Scholarly article on topic 'MOOC-Rec: A Case Based Recommender System for MOOCs'

MOOC-Rec: A Case Based Recommender System for MOOCs Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Keywords
{"Technology enhanced learning" / TEL / MOOC / "Recommendation system" / CBR}

Abstract of research paper on Computer and information sciences, author of scientific article — Fatiha Bousbahi, Henda Chorfi

Abstract Learning processes and learning potentials are continuously investigated. Exploration of the possibilities of Technology Enhanced Learning (TEL) led to the development of many solutions and recently to Massive Open Online Courses (MOOCs). MOOCs are probably the most important “novelty” in the field of e-learning of the last years. MOOCs are capable of providing several ten thousands of learners with access to courses over the web. MOOCs have recently gained much attention especially in leading universities and are now often considered as a highly promising form of teaching. More and more universities are currently working to offer their courses in the form of MOOC providing learners with a wide variety of choices. With MOOCs proliferation, learners will be exposed to various challenges and the traditional problem in TEL “finding the best learning resources” is more than ever up to date. Since information retrieval and searching for the appropriate learning resources is an essential activity in TEL, the development of recommender systems for learning has seen increased attention. Recommender systems permit to respond to the traditional problem. In the present paper, we address this major problem – the difficulty for learners to find courses which best fit their personal interests. We propose a system that recommends appropriate MOOCs in response to a specific request of the learner. Using the Case Based Reasoning (CBR) approach and a special retrieval information technique, the system proposes to the learners the most appropriate MOOCs (from different providers) fitting her/his request based on learner profile, needs and knowledge.

Academic research paper on topic "MOOC-Rec: A Case Based Recommender System for MOOCs"

Available online at www.sciencedirect.com

ScienceDirect

Procedia - Social and Behavioral Sciences 195 (2015) 1813 - 1822

World Conference on Technology, Innovation and Entrepreneurship

MOOC-Rec: A Case Based Recommender System for MOOCs

Fatiha Bousbahia*, Henda Chorfia

a ' Information Technology Department, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia

Abstract

Learning processes and learning potentials are continuously investigated. Exploration of the possibilities of Technology Enhanced Learning (TEL) led to the development of many solutions and recently to Massive Open Online Courses (MOOCs). MOOCs are probably the most important "novelty" in the field of e-learning of the last years. MOOCs are capable of providing several ten thousands of learners with access to courses over the web. MOOCs have recently gained much attention especially in leading universities and are now often considered as a highly promising form of teaching. More and more universities are currently working to offer their courses in the form of MOOC providing learners with a wide variety of choices. With MOOCs proliferation, learners will be exposed to various challenges and the traditional problem in TEL "finding the best learning resources" is more than ever up to date. Since information retrieval and searching for the appropriate learning resources is an essential activity in TEL, the development of recommender systems for learning has seen increased attention. Recommender systems permit to respond to the traditional problem. In the present paper, we address this major problem - the difficulty for learners to find courses which best fit their personal interests. We propose a system that recommends appropriate MOOCs in response to a specific request of the learner. Using the Case Based Reasoning (CBR) approach and a special retrieval information technique, the system proposes to the learners the most appropriate MOOCs (from different providers) fitting her/his request based on learner profile, needs and knowledge.

© 2015TheAuthors.PublishedbyElsevierLtd.This is an open access article under the CC BY-NC-ND license

(http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of Istanbul Univeristy.

Keywords: Technology enhanced learning; TEL; MOOC; Recommendation system; CBR

* Corresponding Fatiha Bousbahi. Tel.: +0-966-565795014. E-mail address: fbousbahi@ksu.edu.sa

1877-0428 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of Istanbul Univeristy.

