A M-Learning Content Recommendation Service by Exploiting Mobile Social Interactions

1939-1382 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. Seehttp://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citationinformation: DOI 10.1109/TLT.2014.2323053, IEEE Transactions on Learning Technologies

IEEE TRANS. ON LEARNING TECHNOLOGIES, VOL. , NO. , 2013 1

A M-Learning Content RecommendationService by Exploiting Mobile Social

InteractionsHan-Chieh Chao, Senior Member, IEEE, Chin-Feng Lai, Member, IEEE, Shih-Yeh Chen

and Yueh-Min Huang, Senior Member, IEEE

Abstract—With the rapid development of the Internet and the popularization of mobile devices, participating in a mobile

community becomes a part of daily life. This study aims the influence impact of social interactions on mobile learning

communities. With m-learning content recommendation services developed from mobile devices and mobile network techniques,

learners can generate the learning stickiness by active participation and two-way interaction within a mobile learning community.

Individual learning content is able to be recommended according to the behavioral characteristics of the response message of

individual learners in the community, and other browsers not of this community are attracted to participate in the learning content

with the proposed recommendation service. Finally, as the degree of devotion to the community and learning time increases,

the learners’ willingness to continue learning increases. The experiment results and analysis show that individualized learning

content recommendation results in better learning effect. In addition, the proposed service proved that the experiment results

can be easily extended to handle the recommended learning content for learners’ time-varying interests.

Index Terms—M-Learning, Content Recommendation Service, Mobile Social Community, Social Interactions

✦

1 INTRODUCTION

W ITH increasing computer and network tech-nologies, social network sites (SNSs), such as

Facebook, and Plurk, Twitter, become an online virtualcommunity formed by a group of people with thesame interests and activities, where users can inter-communicate using the network. Many SNSs providemultiple interactive modes for users, such as chat,messaging, video and file sharing, blogs, discussiongroups, etc [1], [2], [3]. According to statistical data,people now spend more than 20 minutes on Facebookon average per day. The situation illustrates the usagerate of SNSs is very high. Therefore, as styles oflearning and education gradually evolve, SNSs haveentered educational circles, where they have becomepopular. At present, the learning behaviors developedfrom SNSs shows three main trends: (1) informationdiscovery is changing; (2) how we share is changing;(3) social sites have become social webs.

”Information discovery is changing”, meaning alearner’s friend can instantly and actively assist insearching and solving the problems of the learner.For example, when a question is raised in Facebook,

• H.C. Chao is with the Institute of Computer Science and InformationEngineering, National Ilan University, ILan, Taiwan.E-mail: [email protected]

• C.F. Lai is with the Department of Computer Science and InformationEngineering, National Chung Cheng University, Chiayi, Taiwan.E-mail: [email protected]

• S.Y. Chen and Y.M. Huang are with the Department of EngineeringScience, National Cheng Kung University, Tainan, Taiwan.

anyone who sees the message can instantly reply.Therefore, the traditional learning behavior pattern ofsearching for answers on the Internet or in books ischanging [4].

”How we share is changing”, means that when alearner wants to share important information withothers, they need not send individual e-mails as in thepast. The information can be shared easily and rapidlywith the energy of a virtual learning community, thus,more high social energies can be shared. For example,when releasing new homework, using the sharingfunction will transfer information to all learners ofthe same virtual learning community [10].

”Social sites to social web”, means that a learner canuse SNSs to view which websites other learners havebrowsed. For example, when a new theme is issuedto a SNS, and the friends of a learner browsed thistheme, the learner can learn of the new theme fromreal-time dynamic messaging. Consequently, morelearners of this learning community enter the learningwebsite of this theme, thus, promoting interactionbetween students and learning websites [11].

In a SNS, friends may have a common interest,work environment, or come from the same school, anyslight relevancy can be a reason for making friendsin a network. Therefore, a normal phenomenon inSNSs is that a user can have hundreds of friendsor friend links, as it is easier to build friend linksin a virtual network than in real life [12], [13], [30],[31]. However, such virtual learning communities,with massive amounts of information create hugedata volumes everyday, as learners must occasionally




spend extra time finding information regarding theirinterests.

Regarding of mobile social interaction, it refers tolearners could share information and content eachother to build social relationships [43]. Learners inter-ests and favors could be analyzed by characteristicsof shared content and information. The informationwith reference value is able to be used for learningservices to recommend content to learners accurately.In the recent years, various mobile social interactiontechnologies were adopted to connect learners, con-tent and resources to create context-rich learning envi-ronments, and novel learning services were designedthat lead learners could share personal data andknowledge to achieve collaborative learning activities,even recommend learning partners to learners withthe same interests [5], [6], [50]. In that case, mostprevious studies placed emphasis on collaborativerecommendations for information regarding interestsor preferences similar to that of other learners vialearners social interactions [7], [8], [9], [14]. While therecommended items found by such a method havehigh isomorphism with the original preferences ofthe learner, three factors are neglected: 1) the learneris more curious about friends items of interest thanthe items recommended by the recommender system;2) the learner will change interests and preferenceswith time and learning environment changes; 3) thelearner’s latent learning direction and interest areinitiated as popular subjects rise.

Therefore, this study does not place emphasis onusing similarity as a unique consideration in complexinterpersonal networks with numerous target users,but focuses on multi-aspect social network analysisto determine friends with highly correlated interestswith the target users, taken from great friend links.Such recommended items are no longer limited tosimilar users, but to recommend the interesting itemsto target users from other non-similar in interper-sonal networks through social interactions. Such arecommendation mode better enables target users toreceive more diversified knowledge sharing throughinterpersonal networks.

