that the current learning sequence is not interrupted and the recommended items always remain within the learning context. The remaining part of this paper is organized as follows. In Section 2, literature survey of the existing e-learning recommender system is presented. Section 3 introduces the overall system architecture and describes the proposed method which includes the recommender framework and the mathematical model used for recommending the items. Finally, Section 4 provides the concluding remarks.

Abstract An enormous amount of learning materials in elearning has led to the difficulty on locating suitable learning materials for a particular learning topic, creating the need for content recommendation tools within learning context. In this paper, we aim to address this need by proposing a novel framework for an e-learning recommender system. Our proposed framework works on the idea of recommending learning materials based on the similarity of content items (using Vector Space Model) and good learners’ average rating strategy. This paper presents the overall architecture of the proposed system and its potential implementation via a prototype design.

1. Introduction The popularity of e-learning has created huge amount of educational resources and to locate suitable learning references is a big challenge. In normal university e-learning environment, the learners are equipped with large amount of mandatory learning materials and extra references. The success of the implementation of recommender system in e-commerce has gained a lot of interest in e-learning [3][5][6]. Recommender system in e-learning different from other domain in such a way that the recommender system must recommend items without affecting the learning sequence and the recommended items must remain in the current learning context. According to social learning theory [8], people can learn by observing the behavior of others and outcome of those behavior. This theory strongly supports the proposed idea of learning by knowing the good learners average rating on particular items where all other learners know the outcome of the behavior (focus on highly rated items by good learners) of good learner

978-0-7695-3711-5/09 $25.00 © 2009 IEEE DOI 10.1109/ICALT.2009.161

2. Literature survey Recent trend on e-learning recommender system shows that most of the researchers use data mining approach and information retrieval technique as the recommendation strategies (reported in [1]-[6]). Table 1 summarizes the recommendation strategies proposed by current researchers on e-learning. To date, none of the e-learning system (reported in [1]-[7]) have attempted to recommend learning resources or learning items based on the rating of good learners. If this is implemented in an e-learning system, it can benefit learners in their learning process as it helps them to locate good quality of learning materials. The 194

recommendation system proposed in the following section aims to address this issue.

The viewing document is retrieved from the document repository. The content profile builder calculates the document weight that will be used later to calculate the item similarity. The rating repository stores all the good learners’ ratings. Finally the rating profile builder will query the rating repository for the ratings. Both the document weight and the ratings are then used by the recommendation engine to calculate the similarity of the items and the good learners’ average rating.

Table 1. Recommendation strategies, input, and output of the current research. Researchers [1], [2], and [4] [2], [3], [4], [5], and [6] [2] and [5] [5], [6], and [7]

Recommendation Strategies Data mining techniques Collaborative filtering Content-based filtering Clustering, Knowledge discovery, metadata, Item repository theory, Rule-based expert system, Artificial neural network

Input

Output

learner’s activities/ access history, learners rating, item attributes

related items/ documents, related links, learning activities, courseware module

3.1. Recommendation framework The recommendation process is depicted in Figure 2.

3. Proposed methodology Figure 1 shows the overall system architecture of the proposed e-learning recommender system. The instructor is responsible to create learning material (document) using the converting and authoring tool. The converting tool is used to convert the documents, such as presentation and word document into image file (or flash file). Converting such documents into image preserves the fidelity of the original documents and save time to re-write the document using other authoring tool. On the other hand, the authoring tool is used to embed the image into a predefined html template page, set the link between html pages, and to provide keywords for the html page. All the documents are stored in document repository. Learner is able to view and rate the viewing document.

Figure 2. The recommendation process flow.

The attributes table is used for similarity value calculation between the items and the result that are stored in the item-item matrix. The similarity calculation is calculated using vector space model as elaborated in Section 3.2. Top-N recommended items that exceed the similarity value threshold will be used to calculate the predicted rating. The rating matrix initially stores the predicted rating and the predicted rating will be replaced once the user provides the rating. The predicted rating is then stored in the rating log. The process stack includes a set of predefined instruction that will be executed during one of the following events: new item is inserted, existing item is deleted, item attribute is updated, item is rated, and item is rerated. The evaluation table which stores the good learners evaluation mark will query the rating matrix for good learners rating. The average ratings of good learners are then used for item recommendation.