doi: 10.1016/j.sbspro.2015.06.395

1. Introduction

The Technology Enhanced Learning (TEL) is an interdisciplinary research field to which several disciplines contribute (M Kalz & M Specht, 2013) Learning processes and learning potentials are continuously investigated. Exploration of the possibilities of TEL led to the development of many solutions and recently to Massive Open Online Courses (MOOCs). MOOCs are capable of providing several ten thousands of learners with access to courses over the web (McAuly & al, 2010). The latest spread of MOOCs is enabling people to satisfy their learning needs. MOOCs provide a new way of learning, which is open, participatory, distributed and lifelong (Cormier, 2013). MOOCs have recently gained much attention especially in leading universities and are now considered as a highly promising form of teaching. Many universities are offering courses in the form of MOOC providing learners with a wide variety of choices. On the other side, this huge number of available and open resources led us to think on the way to help learners to not be lost. Learners need to find the most appropriate resource, among all proposed on the web, and to prevent them from being overwhelmed by the huge amounts of resources. Since information retrieval and searching for the appropriate learning resources is an essential activity in TEL, the development of recommender systems for learning has seen increased attention. Recommender systems permit to respond to a traditional problem in TEL which is "finding the best learning resources" for the learner. With MOOCs proliferation, this traditional problem is more than ever up to date. Learners will be exposed to various challenges with this excess of learning resources: which provider they have to choose to search for a specific MOOC? How to search for a specific MOOC? When a MOOC is found, how learner can be sure that it is the best one? Hypothetically, all MOOCs users would try to find services that help them identify suitable learning resources from this overwhelming variety of choices. Consequently, in this context, the concept of recommender systems can be particularly appealing. In this paper, we address these questions. We propose a system that recommends appropriate MOOCs in response to a specific request of the learner. Using the Case Based Reasoning (CBR) techniques, the system proposes to the learners the most appropriate MOOCs (from different providers) responding to his request. In the following, we will give an overview on Open Educational Resources (OERs) and MOOCs. In section 3, a brief review of the literature on the recommender system for learning is presented. The used approach namely the Case Based Reasoning and its relation with recommender system are presented in the next section. The proposed system is described in Section 5. Finally, Section 6 is dedicated to the conclusion and future works.

2. Open Educational Resources and MOOCs

Open Educational Resources are teaching, learning and research resources that reside in the public domain or have been released under an intellectual property license that permits their free use or re-purposing by others (Atkins & al, 2007). "Open Educational Resources are digitized materials offered freely and openly for educators, students and self-learners to use and re-use for teaching, learning and research" (Hylen, 2005). The term was first introduced at the first Global OER conference hosted by UNESCO in 2002. By providing open access to course content, the development of OER initiatives has paved the way for free online courses, such as Open Course Ware (OCW) and Massive Open Online Courses (MOOC).

MOOCs have recently gained much attention especially in leading universities and are now considered as a highly promising form of teaching. Many universities are offering courses in the form of MOOCs providing learners with a wide variety of choices. Fig. 1 below shows the fast growth of MOOCs in the world (European MOOCs, 2014).

Fig. 1: The fast growth of MOOCs in the world.

3. Recommender System for Learning

The term recommender system was coined to refer to a system using the opinions of a community of users to identify more effectively content of interest from a potentially overwhelming set of choices (Adomavicius & Tuzhilin, 2005). These types of systems have been used by online e-commerce sites to recommend items to their customers since 1990s (Resnick & Varian, 1997). However, it was only in the beginning of 21th century when the first notable applications arise in the domain of education (Manouselis & al, 2012). The proliferation of learning resources in the internet requires the use of recommender systems that support learners in finding their way through the multiple thousands of learning resources offered. Burke (Burke, 2002) defines recommender system for learning as any system that produces personalized recommendations as output or has the effect of guiding the learner in a personalized way to interesting or useful learning resources in a large space of possible options. However, recommending suitable resources for learning processes is more complex than recommending commercial items. Therefore designing TEL recommender system requires considering some particularities regarding user tasks and goals (Herlocker & al, 2004).

In general, recommendation can be user-based or content-based approaches or both (i.e hybrid approach). In the user-based approach, the users' preferences are analyzed and aggregated from the users' profile (Resnick et al. 1997, Goldberg et al. 2001). In the content-based approach, the system finds patterns or similar patterns related to previous experiences in order to recommend new ones (Ricci et al. 2011). The later approach is a form of case-based reasoning approach. Next section deals with this technique.