2 BACK GROUND AND THEORY-GUIDED

RESEARCH

2.1 Activity Theory - Situated Learning Theory

Situated learning is a learning style proposed by Prof.Jean Lave of the University of California, Berkeley,and independent researcher Etienne Wenger [24]. Ac-cording to the situated learning theory, learning is notonly a psychological process of individual meaningconstruction, but also a social, practical, and differ-ent resource mediated participation process [25], [26].The meaning of knowledge and the learners’ ownconsciousness and roles are derived from interactionsbetween learners and learning situations. Therefore,

the creation of a learning situation aims to regress thelearners’ status and role consciousness, life experienceand cognitive tasks to overcome lacks of traditionalschool learning [27], [28].

According to the situated cognition theory, themeaning of situated teaching is that a teacher candesign diversified teaching situations according tothe different learning characteristics of learners, andintellectual and perceptual knowledge learning is ob-tained by participating in learning [19], [20], [49].In other words, situated teaching centers on learn-ers, where the learners are in the teaching situation,and participate in mobile learning, exploration, andfeedback. The purpose is to allow learners interactin diverse environments, and adapt themselves todevelop and build their own knowledge meaningfully[47]. In short, the concept of situated learning is toreceive learning in actual or simulated situations,where students can apply the learned knowledgemore efficiently to real life through interactions be-tween learners and situations. The situated learningtheory has three teaching styles: cognitive appren-ticeship, learning community, and anchored instruc-tion [21], [22], [23]. The teaching style adopted inthis study is learning community, which is a sort ofcultural situation. In this situation, each individualnot only actively studies, but also bears responsibilityfor their learning, and the collective learning of all,through the mutual support of interpersonal inter-actions. The learning community has the followingbasic elements: 1) The teaching situation must bebased on actual tasks; thereby, students can createconnections between abstract scientific concepts andapplication in real world. 2) The students must de-velop the interdependence of group cooperation inorder to promote interdependence by idea sharing,discussions, and disputation. 3) The students andteachers must discuss and negotiate mutual ideas andcomprehension. 4) The students and teachers mustshare ideas and thoughts in public in order to assistthe students to explore and correct their conceptsand comprehension. 5) The students must cooperatewith experts outside classroom. 6) To jointly bear theresponsibility of learning and teaching, all members(including teachers and students) must actively par-ticipate in ”teaching” and ”learning” activities.

2.2 Social Presence - Mobile Social Community

There are many definitions of traditional communi-ties. H. Stern indicated ”the community is a placewhere people build relationships; it is a dynamic,continuous, and non-rigid construction process” [15].D.C. Teng thought ”The community is a social organ-ism formed based on free will, such as emotion, habit,memory, mentality, or consanguinity” [32]. E. Graybriefly indicated ”the community is the assembly ofa group of people”. The opinions of most scholars




on community are sorted to be brief, a community is”a group of people assembled for mutual exchangeof interest in a specific region. The members of thisgroup trust each other, and have the sense of be-longing and identity” [33]. S. Redfern indicated thatthe nature of community could be divided into Geo-graphical Community and Relation-related Commu-nity. The Geographical Community refers to the pop-ulation established in the same region. The Relation-related Community is built on intangible qualitiesof interpersonal relationships, interests, or ambitions,and is free from the restriction of geographic area [34].However, with the development of the Internet andbreached time and space limits, the Internet connectsVirtual Community emerges accordingly. Armstrongand Thoms, N. Garrett defined virtual communityas ”an interactive space built on the Internet by agroup of community members aiming at common andspecific subjects”. The nature of a virtual communityis relatively similar to the aforesaid Relation-relatedCommunity; they have the natures of interaction,sharing, exchange, and sense of belonging. The onlydifference is that they can contact other memberswithout in-person contact [35]. The concept of a Mo-bile Community is integrated with the element ofmobile equipment, further enhancing the convenienceof community communication. C. Prykop indicated”the Mobile Community is formed of independentregional groups, it is held by clear common interest,and social interaction is undertaken through mobilechannels” [36]. Although a Mobile Community isregarded as an extension of a virtual community, thereare actually some differences between them.

1) A Mobile Community can be undertaken bymobile equipment: the Mobile Community providesservice via various mobile hand-held devices. In short,the interface of Mobile Community extends from com-puter to mobile hand-held devices [37], [38].

2) The Mobile Community can provide new com-munication services based on the application of amobile network: Beale and Lonsdale indicated that,due to the characteristics of mobile technology, MobileCommunity service can provide some communicationservices that cannot be implemented in a network.For example, the users are provided with real-timeinformation of their area [39].

3) The Mobile Community and virtual commu-nity have different operating modes: compared witha virtual community, the interaction among MobileCommunity members is subtler. Although they havenot met each other in-person, their interactive rela-tionship can be deepened by the characteristics ofmobile equipments. C. Prykop thought that mobileequipment has four characteristics, including locationawareness, ubiquity, identification, and immediacy[36].

2.3 Constructivism - Personal learning content

As e-learning is learner-oriented, personal learningcontent has been an inevitable development trend[48]. We can find that many m-Learning integrated ap-plication platforms, including Learning ManagementSystem (LMS), have tried to integrate personal ele-ments into systems. However, most current e-learningsystems lay particular stress on learning content pro-duction, storage, management, delivery, and presen-tation, as well as on the personal management func-tions of learner data, class authorization, and learningprogress tracking. Learners’ learning behavior benefitanalysis is lacking here.

If all e-learning materials (including authentica-tion, attribute, and category data) and learner data(including basic data, learning courses, and coursebenefit) can be built in a unified database, we canuse data mining technology for statistical analysis ofthe massive learning behavior information accordingto the model of personal service recommender system[40], [41], [42].

After extensive learner behavior analysis, the rec-ommender system can recommend teaching materialsand courses meeting the personal needs of learners.Personalized recommendations recommend teachingmaterials and courses meeting the learners’ needsaccording to preference, and learning benefits accord-ing to the group similar learners’ needs, commoninterests, and common experiences. Based on long-term behavior data accumulation, personalized rec-ommendation will become increasingly accurate. Onthe other hand, in the research area of e-learning,how to increase the reusability of different formsof learning resources (e.g. webpage, learning object,video animation, etc.) is very important .

