《电子商务 E-business》阅读文献：Smart E-Learning Using Recommender System

Smart E-Learning Using Recommender System Nuanwan Soonthornphisaj, Ekkawut Rojsattarat, and Sukanya Yim-ngar Department of Computer Science, Fact cience, Kasetsart University fscinws aku Department of Computer Science, Faculty of Resource and Environment, niversity Sri-racha( honburi. Thailand ekkawut@src. ku. ac th, sukanya@src. ku. ac th Abstract. We develop an e-learning web application that integrates the materi- als recommender system to facilitate the learners during the learning proces The system evaluates each learner via the quiz generator by randomly selecting a set of questions that are created by the instructor. Our smart e-learning system helps instructors to create and maintain both compulsory materials and ques- tions. We implemented the system at the faculty of Resource and Environment, Kasetsart University at Sri-racha campus and found that our system got a very good response from the instructors and learners. Furthermore, we propose the global e-learning framework using web service that has an ability to aggregate the recommended materials from other e-learning web sites and predicts more uitable materials to learners 1 Introduction Recommender systems use the opinions of a community of users to help individuals in that community more effectively identify content of interest from a potentially overwhelming set of choices [1. One of the most successful technologies for recom- mender systems, called collaborative filtering, has been developed and improved over the past decade to the point where a wide variety of algorithms exist for generat ing recommendations. We propose a smart e-learning web application that allows all learners to collaborate their expertise in order to predict the most suitable learning materials to each learner. Our system has two main features which are the quiz gen- erator and the material feeder. The quiz generator creates a set of questions from the warehouse with three difficulty levels(beginner, medium, expert). The material feeder module provides two sets of materials which are the compulsory set and the recommended set. Note that the compulsory materials are learning documents given by an instructor in each subject. Furthermore, we introduce the recommender system in the context of web service, nce we believe that the global e-learning system could bring more benefit from a larger set of co-operative learners which make more predictive power on the recom- mender system D-S Huang, K Li, and G w. Irwin(Eds ) ICIC 2006, LNAI 4114, pp. 518-523, 2006 O Springer-Verlag Berlin Heidelberg 2006

D.-S. Huang, K. Li, and G.W. Irwin (Eds.): ICIC 2006, LNAI 4114, pp. 518 – 523, 2006. © Springer-Verlag Berlin Heidelberg 2006 Smart E-Learning Using Recommender System Nuanwan Soonthornphisaj1, Ekkawut Rojsattarat2 , and Sukanya Yim-ngam2 1 Department of Computer Science, Faculty of Science, Kasetsart University, Bangkok, Thailand fscinws@ku.ac.th 2 Department of Computer Science, Faculty of Resource and Environment, Kasetsart University Sri-racha campus, Chonburi, Thailand ekkawut@src.ku.ac.th, sukanya@src.ku.ac.th Abstract. We develop an e-learning web application that integrates the materi￾als recommender system to facilitate the learners during the learning process. The system evaluates each learner via the quiz generator by randomly selecting a set of questions that are created by the instructor. Our smart e-learning system helps instructors to create and maintain both compulsory materials and ques￾tions. We implemented the system at the faculty of Resource and Environment, Kasetsart University at Sri-racha campus and found that our system got a very good response from the instructors and learners. Furthermore, we propose the global e-learning framework using web service that has an ability to aggregate the recommended materials from other e-learning web sites and predicts more suitable materials to learners. 1 Introduction Recommender systems use the opinions of a community of users to help individuals in that community more effectively identify content of interest from a potentially overwhelming set of choices [1]. One of the most successful technologies for recom￾mender systems, called collaborative filtering, has been developed and improved over the past decade to the point where a wide variety of algorithms exist for generat￾ing recommendations. We propose a smart e-learning web application that allows all learners to collaborate their expertise in order to predict the most suitable learning materials to each learner. Our system has two main features which are the quiz gen￾erator and the material feeder. The quiz generator creates a set of questions from the warehouse with three difficulty levels (beginner, medium, expert). The material feeder module provides two sets of materials which are the compulsory set and the recommended set. Note that the compulsory materials are learning documents given by an instructor in each subject. Furthermore, we introduce the recommender system in the context of web service, since we believe that the global e-learning system could bring more benefit from a larger set of co-operative learners which make more predictive power on the recom￾mender system

