Scholarly article on topic 'A Hybrid Attribute–based Recommender System for E–learning Material Recommendation'

A Hybrid Attribute–based Recommender System for E–learning Material Recommendation Academic research paper on "Materials engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
IERI Procedia
OECD Field of science
Keywords
{"Personalized Recommendation" / "Collaborative Filtering" / "Learning Material" / E–learning / "Adaptive Recommender"}

Abstract of research paper on Materials engineering, author of scientific article — Mojtaba Salehi, Isa Nakhai Kmalabadi

Abstract Recommendation system is a significant part of e–learning systems for personalization and recommendation of appropriate materials to the learner. However, in the existing recommendation algorithms, attributes of materials that can improve the quality of recommendation are not fully considered. For addressing this problem, a new material recommendation approach is proposed based on modeling of materials in a multidimensional space of material's attribute. Herein, each learner is modeled by a matrix that can take into account multi–attribute of materials. The recommender is adaptive to individual learner's preference as well as one's changing interest. Recommendation is generated by content–based filtering, collaborative filtering and some hybrid approaches. In attribute–based approach, the learner's real learning preference can be satisfied accurately according to the real–time up dated contextual information. The main contribution of this paper is modelling of learning material attributes in an effective recommendation framework

Academic research paper on topic "A Hybrid Attribute–based Recommender System for E–learning Material Recommendation"

Available online at www.sciencedirect.com

SciVerse ScienceDirect Procedía

IERI Procedía 2 (2012) 565 - 570

www.elsevier.com/locate/procedia

2012 International Conference on Future Computer Supported Education

A Hybrid Attribute-based Recommender System for E-learning

Material Recommendation

Mojtaba Salehia*9 Isa Nakhai Kmalabadia 9

_Faculty of Engineering, Tarbiat Modares University, 14115, Tehran, iran_

Abstract

Recommendation system is a significant part of e-learning systems for personalization and recommendation of appropriate materials to the learner. However, in the existing recommendation algorithms, attributes of materials that can improve the quality of recommendation are not fully considered. For addressing this problem, a new material recommendation approach is proposed based on modeling of materials in a multidimensional space of material's attribute. Herein, each learner is modeled by a matrix that can take into account multi-attribute of materials. The recommender is adaptive to individual learner's preference as well as one's changing interest. Recommendation is generated by content-based filtering, collaborative filtering and some hybrid approaches. In attribute-based approach, the learner's real learning preference can be satisfied accurately according to the real-time up dated contextual information. The main contribution of this paper is modelling of learning material attributes in an effective recommendation framework.

© 2012 Published by Elsevier B.V.

Selection and peer review under responsibility of Information Engineering Research Institute

Keywords: Personalized Recommendation, Collaborative Filtering, Learning Material, E-learning, Adaptive Recommender

1. Introduction

By increasing learning materials available on the e-learning systems, the delivery of appropriate learning material to learners is difficult using keyword searching method. Hence, locating the suitable learning materials has become a big challenge. One way to address this challenge is the use of recommender systems (Adomavicius and Tuzhilin, 2005). In addition, up to the very recent years, several researches have addressed

* Corresponding author. Tel.: +98-2182884323; fax: +98-2182884323, E-mail address: Mojtaba.salehi@modares.ac.ir.

ELSEVIER

2212-6678 © 2012 Published by Elsevier B.V. Selection and peer review under responsibility of Information Engineering Research Institute doi:10.1016/j.ieri.2012.06.135

the need for personalization in the e-learning environment. In fact, one of the new forms of personalization in e-learning environment is to give recommendations to learners in order to support and help them through the e-learning process (Khribi et al., 2009).