3 SERVICE DESIGN AND ARCHITECTURE

In order to attain the purpose of research, this studydeveloped a set of m-learning content Recommenda-tion Services by Exploiting Mobile Social Interactions.The entire M-Learning Content Recommendation Ser-vice contains a mobile learning community function,GPS function, mobile network function, and punchfunction, as well as a student database, learning objectdatabase, and learning process database. Figure 1shows the architecture of m-learning content recom-mendation service.

In a learning environment, a learner uses a mobilecarrier to observe learning objects in a real-time en-vironment and punches the learning point, uses thephoto or file upload function in the service system toupload the data of the learning object to the system,and uses the mobile learning community function torelease posted messages to the learner’s friends, thus,interesting the learner’s friends in viewing the dataof the learning object. This service system displaysdifferent dynamic prompts according to the degree of




Content Recommendation

Learning ExperienceMobile Social Interactions

Content Database

Check-In Function

GPS

Check-In Function

GPSndation

e

Interactions Data

Fig. 1. Architecture of M-Learning Content Recom-mendation Service

interaction between the learner and friends. For exam-ple, if the interactive relationship has high correlation,any action in the community will be reported by thesystem, such as releasing a learning object, liking alearning subject, giving an opinion on a learning sub-ject, etc. The knowledge is automatically organizedand constructed by the observations and interactionsbetween the learner and mobile learning communityfriends, thus, assisting in initiating the learner’s inter-est in learning, while personal knowledge is graduallyconstructed. Thus, the learner can rethink existingknowledge, while observing the interactions of themobile learning community. In the course of obser-vation, the constructed knowledge varies due to thedifferent learning effects of learners.

3.1 Infrastructure for Mobile Social Interactions

The rapid development of the Internet means there arenumerous open resources, which present another dis-cussed subject, namely, the ”Information Overload”of the Internet. People search for required resourcesin a network with the assistance of search engines.Compared with search engines, the recommendersystem actively provides learners with required in-formation. The recommender system can recommendpotential information, services, or products requiredby learners, and according to learners’ interests, pref-erences, behaviors, or other needs [16], [17], [18].The computing process of a recommender system isdivided into three major phases, as follows: 1) Toextract learners’ profile, e.g. interests, purchase lists,bookmarks, favorites, preferences, etc. 2) The collectedlearner information is analyzed and appropriate rec-ommended items are generated, e.g. similarity calcu-lation. 3) To appropriately modify the recommendersystem according to the learners’ satisfaction with rec-ommended information. The extraction of a learner’sprofile can be classified into two modes:

(1) active mode: the learner actively provides per-sonal information, or the system requires the learnerto provide personal preference information directly tothe system for calculation. For example, J.S. Huang re-quired a social learner to score Trust in friends, and setthe recommender system to recommend higher trustin a collection of friends to the learner [44]. Althoughthe active data acquisition mode can obtain accurateanalytic information directly from the learner, the factthat it requires the learner to provide information islikely to cause discomfort for the learner;

(2) passive mode: the required analytic informationis collected from the learner’s behavior. For exam-ple, probable learning content of the future are rec-ommended according to the historical mobile socialinteractions of the learner [45], [46]. However, the pur-pose of this study is to seek multiple recommendedcontent, as based on factors other than similarity.Therefore, in addition to the similarity calculation, thisstudy uses various indices to evaluate the degree ofcorrelation between each learner and a target learner,and seeks the learning content of other learners highlycorrelated with the target learner in the interpersonalnetwork extended from the target learner. Afterwards,the recommendation is calculated according to therecent interest of the target learner and the learningcontent.

This study is divided into two phases: (1) socialnetwork strength analysis of target learner; (2) to de-termine the learning content of the recent preferencesof the target learner considering time factors.

3.1.1 Mobile learning community strength analysisphase

The score of correlation between target learner andeach learner is calculated. The recommender systemmostly uses titles or learning content for similaritycalculation to determine similar learners in the past;whereas, the number of common friends in the ex-isting community platform of Facebook or Twitterbecomes a base of recommended friends in socialnetworks. Therefore, this study adopts this concept inorder to discover the person or favorite learning con-tent, which is probably dissimilar to the target learner,but interesting to the target learner. The considera-tion of popularity evaluates the value of followingthe learner. If popularity is high, likely meaning thelearner is a famous person or the learner’s learningcontent can interest numerous people, this person islikely followed by multiple learners. Figure 2 showsthe concept of mobile learning community strengthanalysis and the index equations of (1) to (3) are, asfollows.

ELCS(α, β) =∑

l∈L

Fα,l ∗ Fβ,l√

F 2

α,l ∗√

F 2

β,l

(1)




Target user

Level 2

Level 2

Level 2

Level 2

Level 2 Level 2

Level 2

Level 1

Level 1

Level 1

Level 1

Fig. 2. Concept of Mobile Learning CommunityStrength Analysis

l:Specific M-Learning ContentL:Set of M-Learning ContentFα,l:M-Learning Frequency of Learner α to LearnedContent lFβ,l:M-Learning Frequency of Learner β to LearnedContent l

Common Friend =

Number of Common Friends

Number of Target Learner Friends

(2)

Popularity =

Number of Target Learner′s Followers(3)

The various index values in Eqs. (1)-(3) are derivedfrom the data of different references. Eq. (1) is usedto calculate m-learning content-based similarity, Eq.(2) is the measurement mode for number of commonfriends, and Eq. (3) is designed as the popularitymeasurement mode. In order to calculate the weightsof the various values, and to avoid a too large or toosmall eigenvalue influencing overall data accuracy,the normalization calculation of Eq. (4) is conducted tomake the various attribute values fall within [0,1]. Thecomputing mode is that each index minus minimumvalue of the index is divided by the maximum minusminimum value of the index. Weights are divided intothree parts by the M-Learning Content-based Simi-larity, Mobile Social Interactions (number of commonfriends, number of friends’ responses) and Popular-ity. Finally, the normalized index is given differentweights, according to Eq. (5), to determine the finalscore, and the interpersonal network of higher corre-lation with the target learner is sorted after ordering.

