forrja: Effective Trust-aware E-learning Recommender System based on Learning

 Effective Trust-aware E-learning Recommender System based on Learning

Introduction 

Due to unprecedented proliferation of information and communication technologies in recent years, e-learning has become more and more popular in academics as well as in commercial environments (Zaiane, 2002). 

E-learning provides opportunities for learners not only to study courses or to learn professional knowledge without time and space constraints, but also to train themselves at their own pace through the asynchronous and synchronous learning network models (Chao and Chen, 2009). 

Due to enormous amount of learning resources in e-learning environment, learners face difficulties in searching appropriate resources according to their need (Ghauth & Abdullah, 2010). 

In this situation, recommender system (RS) seems a proficient solution for dealing with this resource overload in elearning environment (Khribi et al., 2009). 

Styles and Knowledge Levels

1- ABSTRACT 

In the age of information explosion, e-learning recommender systems (ELRSs) have emerged as the most essential tool to deliver personalized learning resources to learners. Due to enormous amount of information on the web, learner faces problem in searching right information. 

ELRSs deal with the problem of information overload effectively and provide recommendations by taking into consideration the learners’ preferences such as learning styles, goals, knowledge levels, learning paths etc. 

In this paper, we propose a weighted hybrid scheme to recommend right learning resources to a learner by incorporating both the learners’ learning styles (LSs) and the knowledge levels (KLs). Further, by elicitation of trust values among learners, we develop a scheme such that for a given active learner, the .trustworthy learners 

2- Having greater knowledge and 

Similar learning style patterns as that of the active learner have greater weightage in recommendation strategy. Experimental results are presented to demonstrate the effectiveness of the proposed scheme

3- Recommender systems 

(RSs) are one of the most promising technologies of web personalization to alleviate the problem of information and product overload. They provide personal, affordable and effective recommendations to users based on their preferences expressed, either explicitly or implicitly (Adomavicius & Tuzhilin, 2005; Al-shamri & Bharadwaj, 2008; Milicevic et al., 2010). 

E-learning recommender systems (ELRSs) deal with information about learners and their learning activities and recommend items such as articles, web pages, etc. (Nghe et al., 2010). Collaborative filtering provides recommendations to learner based on those learners who have similar preferences. 

4- Since CF is able to capture the particular preferences 

Of a user so it has become most widely accepted technique in RSs for recommending web pages, music, books etc(Symeonidis et al., 2008). It has also been successfully employed in ELRSs (Manouselis et al., 2010; Bobadilla et al., 2009; Dwivedi & Bharadwaj, 2011). 

RS is strongly context/domain dependent, so it is not feasible that recommendation strategy for one context/domain is transferable to others (Drachsler et al., 2007). The reason why the thriving application of movie or .

Joke recommendation strategies has not had such an efficacy in e-learning because modeling accurate learner profile is a much harder task than in other application domains. Two important open research issues in ELRSs are as follows: ˙ 

Learner’s point of view : Recommended resources should be interesting to learners, according to their needs as well as their characteristics. ˙ 

Designer’s point of view : How to design learning materials considering learners’ preferences and how to recommend these resources in a specified sequence so that learners’ performance can be enhanced.

5- We designed our proposed 

ELRSs based on learner’s point of view by taking into consideration learner’s characteristics namely learning styles and knowledge levels etc. In e-learning, learners are characterized on the basis of their learning styles, emotions, knowledge levels and goals etc. (Drachsler et al., 2007). 

Learning style of a learner can be considered as a valuable factor for enhancing the individual learning that would affect the recommendation task. Learning style (LS) indicates how a learner learns and likes to learn. 

It can be analyzed or collected from the learning behavior of learner during study (Chang et al., 2009; Garcia et al., 2008). Bobadilla et al. (2009) suggested that learners with greater knowledge should have greater weight in the computation of recommendation than the learners with less knowledge among all neighbors of an active learner in collaborative filtering framework. 

6- Therefore knowledge level of learners is an

important factor in addition to LSs. Therefore, we are providing a hybrid ELRS which offers resource recommendations by acclimatizing automatically learners’ learning styles and knowledge levels that would favor and improve the learning. 

The following assumptions that motivated us for the adaptation of learning style and knowledge level in ELRS are: 

• Learners with different learning style generate different perspectives on effective strategies for dynamic group interactivity (Kolb, 1976). As a consequence, learners can be grouped on the basis of learning styles to have an impact on recommendation task in our work.

 • Researchers believe that learning style is a good predictor of an individual’s preferred learning behavior (Bostrom et al., 1990). 

• Milicevic et al. (2010) recognize the different patterns of LSs in PROTUS system which provides effective personalized recommendation of learning contents after processing the clusters based on different learning styles and mining frequent patterns for the habits and interest of learners. 

As a consequence, we generate effective clusters of learners’ LSs by utilizing the GA K-means algorithm in our work and develop a collaborative framework (CF-LS) based on these clusters. 

• Paechter et al. (2010) suggest that only few variables like exchange of knowledge with peer learners contribute to perceived learning achievements in a course. 

• Bobadilla et al. (2009) have also showed that the incorporation of knowledge of other learners provided the better recommendations. B

7- Besides LSs and KLs, 

we extend our hybrid system (CF-LS-KL) through the incorporation of trustworthiness of learners in recommendation task. The reason behind the incorporation of trustworthiness in recommendation task is that some similar learners may be malicious for the recommendations in collaborative filtering environment and the recommendation provided by them cannot be effective. 

The work presented in this paper, regarding to above aspects, is an effort towards developing a trust-aware ELRS utilizing both LSs and KLs of learners. The main contributions of this paper are three fold: 

• First of all, a collaborative filtering using LSs (CF-LS) is designed utilizing clusters of different learning styles generated through Genetic K-means algorithm.

 • Second, a collaborative filtering scheme based on KLs (CF-KL) is developed. Thereafter, a weighted hybrid scheme (CF-LS-KL) is presented to take possible advantages of learner preferences. 

• Finally, in order to give weightage to trustworthiness of learners, a trust-aware weighted hybrid scheme (TRCFLS-KL) is proposed.

Conclusion

The rest of the paper is organized as follows: We first give a brief summary of related work and literature survey on LSs, KLs and clustering technique in Section 2. Section 3 elaborates collaborative filtering framework utilizing LSs and 

KLs in trust aware e-learning recommendation environment. Section 4 describes the experimental setup, evaluation metrics and results of the evaluation followed by the discussion in Section 5. Finally, Section 6 provides concluding remarks and suggests some future research directions.