4. CBR related to Recommender System

Case-Based Reasoning (CBR) (Wilke & Bergmann, 1998) is a form of memory-based reasoning (Richter & Weber, 2013). Broadly construed, CBR is the process of solving new problems based on the solutions of similar past problems. In that, CBR approach is related to recommender system approaches. Case base recommenders are a form of content-based recommendations. They rely on descriptions of the items as the basis for recommendation. Case-based reasoning systems are distinguished from other forms of content-based recommendation systems by using fairly well-structured descriptions of those items (Manouselis & al, 2011).

The principle of CBR is that similar problems have similar solutions (Cafias & al, 1999). Most case-based reasoning systems represent problems and solutions as cases, and they maintain a library called a Case Base. Each case has two parts, a problem specification that describes the problem in question and a solution part. When a case based reasoning system is faced with solving a new problem, it tries to retrieve a case that is similar to the problem being solved, and then reuse that similar case's solution to adapt the solution for the purpose of the target problem. Case base reasoning systems have proven to be very useful, particularly, in domains where there is not a strong

problem solving knowledge (Richter & Weber, 2013) as the case of our system. In the following, we will detail how techniques of CBR are applied in our proposed system MOOC-Rec.

5. System description

Facing a wide range of learning resources provided by an increasing number of MOOCs, learners search for suggestions. MOOC-Rec is a web based application which takes into account the user's interests expressed in a query as a problem's features and provides suitable learning resources among MOOCs providers.

5.1 System Architecture

The architecture of MOOC-Rec is shown in Fig. 2. There are three layers: system data layer, system function layer and user interface layer.

Fig. 2: System architecture

• User interface layer : is the interface of MOOC-Rec system, whose main function is receiving requests that users input from interface.

• System function layer : This layer includes creation and processing of cases (case search, case retrieval and case adaptation) as well as the recommender reasoning function.

• System data layer : it includes three parts: case base, the user case and similar user case. The case base is accumulated and renewed constantly during the process.

5.2 Interface Layer

In MOOC-Rec, the user chooses the learning resource features according to their relevance from her/his point of view. The user selects five features: course title (keywords), fees, course's availability, language and location of the Open University. The features are given weight values in decreasing order; between 5 and 1 so that the first feature chosen has the high value of weight and the last one has the least value. Figure 3 shows an example where the user is requesting a course about the web, without any fees, in English and will be provided soon. As soon as the user enters his/her request, a number (from 1 to 5) is displayed to show to the user how the features are weighted. The example below (see Fig. 3) shows that the user entered by order of preferences the title, the language, the fees, and the availability.

Owt rrf ertler fe&turn Pf e-rtor ofjrtfr

■Ù 0

Fig. 3: System Interface

5.1 System Data Layer

The main function of this part is gathering data to be used by the CBR Recommender System.

5.1.1 MOOCs' Crawler

MOOCs' Crawler is a topical crawler. It regularly trawls the web MOOCs' home pages and forms indexes to be used by the Recommender search engine for fast recommendations. MOOCs' Crawler exploits rich cues and sources such as XML, Markup and lexical (text) signals within Web pages (Cho et al. 1998, De Bra et al. 1994). The pages of interests are visited at query time.

To populate the case base and to solve the problem of cold start when a new problem occurred, MOOC-Rec creates cases by extracting learning source features from XML pages (treated as cases) of MOOCs indexed previously by the crawler and listed in certain online directories dedicated to MOOCs such openculture.com shown in Fig. 4.

<lixa href="https ://www. edx.org/course/sniJx/snux-sn - 345 - 2x- robot -met h an ic s -1700#.U3 Ra86ldUYQ"> Robot Mechanics and Control, Part II </a>(CC) &# 8211; Seoul National University on edX - Dune 2 (6 weeks)</li>

<lixa iiref="https;//www.canvas.net/courses/math-ref resher-with-college-success-tips-l">Matli Refresher with iollege $uççess Tipj</a> (NJ) - Utah Valley University S»8211; lune 2 (9 weeks)</li> <lixa href="https ://www.canvas.net/courses/5-habits-of-highly-creative-teachers">5 Habits of Highly Creative Teachers</a> (N1) - Northwest Colorado BOCES on Canvas - Hune 2 (5 weeks)</li> <lixa href="https ://www.canvas.net/courses/risk-mansgement-in-higher-education">Risk Management in Higher Education </a>(NI) &#8211; Canvas Network &# 8211; June 2 (5 weeks)</li> <liXa

href="https://www. coursera .org/course/researchinetho ds">Jnderstanding Research Methods</a> (SA> â#8211; University of London on Coursera &#8211; ]une 2 (6 weeks)</li>_