X ′(i) =X(i)−Min(X(i))

Max(X(i))−Min(X(i))(4)

X ′(i): Normalized Attribute

RelevanceScore =

(Weights ∗ ELCS(α, β)) + (Weights ∗ Common Friend)

+ (Weights ∗Mutual Follow) + (Weights ∗ Popularity)(5)

The social network strength relevancy score com-puting mode consists of Steps 1 to 3: Step1, Eqs. (1) (3)are implemented to evaluate the network strength ofvarious learners and target learner. Step 2, the valueobtained from Step1 is normalized by Eq. (4). Step3, the normalized value is substituted in Eq. (5) tocalculate total score.

3.1.2 Build interest profiles according to the recom-

mendation of the target learner: strengthen recent

interest

Learners may have different preferences for learningin different periods. Considering the accuracy of arecommender system, this study determines recentlypreferred learning content, according to the learningsituation of learners in the recent period. Markedfrequency and time weights are considered in thisphase. The frequency of using the learning contentof a learner can be regarded as a preference for thecontent. The higher the frequency, the higher thepreference of learners for the learning content. Thecloser the time point of using the learning contentto the present, the higher the time weight, and thefarther the time point of using the learning content tothe present, the lower the time weight, according toequations (6) and (7). Finally, the two indices of Eqs.(6) and (7) are integrated to determine the preferencescore (PS) of learners for each learning content in aspecific period, as expressed in Eq. (8).

T ime Weight =

1−Number of Days to Learn Content Until Now

Number of Observation Days(6)

Frequency =

Frequency of Using Content in the Day

Frequency of Using All Content in a Period

(7)

Content Preference Score =∑

(T ime Weight ∗ Frequency)(8)

3.1.3 Calculate M-learning Content Recommenda-

tion Weight

The concept of social network analysis is evaluatingthe impact of independent nodes among the networkand discovering the social interaction between theusers. In this concept, the trend can be found thatthe social network analysis is able to be adopted toevaluate the quantitative indicators of interpersonal




interaction, whereby potential interactions or poten-tial friends can be discovered. Anyhow, the three basicelements should be considered in social networks.One is Node that represents Actor in the communities,two is Relationship that means connection betweenActors and T ie is the behavior strength betweenActors. Therefore, naive Bayes classifier was adoptedfor mobile social interactions analysis. To apply thenaive Bayes classifier to the recommendation processof the target learner m, the content is referred to theclass m-learning content and all of the learned contentin the candidate tag model for learner m, CCMw(m)= c1, c2, . . . , cw, are used as features. Suppose thatthere are n class m-learning content, E = e1, e2, . . . ,en. Given a learned content instance cj in CCMw(m)as a set of feature variables, the classifier predicts thatCCMw(m) belongs to the class item having the high-est posterior probability conditioned on CCMw(m).We make a similar naive Bayes assumption that thetags are independent given the items. The posteriorprobability, as a preference probability Pm,k of userm with CCMw(m) for content ek is obtained.

The target learner’s recent preference tag weightand social network RS are considered, the influenceis the RS of the learner having a collection the targetlearner wants, as expressed by Eq. (9). A higher scorerepresents stronger influence, namely, this item ismore probable to be learned by the target learner, andthe content preference score of the target learner ofthis item represents a recent preference of the targetlearner for the learning content.

Pm,k = P (ey|CCMw(m)) (9)

3.2 Mobile Community Learning System Interface

The main menu of this system contains punch andlearn, recommended learning, learning process, mo-bile edition, and login. The learners use portablelearning tools, a 3G wireless network, and the punchcharacteristic of a Mobile Community. The form fea-tures are as shown in Figure 3 to Figure 5.

Fig. 3. Network and Device Aware Bayesian Network



4 EXPERIMENTAL RESULTS

In order to validate the reference value of the m-learning content recommendation service, as pro-posed in this study, this experiment randomly selected60 primary school students as target learners. Thetotal number of Level 1 friends of these learners is587 through the numbers of mobile social interac-tions, and the maximum RS with target learners isdetermined from the 587 persons. Therefore, there are53 persons of Level 1 extending to higher correlationlevel (Level 2). There are 2521 persons at Level 2. Thetotal number of persons of the two levels is 3108. Thefollowing statistical validation is conducted.

1) Statistical analysis results of common friends,popularity, and similarityThe popularity recommendation is a constantly usedrecommendation mode, thus, the indices proposed inthis study are compared with the popularity recom-mendation. As shown in Figure 6, according to thecomparison between various levels and non level, thenumber of common friends is higher correlation iswith similarity, rather than popularity. In addition,Figure 7 shows that two learners following each otherhave higher similarity, as compared with one-wayfollowing.

2) Statistical analysis results of number of targetlearners’ friends and common friends




0

0.1

0.2

0.3

0.4

0.5

0.6

Level 1 Level 2 Non Level

Correlation of Common Friends and Popularity with Similarity

Popularity Common Friends

Fig. 6. Correlations of Common Friends and Popularitywith Similarity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Level 1 Level 2 Non Level

Comparisons with Following Relations and Similarity

One-Way Following Following Each Other

Fig. 7. Comparisons with Following Relations and

Similarity

As shown in Figure 8, according to the validationof relevance between the number of friends and thenumber of common friends of 60 learners; when thelearners have a larger number of friends, the numberof common friends is larger, and the correlation coef-ficient is 0.602. Figure 9 shows another case, the largerthe number of friends (Level 2) of Level 1 friends oflearners, the larger the number of common friendswith target learners, and the correlation coefficient is0.682.

3) Analysis of keywords in recommended contentresultsThis study evaluates the recommendations for 60target learners. The first stage of the experiment ispresently conducted, the recommendation mode se-lects one person with a maximum RS from the Level2 following of the target learners, and the recentlyrecommended learning content is recommended tothe target learners.