Recommender systems that have been deployed usually in e-commerce entities for expressing customer's interests use three strategies for recommendation including content-based (CB) (Khribi et al., 2009; Klasnja-Milicevic, et al., 2011), collaborative filtering (CF) (Liang et al., 2006; Romero et al., 2009; Luo, et al., 2010), and hybrid recommendation (Garcia, et al., 2011). One of the most important drawbacks of existing recommendation systems is that they usually use only rating matrix as useful information and not fully consider contextual information such as attribute of materials for improving recommendation. By implementing an attribute-based multidimensional approach for recommendation, this paper try to take into account multi-attribute of materials and dynamic preferences of learner simultaneously. Therefore, in this paper, to address the drawbacks of existing material recommendation algorithms and have a good recommendation results for e-learning material, a new material recommender system framework and relevant recommendation algorithms are proposed.

In order to reflect learner's complete spectrum of interests, Matrix-Preference (MP) is introduced to consider multi-dimensional attributes of materials. Truly, Matrix-Preference is built based on target learner's historical access records and multi-dimensional attributes of materials. In addition, an updating approach for Matrix-Preference is introduced to consider the dynamic preference of learner. Finally for improving of recommendation some hybrid approach is proposed. Experiments are being formulated to illustrate the system's capability. Learning material and learner profiling section describes modeling of leaner preferences. Then CB and CF are presented for recommendation. Recommendation section describes hybrid methods for recommendation. Experiment section evaluates the proposed approach and also describes the results. Finally, Conclusion section provides the concluding remarks.

2. Learning material and learner profiling

This research develops a recommendation algorithm that considers attributes of learning materials. The importance of specific attribute for each learner can be determined based rating of learner's accessed educational materials. According to the specified attributes, learning material profile is defined as a vector in which the values of attributes are assigned to a material. The attribute-based model is defined as a multidimensional vector I = (A1, A2,...,AK) where Ak indicates the k-th attribute's name. In addition, we can

consider a weight for each attribute to indicate its importance where AWk denotes the appropriate weight value and also ^ AWt =1. For example you may select subject, education level, price and author as attributes. In this situation we have: Booiii) =[(subject= Neurahetwork(Educatiokvel= PhD,,(Price=low),{AuthorA1)\-

r,l r, 2 .... riT

4 Sil2 SilT

A SiKl SiK 2 SiKT

Fig. 1. Personal preference matrix

According to learner ratings on materials, the attributes values of materials transfer to learner profile. For each learner, system makes a Personal preference matrix (shown in Fig. 1) including K rows corresponding to the attributes of category and T columns corresponding to T visited material. In this matrix Sjkj is the score of the attribute Ak by learner i in the observation of material,/' that is calculated as follows:

Where ri- is rating of learner i calculated by equation (3) for the material,/' that has been visited.

3. Content based recommendation

In this work, the similarity between learner behaviour and a special material, p is calculated using the following equation:

Z w'j-E(-mP ' L>j)

sim(L,, M„) = ----(2)

v 1' p' T K

In which Wy is a weighting value for observation of material y by learner i and is normalized with 1 -norm.

Since learner's recent accessed material preference plays an important role to the future interests. The relative

importance of each observation pre-determined as follows:

w = e) (3)

Where t(l. ) the order of material,/' in the recent observation by learner i and A is an adjustable parameter used to describe the change rate of learner's preference. This formula gives more weight to recent visited materials. m, (M L..) is a matching function between £-th attribute of material M and L.. that is calculated

* P5 P J

as follows:

m = jl if value (Ak, Mp) = value{ Ak, Li])

k p 'J [o otherwise Therefore, if value of £-th attribute for M and L,, be same m, (M ,L )gets 1 otherwise 0. Materials are

P y k \ p' ij s

ranked by calculating the similarity between the preference matrix made up of weighted behavior attribute sets and Materials. Highly ranked materials are then recommended to the learner.

4. Collaborative Filtering recommendation

For improving the accuracy and quality of recommendation, our research CF is implemented as follows: Step 1: For reflecting the similarity between the preferences of two learners, the similarity sim(Li, Lj) is

calculated as follows:

Z Z -E -SW mk (Ljk ' LH )

sim m, L,) = ---(5)

v 1 J y j* A

m L ^ = j1 ifvalue(Ak, Lih) = value(Ak ,LJf) [o otherwise

Step 2: In this step, system determines explicit attribute-based neighborhoods of learner a, N(La) according the calculated similarity in Step 1.