Fig. 4: Sample of XML sources

5.2 System Function Layer 5.2.1 Overview

The main hypothesis behind CBR is simply that similar problems have similar solutions or that you can reuse the solution of a similar problem in order to solve your actual problem. A case is the most basic element representing an experienced situation. As shown in Fig. 5, the main techniques that make up the CBR system are: case representation, case retrieval, case adaptation and case reuse

Fig. 5: CBR system

The following sections give details on the CBR process.

5.2.2 CBR Recommender System

In MOOC-Rec, the CBR recommender system is very similar to generic CBR problem solving system including the classical four steps of CBR methodology (retrieve, reuse, adapt and retain) (Wess & al, 1993). This starts with a new problem, retrieves similar cases from the case base, suggests the retrieved solution(s) to the user or adapts the solution(s) to better solve the new problem and terminates the process by retaining the new case. CBR Recommender system guarantees that it retrieves the k cases that are maximally similar to the target problem by computing the similarity of the target problem to every case in case base. To reduce retrieval time, MOOC-Rec relies on the organization of cases in memory as was proposed in (Aamodt & Plaza, 1994). The organization of the case base is based on similarities between cases. A binary tree called a k-d tree is used to split the case library into groups of cases in such a way that each group contains cases that are similar to each other according to a given similarity measure. To ensure that the most similar cases are retrieved, the retrieval algorithm computes similarity bounds to determine which groups of cases should be considered first. In the following sections we describe the process steps of the CBR recommender system reasoning.

5.2.3 Case Representation

There are three main types of CBR that differ significantly from one another concerning case representation and reasoning: structural (ontology), textual (string) and conversational (list of questions) (Richter & Weber, 2013).The Case Base in MOOC-Rec is a finite data source. The cases are represented in a flat form by a set of attributes-value pairs i.e. an attribute-value vector. Each case is in general, a couple of (problem, solution). The problem is the user's query. It is described by five attributes: course's title, location (university where has being giving this course), language, availability and fees. Values of attributes are string for the features Title, Language, Location and Availability and Boolean for fees (1 if the proposed course is free).The solution is a set of MOOCs' URL responding to the user request. Each attribute in the problem is assigned a weight value according to its relevance expressed in the user's query when s/he chooses features of the desired learning resource. Fig. 6 shows a sample of a

case and its features in MOOC-Rec.

Problem Features Title: web intelligence and big data Location: India Language: English Free: True Availability: soon

Fig. 6: Sample of a case in MOOC-Rec.

5.2.4 Similarity and Retrieval

The retrieval step attempts to retrieve the k-cases (candidate cases) from the case base, that their problem is similar to a given new one. This is done by a reasoning process based on the degree of similarity. To assess similarity between the query and the cases, MOOC-Rec compares their attributes one by one. Each attribute has its relevance in the query and requires its own similarity function. Recall we mentioned that, in MOOC-Rec, the user is able to choose the learning resource features according to their importance from her/his point of view. The user selects five features such course title (keywords), fees, course's availability, language and location of Open University. Even though, we are in an on-line education context, we think that the location might be a distinguishing indicator in case of possible face-to-face final exams.

MOOC-Rec uses the Levenshtein distance to measure local similarity between cases' attributes. The Levenshtein function (Levenshtein, 1966) is a type of transformational measure which counts the number of changes needed to transform one string into another one. The possible change actions are: insertions / substitutions of characters to move from one string to another. The global similarity of each case to the target problem is calculated by taking weighted sums of local similarities of n attributes. Attributes weights Oi are crucial. They influence the global similarity between cases (Richter & Weber, 2013).

Sim(QueryP,CaseP)= ^(ai * LD(xi, yi)) , i in [1:n]

LD(xi,yi) is the levenshtein distance function to evaluate similarity between attributes-value pairs. It is limited to 255 characters but very fast. If xi equals yi, then LD(xi,yi)=0, no transformation are needed, the strings are identical. If LD(xi,yi) > 0, the strings are not equals. More the distance is higher plus the strings are different.