There are three evaluation modes: (1) to checkwhether the owner of the learning content to be

0

0.2

0.4

0.6

0.8

1

1.2

0 10 20 30 40 50 60 70

Correlations of Target Learner's Friends Number and Average

Common Friends Number

Number of Target Learner's Friends Number of Average Common Friends

Fig. 8. Correlations of Target Learner’s Friends Num-ber and Average Common Friends Number

0

0.2

0.4

0.6

0.8

1

1.2

0 10 20 30 40 50 60 70

Correlations of Level 2 Friends Number and Average Common

Friends Number

Number of Level 2 Friends Number of Average Common Friends

Fig. 9. Correlations of Level 2 Friends Number and

Average Common Friends Number

recommended is already followed by target learners;(2) to check whether the keywords in the learningcontent to be recommended overlap the first 10 itemsof frequent learning content or recently used 19 itemsof the learning content of the target learners; (3) tocheck whether the keywords in the learning contentto be recommended overlap the keywords in recentlearning content of the target learners. According tothe aforesaid three areas of evaluation comparisonfrom among the 60 target learners, eight persons meetthe matching mode (1); four persons meet the match-ing mode (2); two person meets the matching mode(3); the results show that only three persons have nokeyword or content overlap with target learners insuch a recommendation mode.

4) Analysis results of content instance selectionsRegarding of content selections, the target learnersposts the preferred course instances to his friendsthrough the recommendation service, and the inter-actions are recorded by system once his friends clickthe content sharing message. The response results




collected in this study are shown in Figure 10. Forthe specific content instance, the content attract moreresponses from the Level 1 friends in the beginningstage. It means the level 1 friends are able to sense thenovel content posted in the communities immediately.However, the response times of the level 1 friends islimited by the target learner’s friends amount, andthe response times of level 2 friends arise substantiallythrough the diffusion of the level 1 friends’ responses.Furthermore, in Figure 11, [44] is refereed to comparewith the proposed service. In the second part, thelearners are able to get the recommended content bySocial Learning Networks. All issues the learners canlearn only depend on what their friends provide. Thecondition maybe trends the issues fall in the limitedlearning areas through a few incentives. And the firstpart shows that content instance selection proportionthat the learners select; the recommendation serviceprovide the content instance lists from their friendsand popularity issues to the learners, more than halfthe content instance selections are from level 1 friends,and 13% are referred to the popularity issues thataren’t followed by their friends.

0

100

200

300

400

500

600

700

800

900

5 10 15 20 25 30 35 40 45 50 55 60

Re

spo

nse

s (T

ime

s)

Time (Mins)

Response Analysis for Specific Content Instance

Level 1 Friends

Level 2 Friends

Non Level Friends

Fig. 10. Response Analysis for Specific Content In-

stance

5 CONCLUSIONS AND FUTURE WORK

Web 2.0 implements the existence and collective cre-ation of virtual social networks. The platform value ofa virtual community is completely co-produced andcreated by the members, which phenomenon meetsservice dominant logic, and place emphasis on co-producing and co-creating value during the recom-mendation learning content service delivery process.Individual learning content is recommended accord-ing to the behavioral characteristics of the responsemessage of individual learners in the mobile socialcommunity, and other browsers not of this communityare attracted to participate in the learning content.In addition, the functions of a mobile social network

0% 10% 20% 30% 40% 50% 60% 70%

Content Instance Selection ProportionPopularity Issues

(Proposed Service)

Level 2 Friends

(Proposed Service)

Level 1 Friends

(Proposed Service)

Level 2 Friends

(Social Learning

Networks)

Level 1 Friends

(Social Learning

Networks)

Fig. 11. Content Instance Selection Proportion

can be divided into two types, social support, anda reference function. The social support refers to themembers’ physiological, psychological, tools, or ma-terial assistance to others, which allows individuals torespond to pressure and adapt to the environment. Interms of three mobile social networks, the friendshipnetwork has the strongest social support. On thecontrary, the reference function means the membersobtain messages, resources, and knowledge from amobile social network. The reference function is espe-cially applicable to information and advice networks,due to consultation, emotion, and information.

The aforesaid analytical data can validate severalfindings in a mobile social interactions: (1) the numberof common friends is positively related to similarity;(2) the larger the number of friends, the larger thenumber of common friends; (3) recent learning con-tent is recommended according to the RS proposed inthis study, there are 73.3% (40 of 60 persons) havinginterest overlap with target learners. Compared withprevious similarity calculation-based learning contentrecommendation services, the aforesaid findings havethe following advantages: (1) if the target learnershave not learned any content, they still have recom-mended items according to the social network score;(2) the number of common friends and followerscan be used as filter indices of the recommendationservice. The larger the number of friends, the largerthe reference value of the number of common friends;(3) recent interest recommendation meets the learners’needs.

This study has the following research limitationsand suggestions, as the learners of the mobile socialcommunity are of various age-brackets, this studyadopted convenience sampling, there may be biasederror in inference. Therefore, in order to increase therepresentativeness of samples, the number of sam-ples, and the diversity of sample sources, shall beincreased. In addition, there is a substantial differencebetween a mobile social community and a traditional




community, there is a group in the network world.They only browse the content provided by othercommunity members, and seldom or never partic-ipate in community activities. They merely browsewebpages as spectators, and probably do not con-tribute to the co-production process of the network.Secondly, mobile social networks have more than onepurpose in general, thus, it is difficult to distinguishthe exact purposes, and findings are not always ap-plicable to explain all types of virtual communities.It is suggested to finely and precisely classify thequestion items developed. Later scholars can usestructured and quantitative Social Network Analysis,actual quantitative data and qualitative interviews,matrices, and graphs to present the results in orderthat the relationships among learners of a mobilesocial community are closer to practical situations.Besides being a social network, the ties can influencethe interactions and behaviors of learners. Accordingto the strength of weak ties of sociologist, a weaktie can transfer messages between two groups, andprovide people with resources outside their life circle.A strong tie connects two groups with the samecharacteristic, provides emotional support and socialintegration, and transfers important information, thus,it is likely to form a closed cluster relationship. There-fore, a weak tie is more effective than a strong tieon information transfer. Future studies can conductfurther discussions.