Smart E-Learning Using Recommender System 519 The following section presents our smart e-learning framework. Section 3 discusses recommender systems. The recommender web service is introduced in Section 4 Finally, Section 5 presents some concluding remarks 2 Smart E-Learning framework Web-based learning environment plays an important role for todays education. Cur- ently, learners have more options in studying compared to the previous time. Many e-learning applications have been developed such as Virtual-U [2] and Web-CT [3] In a virtual classroom, the instructor provides an online course which consists of the learning materials, self quiz and learning path. Typical learners study course material and do self quiz in order to evaluate their knowledge. The most important thing for the success learners is related to the quality of the learning materials which are not only depend on the given materials provided by the instructor but also depend on other learners recommendations. Our smart e-learning system applies the collabora- tive filtering approach that has an ability to predict the most suitable documents to the learner. All learners have the chance to introduce new material by uploadin the documents to the server or pointing out the web link from the Internet and rate the urrently available materials As shown in Figure 1, the instructors are responsible for the course materials preparation and creating a set of questions for the quiz. In order to create the question our system provides an easy to use tool for the instructor to create the multipl choices and solution. The instructors can indicate the level of difficulty for each ques- tion. The quiz generator's function is to randomly select a set of questions based on the difficulty level to generate the quiz for each lesson. The learning path for each learner depends on his/her quiz's score Fig 1. The smart e-learning framework 3 The Recommender System The recommender system is a software agent that gathers the rating information from all users in order to predict or recommend the most suitable materials to each user There are two traditional approaches for constructing recommender systems. The recommender systems were originally defined as ones in which"people provide

Smart E-Learning Using Recommender System 519 The following section presents our smart e-learning framework. Section 3 discusses recommender systems. The recommender web service is introduced in Section 4. Finally, Section 5 presents some concluding remarks. 2 Smart E-Learning Framework Web-based learning environment plays an important role for today’s education. Cur￾rently, learners have more options in studying compared to the previous time. Many e-learning applications have been developed such as Virtual-U [2] and Web-CT [3]. In a virtual classroom, the instructor provides an online course which consists of the learning materials, self quiz and learning path. Typical learners study course material and do self quiz in order to evaluate their knowledge. The most important thing for the success learners is related to the quality of the learning materials which are not only depend on the given materials provided by the instructor but also depend on other learners recommendations. Our smart e-learning system applies the collabora￾tive filtering approach that has an ability to predict the most suitable documents to the learner. All learners have the chance to introduce new material by uploading the documents to the server or pointing out the web link from the Internet and rate the currently available materials. As shown in Figure 1, the instructors are responsible for the course materials preparation and creating a set of questions for the quiz. In order to create the question, our system provides an easy to use tool for the instructor to create the multiple choices and solution. The instructors can indicate the level of difficulty for each ques￾tion. The quiz generator’s function is to randomly select a set of questions based on the difficulty level to generate the quiz for each lesson. The learning path for each learner depends on his/her quiz’s score. Fig. 1. The smart e-learning framework 3 The Recommender System The recommender system is a software agent that gathers the rating information from all users in order to predict or recommend the most suitable materials to each user. There are two traditional approaches for constructing recommender systems. The recommender systems were originally defined as ones in which “people provide

N. Soonthornphisaj, E. Rojsattarat, and S. Yim-ngam commendations as inputs, the system then aggregates and directs to appropriate recipients"[1]. Numerous recommender systems have been built for bo and applications. The algorithms used in these systems can be categorized into 2 categories which are collaborative filtering and content-based filtering. The concept of collaborative filtering is to use"word of mouth"process. It means that each user provides the system with the evaluations of items that may be used to build a profile of her likes or dislikes. The system aggregates all of the evaluations and creates top N lists of most popular items Most of the collaborative filtering systems trace their roots to Tapestry [4]. It is e collaborative filtering system in which a user takes a direct role in the process of deciding whose evaluations are used to provide his recommendations. The system operates the set of email and Usenet news group and allows the user to create rules or queries that indicate their preferences Another way to do collaborative filtering called automated collaborative filterin Its algorithm can automatically handle the process of user personalization. It means that the algorithm forms a set of individualized neighborhood for each user. The neighborhood consists of a subset of users whose opinions are highly correlated with the individual user. There are several systems such as GroupLens project [5, 6] MovieLens [7, 8, 9), Video Recommender [10], Audio CD recommender [111 Table 1. The algorithm for recommending materials in e-learning system Algorithm: Neighborhood-based Step 1: Weight all users with respect to the similarity of the active learner using Pearson correlation between their rating vector(see the following equation) ∑m(,-元)x:(n,- Where rails the rating given to material i by user a ra is the mean rating given by user a m is the total number of materials Step 2: Select n users that have the highest similarity to the active learner in order to form the neighborhood. Step 3: Compute a prediction using a weight combination obtained from the neighborhood using the following equation (,-)x Pai=ra Where Pai is the prediction for the active learner a for material pau is the similarity between the active learner a and u n is the number of learners in the neighborhd