Step 3: The predication rating of material i by using implicit attribute based method for active learner is P(La,z')that is gained by the rating of La neighbourhood, N(La), that have rated i before. The computation formula is as the follows:

X sim (La, Lj) X {Rlj(i) - Rij )

P (La,») = Rl. + ^>--(7)

^ sum (La, Lj)

je «( )

Where RL¡t and RL¡ average rating of items rated by active learner La and Z . respectively.

5. Recommendation

We proposed the following recommendation approaches: Content Based Filtering (CBF), Collaborative Filtering (CF), CBF-CF, Most Frequently Visited material (MFV) and Most Similar Visited material to target learner (MSV).

Most Frequently Visited materials (MFV): MFV looks into N(L ) and for each neighbor, scans through the database and counts the visit frequency of the materials. Therefore, the score in this method is defined as

Scorep = X Nfv (A > P) (8)

Ls EiV (La)

Scores denotes the visit frequency of material^ by neighbours of La . Inspired from Symeonidis et al. (2008), this method assumes that the more a material is visited, the more popular it becomes.

Most Similar Visited materials to target learner (MSV): In MSV, according to sequential combination model

of hybrid recommender systems, those materials that visited by L. ( l. <= N(La)) and are the most similar to La (Sim(La,Mp)) will be considered, the score is defined as:

Scorep = Sim{ La, p) (9)

Most Similar Visited to the Most Similar Learners (MSV-MSL): since the similarities of the neighbors may vary significantly, for each material which visited by Lt (L. e N(La)), this method admeasures not just the similarity of material to Lt (Sim(L.,M ))= but the similarity of Lt and La (Sim(La,L¡)) as well. In this method, the scores is defined as

Scorep = Sim(L¡, p).Sim(La, L¡) (10)

MFV-MSL: This method, which combines MFV and MSL methods, scores materials according to their visit

frequency by learner Lt within N(La) ((¿.; p) )andthesimilarity of Lt and La (Sin(L ,i¡)), where

Scorep = NPV (Lt, p).Sim(La, Lt) (11)

6. Experiments

In this research, a real-world dataset, learning data records are applied in our experiments. The learning data records dataset come from the usage data of the course management system Moodle. MOODLE (Modular Object-Oriented Dynamic Learning Environment) is defined as a course management system (CMS), a free, Open Source software package designed using pedagogical principles, to help educators by creating effective online learning communities. The used dataset contains 16345 lending records from 676 learners on 3763 books where each record contains timestamp and rating information (as the ratio of certain lending time segment to maximum lending time segment). In addition, it contains books' type information and learners' basic information. For the evaluation of recommender system, the precision and recall have been used by various researchers (Pazzani and Billsus, 2007; Herlocker, 2000). Therefore, in this research we use these measures also. Since increasing the size of the recommendation set leads to an increase in recall but at the

same time a decrease in precision, we can use F1 measure (Shih and Liu, 2008). 6.1. Impact of number of recommended materials and X in CBF Component

In CBF component, we have two parameters that may affect the result of recommendation. To determine the sensitivity of parameter, we performed an experiment where we varied the number of recommendation and X and computed the F1 metric. Our results are shown in Fig. 2. It's observed that the number of recommendation and X affect the quality of CBF based recommendations. Best results is for 5 or 6 numbers of recommendation and X = 0.3 .

Fig. 2. Impact of number of recommended materials and X in CBF Component with respect of F1

6.2. Impact of neighborhood size in the CF component

An experiment is implemented to determine the sensitivity of neighborhood size in the CF component with respect of Fl. We performed where we changed the number of neighbors and computed the corresponding F1 metric. It was observed that the size of the neighborhood does affect the quality of top-N recommendations. According to results, we select 18as the optimal choice of neighborhood size.