ACKNOWLEDGMENTS

The authors would like to thank the National ScienceCouncil and Science Park Administration of the Re-public of China, Taiwan for supporting this researchunder Contract NSC 101-2628-E-197-001-MY3, 101-2221-E-197-008-MY3 and 101-2219-E-197-004.

REFERENCES

[1] S. Zhao, M.X. Zhou, Q. Yuan, X. Zhang, W. Zheng, and R. Fu,”Who Is Talking about What: Social Map-Based Recommen-dation for Content-Centric Social Websites,” Proc. Fourth ACMConf. Recommender Systems (RecSys ”10), pp. 143-150, 2010.

[2] G. Clough, ”Geolearners: Location-Based Informal Learningwith Mobile and Social Technologies,” IEEE Trans. LearningTechnologies, vol. 3, no. 1, pp. 33-44, Jan.-March 2010

[3] E. Duval, R. Vuorikari, N. Manouselis, and F. Van Assche,Proc. Second Workshop Social Information Retrieval for Technology-Enhanced Learning (SIRTEL ”08), vol. 382, 2008.

[4] J. Vassileva, ”Toward Social Learning Environments,” IEEETrans. Learning Technologies, vol. 1, no. 4, pp. 199-214, Oct.-Dec.2008.

[5] B. Thmos, ”A Dynamic Social Feedback System to SupportLearning and Social Interaction in Higher Education,” IEEETrans. Learning Technologies, vol. 4, no. 4, pp. 340-352, Oct.-Dec.2011

[6] P. De Bra, M. Pechenizkiy, K. van der Sluijs, and D. Smits,”GRAPPLE: Integrating Adaptive Learning into LearningManagement Systems,” Proc. World Conf. Educational Multime-dia, Hypermedia and Telecomm, pp. 5183-5188, June 2008.

[7] P. Brusilovsky and E. Milln, ”User Models for Adaptive Hy-permedia and Adaptive Educational Systems,” The AdaptiveWeb, P. Brusilovsky, A. Kobsa, and W. Nejdl, eds., vol. 4321, pp.3-53, Springer, 2007.

[8] M. Friedrich, M. Wolpers, R. Shen, C. Ullrich, R. Klamma, D.Renzel, A. Richert, and B. von der Heiden, ”Early Results ofExperiments with Responsive Open Learning Environments,”J. Universal Computer Science, vol. 17, no. 3, pp. 451-471, 2011.

[9] B. Hu and P. Moore, ”Context Modelling to Support LocationBased Cooperative Mobile Learning,” Proc. International Conf.Computer Supported Cooperative Work in Design, pp. 1372-1376,May 2006.

[10] B.A. Nardi, D.J. Schiano, and M. Gumbrecht, ”Blogging asSocial Activity, or, Would You Let 900 Million People ReadYour Diary?” Proc. ACM Conf. Computer Supported CooperativeWork, pp. 222-231, 2004.

[11] W. Hung, ”Building Learning Communities by EnhancingSocial Presence: Implementing Blended Instructional DeliveryMethods,” SIGGROUP Bull., vol. 24, no. 3, pp. 79-84, 2003.

[12] C.H. Tu and M. McIsaac, ”The Relationship of Social Presenceand Interaction in Online Classes,” The Am. J. Distance Educa-tion, vol. 16, pp. 131-150, 2002.

[13] Neil Y. Yen, Timothy K. Shih, Louis R. Chao, Q. Jin, ”RankingMetrics and Search Guidance for Learning Object Repository,”IEEE Trans. Learning Technologies, vol. 3, no. 3, pp. 250-264, July-Sept., 2010.

[14] Flora Chia-I Chang, Lun-Ping Hung and Timothy K. Shih, ”ANew Courseware Diagram for Quantitative Measurement ofDistance Learning Course,” J. Information Science and Engineer-ing(JISE), Vol. 19 , No. 6, pp. 989-1014, Nov. 2003.

[15] H. Stern, R. Kaiser, P. Hofmair, P. Kraker, S.N. Lindstaedt, andP. Scheir, ”Content Recommendation in Aposdle Using theAssociative Network,” J. Universal Computer Science, vol. 16,no. 16, pp. 2214-2231, 2010.

[16] H. Drachsler, T. Bogers, R. Vuorikari, K. Verbert, E. Du-val, N. Manouselis, G. Beham, S. Lindstaedt, H. Stern, M.Friedrich, and M. Wolpers, ”Issues and Considerations Re-garding Sharable Data Sets for Recommender Systems inTechnology Enhanced Learning,” Procedia Computer Science,vol. 1, no. 2, pp. 2849-2858, 2010.

[17] K. Verbert, N. Manouselis, X. Ochoa, M. Wolpers, H. Drach-sler, I. Bosnic and E. Duval, ”Context-Aware RecommenderSystems for Learning: A Survey and Future Challenges,” IEEETrans. Learning Technologies, vol. 5, no. 4, pp. 318-335, Oct.-Dec.2012

[18] Adomavicius and A. Tuzhilin, ”Context-Aware RecommenderSystems,” Recommender Systems Handbook: A Complete Guidefor Research Scientists and Practitioners, L. Rokach, B. Shapira,P. Kantor, and F. Ricci, eds., pp. 217-250, Springer, 2011.

[19] Y. Cui and S. Bull, ”Context and Learner Modeling for theMobile Foreign Language Learner,” System, vol. 33, no. 2, pp.353-367, 2005.

[20] R. Bader, E. Neufeld, W. Woerndl, and V. Prinz, ”Context-Aware POI Recommendations in an Automotive Scenario Us-ing Multi-Criteria Decision Making Methods,” Proc. WorkshopContext- Awareness in Retrieval and Recommendation, pp. 23-30,2011.