520 N. Soonthornphisaj, E. Rojsattarat, and S. Yim-ngam recommendations as inputs, the system then aggregates and directs to appropriate recipients” [1]. Numerous recommender systems have been built for both research and applications. The algorithms used in these systems can be categorized into 2 categories which are collaborative filtering and content-based filtering. The concept of collaborative filtering is to use “word of mouth” process. It means that each user provides the system with the evaluations of items that may be used to build a profile of her likes or dislikes. The system aggregates all of the evaluations and creates top N lists of most popular items. Most of the collaborative filtering systems trace their roots to Tapestry [4]. It is an active collaborative filtering system in which a user takes a direct role in the process of deciding whose evaluations are used to provide his recommendations. The system operates the set of email and Usenet news group and allows the user to create rules or queries that indicate their preferences. Another way to do collaborative filtering called automated collaborative filtering. Its algorithm can automatically handle the process of user personalization. It means that the algorithm forms a set of individualized neighborhood for each user. The neighborhood consists of a subset of users whose opinions are highly correlated with the individual user. There are several systems such as GroupLens project [5, 6] MovieLens [7, 8, 9], Video Recommender [10], Audio CD recommender [11]. Table 1. The algorithm for recommending materials in e-learning system Algorithm: Neighborhood-based Step 1: Weight all users with respect to the similarity of the active learner using Pearson correlation between their rating vector (see the following equation) Where ra,i is the rating given to material i by user a ra is the mean rating given by user a m is the total number of materials Step 2: Select n users that have the highest similarity to the active learner in order to form the neighborhood. Step 3: Compute a prediction using a weight combination obtained from the neighborhood using the following equation Where pa,i is the prediction for the active learner a for material i p a,u is the similarity between the active learner a and u n is the number of learners in the neighborhood ( ) ( ) ∑ ( ) ∑ ( ) ∑ − × − − × − = = = = 2 1 2 1 1 u,i u m a,i a i m i a,i a u,i u m i a,u r r r r r r r r P ( ) a,u n u u,i u a,u n u a,i a P r r P p r ∑ ∑ = = − × = + 1 1

Smart E-Learning Using Recommender System 521 to The problem space of the collaborative filtering can be defined as a matrix of users rsus items. Each cell represents user's rating on a specific item [12]. The objective of the algorithm is to predict the value of the empty cell. We employ an algorithm called neighborhood-based [8 to do material prediction for each learner. The concept of the algorithm is to select a subset of learner based on their similarity to the active learner. Then do the weight aggregation using all ratings in order to generate the prediction to the active learner. The detail of neighborhood- based algorithm can be found in Table 1 ⊙:困psm创,回·□器 Go earnin Material I My maid I Mr RaledI My WrhLiat ySearch materials titie v Recommendation material > Ranae sm1m:留Jms8:ml山n:前ml弄wmn;fm Predict lor you Not Readv sn: m iunanwlda lne: kudt lupin: R 5合合合)mem moola I Ilau: kuda t Not Read y Sun: infills lnu: 1 Iussi I P Search material Tite 食N2m如m体,出m口 食☆ 2 stors v Su nu: a whesuNooe sn lusunn System Prediction or individual learner Fig. 2. The smart e-learning Web application 4 Global E-Learning Using Web service In order to make the recommender system more practical, we implement the recom- mender system web service in order to extend the e-learning system from local learners to global learners. As shown in Figure 3, the web service is responsible for cooperate all learners' recommendations from different e-learning websites. First, each e-learning website administrator must register to be the member of the recommender system web