3.2. Vector space model The vector space model involves two main phases as follows: i) calculate the weight and ii) calculate the

Figure 1. Overall system architecture of the elearning recommender system.

195

cosine similarity. The weight wi , j is calculated using

4. Conclusions

the term frequency/inverse document frequency (TDIDF) with normalized frequency as shown in equation (1). f i, j D (1) wi , j * log max z f z , j di

In this paper, we propose a recommender system for e-learning that is able to recommend similar items to the viewing item and also to recommend based on good learners’ average rating on the viewing items. Both recommendations are crucial to improve the learners learning process and the proposed system addresses both of them. The prototype is currently under testing with several set of presentation slides and word documents as learning materials. The performance analysis and benchmarking comparison are scheduled for the ongoing works

f i , j denotes the frequency a term i occurs in

where

document j. The maxz f z , j is the maximum frequency among all the keyword z that appears in document j. The D is the total number of documents that can be recommended to the learners. The d i is the number of documents that contains term i. The weight obtained from equation (1) is used to calculate the similarity value between the two items. The cosine similarity value is defined as follows. wc .ws (2) cos(wc , ws ) || wc || || ws || where

5. References [1] M.K. Khribi, M. Jemni, and O. Nasraoui, “Automatic Recommendations for E-Learning Personalization Based on Web Usage Mining Techniques and Information Retrieval”, Eight IEEE International Conference on Advanced Learning Technologies, Santander, Spain, July 2008, pp. 241-245.

wc and ws are treated as a vector of content

[2] N. Soonthornphisaj, E. Rojsattarat, and S. Yim-ngam, “Smart E-Learning Using Recommender System”, Computational Intelligence, Springer-Verlag Berlin Heidelberg, 2006, pp. 518-523.

based profile of user c and content of document s. Both

|| wc || and || ws || are the magnitude of the vector wc and

ws .

[3] F. Liu and B. Shih, “Learning Activity-Based E-Learning Material Recommendation System”, Proceedings of the Ninth IEEE International Symposium on Multimedia Workshops 2007, Beijing, China, Dec 2007, pp. 343-348.

3.3. Good learners recommendation strategies The average rating is used to calculate the good learners’ rating on a particular item and is defined as follows. N r i 1 i, j (3) Ri , j

[4] T. Y. Tang and G. McCalla, “Smart Recommendation for an Evolving E-Learning System: Architecture and Experiment”, International Journal on E-learning, vol. 4, no.1, pp. 105-129, 2003.

Nj

[5] T. Kerkiri, A. Manitsaris, and A. Mavridou, “Reputation Metadata for Recommending Personalized E-Learning Resources”, Proceedings of the Second International Workshop on Semantic Media Adaptation and Personalization, Uxbridge, Dec 2007, pp. 110-115.

where ri , j is the rating of good learner i on item j. The N j is the total number of good learners that rated item

j.

3.4. Good learners average rating prediction

[6] C. M. Chen, H. M. Lee, and Y. H. Chen, “Personalized ELearning System Using Item Repository Theory”, Computers & Education, vol. 44, no. 3, pp. 237-255, 2005.

As mentioned in Section 3.3, any item that has not received rating from the good learners will be recommended with a prediction rating, which is calculated as follows. N sim(d i , d n ) * Rn (4) Pi sim(d i , d n ) n 1 where sim (d i , d n ) is the similarity between item i and

[7] D. W. Tai, H. Wu, and P. Li, “Effective E-Learning Recommendation System Based on Self-Organizing Maps and Association Mining”, The Electronic Library, vol. 26, pp. 329-344, 2008. [8] A. Bandura, Social learning theory. Englewood Cliffs, NJ: Prentice-Hall, 1977.

item n and Rn is the good learners average rating on item n.

196

Suggest Documents