[21] M. Derntl and K.A. Hummel, ”Modeling Context-AwareELearning Scenarios,” Proc. IEEE Third International Conf. Per-vasive Computing and Comm. Workshops (PERCOMW ”05), pp.337-342, 2005.

[22] Freya H. Lin and Timothy K. Shih, ”Automatic Trap Detection:A Debugging Mechanism for Abnormal Specification in theIMS Sequencing Controls,” IEEE Trans. Learning Technologies,vol. 1, no. 3, pp. 176-189, July-Sept., 2010.

[23] P. Lehsten, R. Zender, U. Lucke, and D. Tavangarian, ”AService- Oriented Approach Towards Context-aware MobileLearning Management Systems,” Proc. IEEE Eighth Interna-tional Conf. Pervasive Computing and Comm. Workshops (PER-COM), pp. 268-273, 2010.

[24] J. Lave and E. Wenger, ”Situated Learning. Legitimate periph-eral participation,” Cambridge: University of Cambridge Press,pp. 40, 1991

[25] Neil Y. Yen, Timothy K. Shih, Qun Jin, Jason C. Hung, QingguoZhou, Louis R. Chao, ”Adaptive Object Re-Ranking Mecha-nism for Ubiquitous Learning Environment,” J. Multimedia, Vol6, No 2, 129-138, April 2011.

[26] S. Berkovsky, T. Kuflik, and F. Ricci, ”Mediation of User Mod-els for Enhanced Personalization in Recommender Systems,”User Modeling and User-Adapted Interaction, vol. 18, no. 3, pp.245-286, 2008.




[27] H. Drachsler, H. Hummel, and R. Koper, ”Personal Recom-mender Systems for Learners in Lifelong Learning Networks:The Requirements, Techniques and Model,” International J.Learning Technology, vol. 3, no. 4, pp. 404-423, 2008.

[28] M.M. El-Bishouty, H. Ogata, and Y. Yano, ”PERKAM: Person-alized Knowledge Awareness Map for Computer SupportedUbiquitous Learning,” Educational Technology and Soc., vol. 10,no. 3, pp. 122-134, 2007.

[29] J. Luo, F. Dong, J. Cao, and A. Song, ”A Context-Aware Per-sonalized Resource Recommendation for Pervasive Learning,”Cluster Computing, vol. 13, pp. 213-239, June 2010.

[30] L. Shih and K. Swan, ”Fostering Social Presence in Asyn-chronous Online Class Discussions,” Proc. Conf. Computer Sup-port for Collaborative Learning, pp. 602-606, 2005.

[31] K.J. Lee, ”What Goes around Comes Around: An Analysisof Del.icio.us as Social Space,” Proc. 20th Anniversary Conf.Computer Supported Cooperative Work, pp. 191-194, 2006.

[32] D.C. Teng, N.-S. Chen, and C.-H. Lee, ”Enhancing EnglishReading Comprehension by Integrating Direct Access to Dig-ital Materials and Scaffolded Questionings in Paper Prints,”Proc. IEEE 11th International Conf. Advanced Learning Technolo-gies (ICALT), pp. 244-248, July 2011.

[33] E. Gray, ”Informal Learning in an Online Community ofPractice,” J. Distance Education, vol. 19, pp. 20-35, 2003.

[34] S. Redfern and N. Naughton, ”Collaborative Virtual Envi-ronments to Support Communication and Community inInternet-Based Distance Education,” J. Information TechnologyEducation, vol. 1, no. 3, pp. 201-211, 2002.

[35] Thoms, N. Garrett, and T. Ryan, ”Online Learning Commu-nities in the New ”U”,” International J. Networking and VirtualOrganisations, vol. 6, no. 5, pp. 499-517, 2009.

[36] C. Prykop and M. Heitmann, ”Designing Mobile Brand Com-munities: Concept and Empirical Illustration,” J. OrganizationalComputing and Electronic Commerce, vol. 6, pp. 301-323, Jun.2011.

[37] J.H. Chen, H.B. Chang, C.Y. Shen, T.H. Wang, W.C. Chang,Timothy K. Shih, ”Immersive Learning Environment withIntegrated Interactive Video and Ubiquitous Technologies,” J.Convergence Information Technology, vol.5, no.9, pp.61-72, 2010.

[38] Passey, ”Mobile Learning in School Contexts: Can TeachersAlone Make It Happen?,” IEEE Trans. Learning Technologies,vol. 3, no. 1, pp. 68-81, March, 2010

[39] R. Beale and P. Lonsdale, ”Mobile Context Aware Systems:The Intelligence to Support Tasks and Effectively Utilise Re-sources,” Mobile Human-Computer Interaction, S. Brewster andM. Dunlop eds., vol. 3160, pp. 573-576, Springer, 2004.

[40] J.S. Donath, ”Identity and Deception in the Virtual Commu-nity,” Communities in Cyberspace, M.A. Smith and P. Kollock, eds.,pp. 29-59, Routledge, 1999.

[41] J.H. Chen, D. T. Ku, T.H. Wang, W.C. Chang, Timothy K. Shih,”Using Adventure Game to Facilitate Learning AssessmentProcess,” J. Convergence Information Technology, vol.5, no.9,pp.126-134, 2010.

[42] M.M. Wasko and S. Faraj, ”Why Should I Share? ExaminingSocial Capital and Knowledge Contribution in Electronic Net-works of Practice,” MIS Quarterly, vol. 29, no. 1, pp. 35-57,2005.

[43] G. Conole and J. Culver, ”The design of cloudworks: Applyingsocial networking practice to foster the exchange of learningand teaching ideas and designs ,” Computers & Education, vol.54, no. 3, pp. 679-692, 2010.

[44] J.S. Huang, S.J. Yang, Y.M. Huang and I.Y. Hsiao, ”SocialLearning Networks: Build Mobile Learning Networks Basedon Collaborative Services,” Educational Technology & Society,vol. 13, no. 3, pp. 78-92, 2010.