Smart E-Learning Using Recommender System 521 The problem space of the collaborative filtering can be defined as a matrix of users versus items. Each cell represents user’s rating on a specific item. [12]. The objective of the algorithm is to predict the value of the empty cell. We employ an algorithm called neighborhood-based [8] to do material prediction for each learner. The concept of the algorithm is to select a subset of learner based on their similarity to the active learner. Then do the weight aggregation using all ratings in order to generate the prediction to the active learner. The detail of neighborhood￾based algorithm can be found in Table 1. Fig. 2. The smart e-learning Web application 4 Global E-Learning Using Web Service In order to make the recommender system more practical, we implement the recom￾mender system web service in order to extend the e-learning system from local learners to global learners. As shown in Figure 3, the web service is responsible for cooperate all learners’ recommendations from different e-learning websites. First, each e-learning website administrator must register to be the member of the recommender system web System Prediction for individual learner

N. Soonthornphisaj, E. Rojsattarat, and S. Yim-ngam service. The web service maintains the database of materials in order to do the collabo- rative filtering process. The advantage of using web service is to provide more chance to each learner to get better quality of materials. The web service will facilitate all e- learning web sites to get the recommended materials for their learners. These web-sites don' t need to implement the collaborative filtering algorithm and maintain any of their learner's ratings. Therefore, it is very convenient for those web-sites The global e-learning web service provides a wide variety of items F和m such as the text book recommendation We investigate the web service nce by connectingoursystemtotheamazonwebsite(www.amazon.com the book recommendation for our learners The original concept of web service is inspired by the remote procedure call func- tion in programming language. The web service's role is to serve each e-learning web site via the Xml(ExtensIble Markup Language)using Http protocol Recommender system Web service E-Learning Website 1 E-Learning Website 2 E-Learning Website 3 Fig 3. The recommender system web service architechture 5 Discussion and Future work The global smart e-learning system is a new vision for the e-learning. Learners will be benefited from this system. They have more chance to obtain better quality of materi- als and a wide variety of recommended items. The database of material maintained by the Web service can be used for data mining project that will enhance the perform- ance of all e-learning web site members. We plan to extend our work to the data min- ing discipline in the near future

522 N. Soonthornphisaj, E. Rojsattarat, and S. Yim-ngam service. The web service maintains the database of materials in order to do the collabo￾rative filtering process. The advantage of using web service is to provide more chance to each learner to get better quality of materials. The web service will facilitate all e￾learning web sites to get the recommended materials for their learners. These web-sites don’t need to implement the collaborative filtering algorithm and maintain any of their learner’s ratings. Therefore, it is very convenient for those web-sites. The global e-learning web service provides a wide variety of items to learners such as the text book recommendation. We investigate the web service performance by connecting our system to the AMAZON web site (www.amazon.com) in order to get the book recommendation for our learners. The original concept of web service is inspired by the remote procedure call func￾tion in programming language. The web service’s role is to serve each e-learning web site via the XML (Extensible Markup Language) using HTTP protocol. Fig. 3. The recommender system web service architechture 5 Discussion and Future Work The global smart e-learning system is a new vision for the e-learning. Learners will be benefited from this system. They have more chance to obtain better quality of materi￾als and a wide variety of recommended items. The database of material maintained by the Web service can be used for data mining project that will enhance the perform￾ance of all e-learning web site members. We plan to extend our work to the data min￾ing discipline in the near future