Two learning resources with the same title would get the maximum similarity percentage on the metric of title if the user chooses title as the first feature to consider, but may differ greatly on another metric, such as location or availability.

1 J4A ulWV C U41 J "N

Location: wt-1 Lancuici: Enslifh wc—4 Free:Truc ^t-1 Avallabllltv scon mr-2 Title m«b ¡ntellifeneetind t>ig data Location: India LAheuaie Lngiisi^i Frce:Tru« Availabilitv soon

Froviicrl: courscra.org

ri^-A "JllCl . (a) CUIrJ

Pinblnn Fralum A

Title: Web 5- Location: wt-1 Laneu^ei Enelish F»*:Trt№ j-jr— Availabiliry scon uvt-I TiiiervveD deveiopmern: Language^ En^listi Avdilibiliiv- current

th| Provicterl: L»dJtily org Provider2: ^

Fig. 7: Comparison between a query and cases.

In order to illustrate how MOOC-Rec identifies the most similar cases to the current problem, we take an example of a query and two cases into the case base (see Fig. 7).

(a) Sim(QueryP,Case1P)=

1/15 (5* LD(QueryTitle,Case1Title) + 3* LD (QueryFees,Case1Fees) + 4* LD(Querylanguage,Case1language) + 2* LD (QueryAvailability,Case1 Availability) + 0* LD (QueryLoc, Case1Loc)) = 1/15(5*22 +3*0+4*0+2*0+0*5)= 7.33

(b) Sim(QueryP,Case2P)=

1/15 (5* LD(QueryTitle,Case2Title) + 3* LD (QueryFees,Case2Fees) + 4* LD(Querylanguage,Case2language) + 2* LD (QueryAvailability,Case2Availability) + 0* LD (QueryLoc, Case2Loc)) = 1/15(5*11 +3*0+4*0+2*6+0*5)= 4.46

The retrieval step in MOOC-Rec identifies the cases that are most similar to the description of the problem. The retrieved cases are ordered in descending value of similarity.

5.2.5 Reuse and Adaptation

The result of the similarity measure allows the system to decide on the adaptation to perform. Different forms of adaptation exist, such as null adaptation, transformational adaptation (including substitutional and structural adaptation), and generative adaptation (Wolfgang & Bergmann, 1998). In MOOC-Rec, we use null adaptation. The case base is searched for a similar case containing the most similar user requirements and the solution from this case is proposed to the user to solve the current problem without any modification. If no similar solution is found in the case base, a search will be processed on the web; a new solution is constructed and saved in the case base.

6. Conclusion and Future Works

Nowadays, learners are overwhelmed by the huge amounts of resources available on Internet. One way to overcome this difficulty is to use recommender systems and specifically CBR recommender systems. The later are solutions in a world of ever-growing online learning resources. Unlike other recommender systems, CBR recommender systems do not need to store large masses of data about items rating or particular users. Our experience when designing MOOC-Rec has shown that the data base does not need to be excessively large, since we need only enough features to search for similar cases. MOOC-Rec system helps users benefit from the MOOCs proliferation and select in an easy way the learning resources that better fit their interests.

The system proposed in this paper is under implementation and we intend to test it in real-world situation. As a future work, we plan to add more search features such that learning outcomes, course syllabus and course prerequisites.

Acknowledgements

This research project was supported by a grant from the "Research Center of the Female Scientific and Medical Colleges", Deanship of Scientific Research, King Saud University.

References

Kalz, M, Specht, M. (2013). Assessing the cross disciplinarity of technology enhanced learning with science overlay maps and diversity measures. In British Journal of Educational Technology.

McAuly, A., Bonnie, S., Siemens, G., Cormier, D. (2010). The MOOC Model for Digital Practice, Knowledge Synthesis Grants on the Digital Economy, University of Prince Edward Island, http://www.edukwest.com/wpcontent/uploads/2011/07/MOOC_Final.pdf.

Cormier, D. "What is a MOOC?" 8 December 2010, http://www.youtube.com/watch?v=eW3gMGqcZQc, Accessed January 2013.