6.3. Impact of recommendation generation method

To compare the relative performance of attribute based methods including CF, CBF, MFV, MSV, MSV-MSL and MFV-MSL methods in the recommendation generation, we performed an experiment also. As we can see in Fig. 3, the relative performance of these methods in different number of recommendations may change. But four methods including CF, CBF, MSV and MSV-MSL have better performance.

Fig. 3. Comparison of different recommendation approaches with respect of Fl

6.4. Qualitycomparison

Once we obtained the optimal values of different parameters of recommendation methods, we compared the recommendation quality of different attribute-based methods with the memory-based method CF that uses similarity between items for recommendation generation. Experimental results are summarized in Table 1 and show that attribute based methods have better performance and generates better recommendations. Table. 1. Comparison between attribute-based approach and Item based -CF

System Recall Precision F1

CF 0.342 0.674 0.454

CBF 0.357 0.698 0.472

MSV 0.394 0.723 0.510

MSV-MSL 0.401 0.754 0.524

Item based -CF 0.324 0.584 0.417

7. Conclusion

Material recommendation systems have been a significant part of e-learning systems for solving this problem. However, in the most of existing recommendation algorithms, attributes of learning material are neglected. This paper proposes a novel recommendation approach to take into account attributes of material for improving the quality of recommendations. The experiment results show that our algorithm can outperform traditional recommendation approaches significantly in precision, recall and F1 and also could be more suitable for e-learning environments. Based on the proposed algorithm, the learner's real learning preference can be satisfied accurately according to the real-time up dated contextual information.

References

[1]Adomavicius, G., Tuzhilin, A., Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions, IEEE Transactions on Knowledge and Data Engineering, 2005;17(6):734-749.

[2]Khribi, M.K., Jemni, M., Nasraoui, O., Automatic recommendations for e-learning personalization based on web usage mining techniques and information retrieval. Educational Technology and Society, 2009:12(4):30-42.

[3]Ghauth, K. I., Abdullah, N.A., Learning materials recommendation using good learners' ratings and content-based filtering, Educational Technology Research and Development, 2010;58(6):711-727.

[4]Kerkiri, T., Manitsaris, A., Mavridou, A., Reputation Metadata for Recommending Personalized e-Learning Resources, Second International Workshop on Semantic Media Adaptation and Personalization, 2007;110-115.

[5]Liu, F., Shih, B., Learning activity-based e-learning material recommendation system, In Proceedings of the ninth IEEE international symposium on multimedia workshops, 2007;343-348.

[6]Romero, C., Ventura, S., Zafra, A., Bra, P., Applying Web usage mining for personalizing hyperlinks in Web-based adaptive educational systems. Computers and Education, 2009;53:828-840.

[7]Luo, J., Dong, F., Cao, J., Song, A., A context-aware personalized resource recommendation for pervasive learning, Cluster Comput, 2010;13:213-239

[8]Klasnja-Milicevic, A., Vesin, B., Ivanovic, M., Budimac, Z., E-Learning personalization based on hybrid recommendation strategy and learning style identification, Computers and Education, 2011;56(3):885-899.

[9]Garcia, E., Romero, C., Ventura, S., Castroa, C.d., A collaborative educational association rule mining tool, The Internet andHigher Education, 2011;14(2):77-88.

[10]Symeonidis, P., Nanopoulos, A., Papadopoulos, A.N., Manolopoulos, Y. Collaborative recommender systems: Combining effectiveness and efficiency. Expert Systems with Applications, 2008:34(4):2996-3013.

[1 l]Pazzani, M., Billsus, D., Content-based recommendation systems. The adaptive web, 2007;325-341.

[12]Herlocker, J.L., Understanding and improving automated collaborative filtering systems. Citeseer. 2000.

[13]Shih, Y. Y., Liu, D. R. Product recommendation approaches: Collaborative filtering via customer lifetime value and customer demands. Expert Systems with Applications, 2008;35(l-2):350-360.