[45] C. Liu and M. Divitini, ”Providing Recommendations for Mo-bile Learning: Opportunities and Challenges from Context,”Educational Recommender Systems and Technologies: Practices andChallenges, Chapter 13, pp. 302-328, 2012.

[46] C.H. Tua, M. McIsaacb, L.S. Montesa and S. Armfielda, ”Isthere a mobile social presence?,” International Educational Me-dia, vol. 49, no. 4, pp. 247-261, 2012.

[47] Y.M. Huang, P.S. Chiu, T.C. Liu and T.S. Chen, ”The designand implementation of a meaningful learning-based evalua-tion method for ubiquitous learning,” Computers & Education,vol. 57, no. 4, pp. 2291-2302, 2011.

[48] Y.M. Huang, T.H. Liang, Y.N. Su, and N.S. Chen, ”Empower-ing personalized learning with an interactive e-book learningsystem for elementary school students,” Educational TechnologyResearch and Development, vol. 60, no. 4, pp. 703-722, 2012.

[49] M.C. Liu, Y.M. Huang, Kinshuk and D. Wen, ”Fostering learn-ers’ metacognitive skills of keyword reformulation in imageseeking by location-based hierarchical navigation,” EducationalTechnology Research and Development, vol. 61, no. 2, pp. 233-254,2013.

[50] Y.M. Huang and T.T. Wu, ”A Systematic Approach for LearnerGroup Composition Utilizing U-Learning Portfolio,” Educa-tional Technology & Society, vol. 14, no. 3, pp. 102-117, 2011.

Han-Chieh Chao Han-Chieh Chao is a jointappointed Full Professor of the Departmentof Electronic Engineering and Institute ofComputer Science & Information Engineer-ing where also serves as the president of Na-tional Ilan University, I-Lan, Taiwan, R.O.C.He has been appointed as the Director of theComputer Center for Ministry of Educationstarting from September 2008 to July 2010.His research interests include High SpeedNetworks, Wireless Networks, IPv6 based

Networks, Digital Creative Arts and Digital Divide. He receivedhis MS and Ph.D. degrees in Electrical Engineering from PurdueUniversity in 1989 and 1993 respectively. He has authored or co-authored 4 books and has published about 280 refereed professionalresearch papers. He has completed 100 MSEE thesis students and3 PhD students. Dr. Chao has received many research awards,including Purdue University SRC awards, and NSC research awards(National Science Council of Taiwan). He also received many fundedresearch grants from NSC, Ministry of Education (MOE), RDEC,Industrial Technology of Research Institute, Institute of InformationIndustry and FarEasTone Telecommunications Lab. Dr. Chao hasbeen invited frequently to give talks at national and internationalconferences and research organizations. Dr. Chao is the Editor-in-Chief for IET Networks, Journal of Internet Technology, InternationalJournal of Internet Protocol Technology and International Journalof Ad Hoc and Ubiquitous Computing. Dr. Chao has served asthe guest editors for Mobile Networking and Applications (ACMMONET), IEEE JSAC, IEEE Communications Magazine, ComputerCommunications, IEE Proceedings Communications, the ComputerJournal, Telecommunication Systems, Wireless Personal Commu-nications, and Wireless Communications & Mobile Computing. Dr.Chao is an IEEE senior member and a Fellow of IET (IEE). He is aChartered Fellow of British Computer Society.

Chin-Feng Lai Chin-Feng Lai is an assis-tant professor at Department of ComputerScience and Information Engineering, Na-tional Chung Cheng University since 2013.He received the Ph.D. degree in departmentof engineering science from the NationalCheng Kung University, Taiwan, in 2008. Hisresearch interests include multimedia com-munications, sensor-based healthcare andembedded systems. After receiving the Ph.Ddegree, he has authored/co-authored over

100 refereed papers in journals, conference, and workshop proceed-ings about his research areas within four years. Now he is makingefforts to publish his latest research in the IEEE Transactions onmultimedia and the IEEE Transactions on circuit and system on videoTechnology. He is also a member of the IEEE as well as IEEE Circuitsand Systems Society and IEEE Communication Society.




Shih-Yeh Chen Shih-Yeh Chen receivedthe M.S. degree in Master’s Program of E-Learning, Taipei Municipal University of Edu-cation, Taipei, Taiwan in 2012. He is currentlyworking toward the Ph.D. degree in engi-neering science from National Cheng KungUniversity, Tainan, Taiwan. His main researchinterests are in e-learning education, SocialInteractions and Mobile Networking.

Yueh-Min Huang Dr. Yueh-Min Huang is aDistinguished Professor of the Department ofEngineering Science, National Cheng-KungUniversity, Taiwan. He was also chairing thedepartment from 2006 to 2009. His researchinterests include e-Learning, embedded sys-tems, and artificial intelligence. He receivedhis MS and Ph.D. degrees in Electrical En-gineering from the University of Arizona in1988 and 1991 respectively. He has co-edited 3 books published by Springer Verlag

and has published over 180 refereed journal papers. Dr. Huang hasreceived many research awards, for instance, in recent 3 years,such as the Best Paper Award of HumanCom 2010, ICS2009,TANET2009 and the distinguished research award from the NationalScience Council, Taiwan in 2010. Dr. Huang is in the editorialboard of over 10 SSCI- and SCI-indexed journals such as Interac-tive Learning Environments, Wireless Communications and MobileComputing, Journal of Internet Technology, International Journalof Communication Systems, International Journal of Ad Hoc andUbiquitous Computing, etc. He was the technical program chair ofACM MDLT2009, ICWL (2013). He was the general chair of VIP2007,IEEE EPSA2008, PCM2008, ICST2008, ICS2010, AECT-Taiwan-Symposium2013, HumanCom2013. Dr. Huang is a Fellow of BritishComputer Society (FBCS) and senior member of the IEEE.

Documents

A M-Learning Content Recommendation Service by Exploiting Mobile Social Interactions