Atkins, D. E., Brown J. S., & Hammond A. L. (2007). A review of the Open Educational Resources (OER) movement: Achievements, challenges, and new opportunities. The William and Flora Hewlett Foundation. 1-84.

Goldberg K., Roeder T., Gupta D., Perkins C. (2001). Eigentaste: a constant time collaborative filtering algorithm. Inf. Retrieval 4(2), 133— 151.

Hylen, J. (2005). Open educational resources: Opportunities and challenges. OECD-CERI. Retrieved 30 March 2014 from http://www.oecd.org/edu/ceri/37351085.pdf.

European MOOCs Scoreboard. http://openeducationeuropa.eu/en/european_scoreboard_moocs, Accessed April 2014.

Ricci, F., Rokach, L. & Shapira, B., (2011). Introduction to Recommender Systems, Handbook, DOI 10.1007/978-0-387-85820-3_1, © Springer Science+Business Media, LLC.

Adomavicius, G., Tuzhilin A., (2005). Towards the next generation oof recommender systems: a survey oof the state-of-the-art and possible extensions. IEEE Trans. Knowl. Data Eng. 17(6), 734-749.

Resnick, P., & Varian, H.R. (1997). Recommender systems. Commun. ACM 40(3), 56-58.

Manouselis, N., Drachsler, H., Verbert, K., Duval, E. (2012). Recommender Systems for Learning. Springer, Berlin.

Anderson, C. (2006). The Long Tail: Why the Future oof Business Is Selling Less oof More. Hyperion URL http://www.amazon.ca/exec/obidos/redirect?tag=citeulike04-20{\&}path=ASIN/1401302378

Burke, R. (2002). Hybrid Recommender Systems: Survey and Experiments, User Modeling and User Adapted Interaction, 12, pp. 331-370.

Herlocker J.L., Konstan J.A., Terveen L.G., Riedl J.T. (2004) "Evaluating Collaborative Filtering Recommender Systems", ACM Transactions on Information Systems, Vol. 22, No. 1, January 2004, Pages 5-53.

Wilke, W., & Bergmann, R. (1998). Techniques and knowledge used for adaptation during case-based problem solving. Proceedings of IEA-98-AIE. Berlin, Heidelberg, New York: Springer-Verlag.

Richter, M. M., Weber, R.O., (2013). Case-Based Reasoning. Springer-Verlag Berlin Heidelberg. Chapter 2, p.17

Manouselis, N., Drachsler, H., Vuorikari, R., Hummel, H.G.K., Koper, R. (2011). Recommender Systems in Technology Enhanced Learning. In: Kantor, P.B., Ricci, F., Rokach, L., Shapira, B. (eds.): 1st Recommender Systems Handbook. Springer, Berlin.

Cafias, A. Leake, D. Wilson, D. (1999). Managing, mapping, and manipulating conceptual knowledge. Proceedings of the AAAI-99 Workshop on Exploring Synergies of Knowledge Management and Case-Based Reasoning.

Richter, M. M., Weber, R.O., (2013). Case-Based Reasoning. Springer-Verlag Berlin Heidelberg.

Cho, J., Garcia-Molina, H., & Page, L., (1998). Efficient crawling through URL. Ordering. Computer Networks, 30(1-7): p. 161-172.

De Bra, P. & Post R.,(1994). Information retrieval in the World-Wide Web: making client-based searching feasible. Computer Networks, 1994, 27(2): p. 183-192.

Wess S, Althoff K-D, Derwand G (1993). Using kd-trees to improve the retrieval step in case based reasoning, In: Wess S, Althoff K-D, Richter MM (eds) Topics in case-based reasoning. EWCBR-93: first EuropeanWorkshop, Kaiserslautern, Germany, 1-5.

Aamodt. A and E. Plaza., (1994); Case-based reasoning: foundational issues, methodological variations, and system approaches, AI Communications, 7(1):39-59.

Levenshtein, V., (1966). Binary codes capable of correcting deletions, insertions and reversals, SOL Phys Dokl, pp, 707-710.

Wolfgang W., Bergmann, R. (1998). Techniques and knowledge used for adaptation during case-based problem solving, Tasks andMethods in Applied Artificial Intelligence, Lecture Notes in Computer Science Volume 1416, 1998, pp 497-506