Smart E-Learning Using Recommender System 523 Acknowledgment This work is supported by Kasetsart University Research and Development Institute (KURDD References 1. Resnick, P, Varian, H.R.: Recommender Systems. Communication of ACM 40, (1997) 2. Groeneboer, T -CC, Stockley, D: Virtual-u: A Collaborative Uodel for Online L Environments. Proceeding of the Second International Conference on Computer for Collaborative Learning, Toronto, Canada, December(1997) 3.WebctavailableonAugust2002.http://www.webct.com/ 4. Goldberg, D, Nichols, D, Oki, B M, Terry, D. Using Collaborative Filtering to Weave an Information Tapestry. Communications of the ACM 35(12):(1992)61-70 5. Resnick, P, lacovo, N. Suchak, M, Bergstrom, P, Riedl, J. GroupLens: An Open Ar chitecture for Collaborative Filtering of Netnews Proceeding of ACm cscw'94 Confer- ence on Computer Supported Cooperative work. (1994)175-186 6. Konstan, J.A., Miller, B N, Malz, D, Herlocker, J L, Gordon, L.R., Riedl, J GroupLens Applying Collaborative Filtering to Usenet News Communications of the ACM(1997) 7. Dahlen, B., konstan, J.A., Herlocker, J L, Good, N, Borchers, A, Riedl, J. Jump Starting MovieLens: User Benefits of Starting a Collaborative Filtering System with Dead Data, University of Minesota, (1998) 8. Herlocker, J L, Konstan J.A. Borchers, A Riedl J. An Algorithmic Framework for Per forming Collaborative Filtering. Proceedings of the 22nd Annual International Conference on Research and Development in Information Retrieval (1999)230-237 9. Sarwar, B M., Konstan, J.A., Borchers, A, Herlocker J, Miller B. Redl, J. Using Filter- ing Agents to Improve Prediction Quality in the GroupLens Research Collaborative Filter- ing System. Proceeding of the ACM Conference on Computer Supported Cooperative work.(198)345-354 10. Hill. w.. Stead. L. Rosenstein M.. Furnas. G. Recommending and Evaluation Choices in Virtual Community of Use, pp. 194-201. Proceeding of the Conference on Human Fac- tors in computing Systems. (1995)194-201 11. Shardanand, U, Maes, P. Social Information Filtering: Algorithms for Automating Word Mouth, pp 210-217. Proceeding of the 1995 Conference on Human Factors in Comput ing Systems. (1995)210-217 12. Breese, J, Heckerman, D, Kadie, C: Empirical Analysis of Predictive Algorithms for Collaborative Filtering. Proceeding of 14 Conference on uncertainty in Artificial Intelli- gence.(1998)

Smart E-Learning Using Recommender System 523 Acknowledgment This work is supported by Kasetsart University Research and Development Institute (KURDI). References 1. Resnick, P., Varian, H. R.: Recommender Systems. Communication of ACM 40, (1997) 56–58 2. Groeneboer,T.-C.C., Stockley, D.: Virtual-u: A Collaborative Uodel for Online Learning Environments. Proceeding of the Second International Conference on Computer Support for Collaborative Learning, Toronto, Canada, December (1997) 3. Webct, available on August 2002. http://www.webct.com/ 4. Goldberg, D., Nichols, D., Oki, B.M., Terry, D.: Using Collaborative Filtering to Weave an Information Tapestry. Communications of the ACM 35(12): (1992) 61-70 5. Resnick, P., Iacovou, N. Suchak, M., Bergstrom, P., Riedl, J.: GroupLens: An Open Ar￾chitecture for Collaborative Filtering of Netnews, Proceeding of ACM CSCW’ 94 Confer￾ence on Computer Supported Cooperative work.(1994) 175-186 6. Konstan, J.A., Miller, B.N., Malz, D., Herlocker, J.L., Gordon, L.R., Riedl, J. GroupLens: Applying Collaborative Filtering to Usenet News. Communications of the ACM (1997) 7. Dahlen, B.J., konstan, J.A., Herlocker, J.L., Good, N., Borchers, A., Riedl, J.: Jump￾Starting MovieLens: User Benefits of Starting a Collaborative Filtering System with “Dead Data”, University of Minesota, (1998) 8. Herlocker, J.L., Konstan J.A. Borchers, A., Riedl J.: An Algorithmic Framework for Per￾forming Collaborative Filtering. Proceedings of the 22nd Annual International Conference on Research and Development in Information Retrieval. (1999) 230-237 9. Sarwar, B.M., Konstan, J.A., Borchers, A., Herlocker J., Miller B., Redl, J.: Using Filter￾ing Agents to Improve Prediction Quality in the GroupLens Research Collaborative Filter￾ing System. Proceeding of the ACM Conference on Computer Supported Cooperative Work. (1998) 345-354 10. Hill, W., Stead, L., Rosenstein, M., Furnas, G.: Recommending and Evaluation Choices in a Virtual Community of Use, pp. 194-201. Proceeding of the Conference on Human Fac￾tors in computing Systems. (1995) 194-201 11. Shardanand, U., Maes, P.: Social Information Filtering: Algorithms for Automating Word of Mouth, pp. 210-217. Proceeding of the 1995 Conference on Human Factors in Comput￾ing Systems. (1995) 210-217 12. Breese, J., Heckerman, D., Kadie, C.: Empirical Analysis of Predictive Algorithms for Collaborative Filtering. Proceeding of 14th Conference on uncertainty in Artificial Intelli￾gence. (1998)