Improve Performance of Recommender System in Collaborative

Learning Environment based on Learner Tracks

Qing Tang

, Marie-Hélène Abel

and Elsa Negre

Sorbonne Universités, UTC, CNRS UMR 7253, HEUDIASYC, 60200 Compiègne, France

Paris-Dauphine University, PSL Research University, CNRS UMR 7243, LAMSADE, 75016 Paris, France

Keywords: Online Learning, SoIS, Collaborative Learning Environment, Recommender System, Learner Track.

Abstract: Learning with huge amount of open educational resources is challenging, especially when variety resources

come from different System of Information Systems (SoIS). How to help learners obtain appropriate resources

efficiently in collaborative learning environment is still a rigorous problem of research. This paper proposes

a method to calculate learner’s knowledge competency by tracking and analyzing their behaviors in a

collaborative learning environment based on SoIS, and combining other basic learner’s information to build

a personalized recommender system to help learners select appropriate educational resources to improve their

learning efficiency.

1 INTRODUCTION

With the development of Internet information

technology, human society has stepped into an era of

information overload. Owing to the overwhelming

quantity of information, both information providers

and consumers are facing challenges: information

providers are willing to find the information to be

transferred to the target audience while information

consumers are willing to find the information most

relevant to their needs (Wang, 2016).

In a collaborative learning environment of System

of Information Systems (SoIS), acquiring educational

resources from appropriate channels could also be

much tougher (Wang, 2016). In such an environment,

heterogeneous educational resources is collected

from separate systems (Saleh and Abel, 2018). When

faced with so many heterogeneous resources, learners

have information burden problem and difficulty in

resource decision-making. We want to help those

learners who use SoIS-based collaborative learning

environment for learning to efficiently obtain the

most suitable educational resources for their current

learning process.

Recommender systems in online learning is a

branch of information retrieval where learning

resources are filtered and presented to the learners

(Chughtai et al., 2014). However, how to improve the

accuracy of the recommender system is still a

direction worth researhing. This paper proposes a

scheme for calculating learner's knowledge level

from their historical learning tracks in collaborative

learning environment of SoIS, and combining with

basic information (e.g., degree, profession,

preference, etc.) to build a personalized recommender

system. This recommender system is based on

learners interaction tracks and semantic description,

it can help deliver appropriate educational resources

to target learners, thereby reducing information

burden and difficulty in decision-making and

improving their learning efficiency.

The structure of this article is as follows: Section

2 reveals the background and technologies, Section 3

explains the proposed method and technical support,

Section 4 discusses the core ideas and shortcomings,

Section 5 makes a summary and expressed the

perspective for future work.

2 BACKGROUND

This section introduces the three core theoretical parts

of this paper: SoIS, Collaborative Learning

Environment, and Recommender System. And gives

a framework to provide theoretical support.

270

Tang, Q., Abel, M. and Negre, E.

Improve Performance of Recommender System in Collaborative Learning Environment based on Learner Tracks.

DOI: 10.5220/0010214702700277

In Proceedings of the 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2020) - Volume 3: KMIS, pages 270-277

ISBN: 978-989-758-474-9

2.1 System of Information Systems

System of Systems (SoSs) are large-scale integrated

systems which are heterogeneous and independently

operable on their own but are networked together for

a common goal (Carlock and Fenton, 2001)

(Jamshidi, 2011). SoIS is a special type of SoS in

information domain. SoIS are networks of agents

interacting in a specific technology area under a

particular institutional infrastructure for the purpose

of creating, diffusing, and utilising technology

focused on knowledge, information, and competence

flow (Carlsson and Stankiewicz, 1991). SoIS are the

specific clusters of the firms, technologies, and

industries involved in the generation and diffusion of

new technologies and in the knowledge flow that

takes place among them (Breschi and Malerba, 1996).

Based on the definitions provided, SoIS has such

features (Saleh and Abel, 2018):

 SoIS addresses impact of the interrelationships

between different Information Systems (ISs).

 SoIS is concerned with the flow of information

and knowledge among different ISs.

 SoIS is responsible for generating information

from the constituent ISs.

 Information interoperability is a key issue

when designing and implementing an SoIS.

2.2 Collaborative Learning

Environment

Collaborative learning is an educational approach to

teaching and learning that involves groups of students

working together to pursue a same learning goal,

solve a learning problem, or complete a learning task

(Vijayalakshmi and Kanchana, 2020). Collaborative

learning environment is a community to support

group members’ coordination and interaction so that

they complete the task more efficiently, due to the

way of group learning facilitates a comfort

communication between students to share their

resources, discuss their problems and receive the

appropriate solutions (Riyahi and Sohrabi, 2020). A

collaborative learning environment is an

interconnected virtual place of people, systems, and

resources to support learners by using multiple tools

to access educational resources (Gütl and Chang,

2008). Collaborative learning environment consists

of biotic parts which are the learning community

(e.g., teachers, learners, etc.) working together with

abiotic parts which are the learning utilities (e.g.,

technologies, Information Systems, etc.) (Álvarez-

Arregui et al., 2017). The elements of collaborative

learning environment are remix of different forms of

technologies, devices, data repositories, information

retrieval, information sharing, networks and

communication (O’Connell, 2016).

Collaborative learning environment has many

advantages, learners form groups, collaborate with

each other and with educators, and content designed

for interaction (Ouf et al., 2017). Learners do not

evolve alone as single individual, but in a learning

environment that includes the learners and their

physical and social equipments: tools (e.g., notepad,

tablet, etc.), resources (e.g., procedures, methods,

instructions, course materials, notes, document, etc.),

and the partners (e.g., teachers, network of experts,

work colleagues, etc.) (Perkins, 1995). And this

environment can be seen as a virtual learning space in

which technologies that contribute to learning (e.g.,

hardware, software, network, etc.) are used to foster

interactions between communities of actors and

content. Learners who share and use resources and

knowledge information about common interests, and

variety of educational resources are accessible

through the learner’s own memory as well as through

their tools or partners (Saleh and Abel, 2018).

2.3 Recommender System

Recommender systems can be classified according to

three approaches: score estimation method, the data

used to estimate scores or the main objective of the

system (Negre, 2015). Whatever recommendation

technique is used, certain information needs to be

considered in relation to users and this kind of

information usually store in user’s profiles, and the

common algorithms mainly are Content Based,

Collaborative Filtering, Knowledge Based, and

hybrid approaches (Negre, 2015). Fanaeetork and

Yazdi (2013) proposed a Content Based method

based on vector space model which the users’ profiles

are enriched using ontologies, the ontologies are

made by combining the text mining and NLP (Natural

Language Processing) techniques. Yoldar and Özcan

(2019) proposed a Collaborative Filtering method on

an online ad dataset, which is based on bi-clustering

and ordered weighted average aggregation operators,

can address situations such as the lack of implicit

feedback on items. Paradarami et al. (2017) present a

Hybrid method with a deep learning neural network

framework that utilizes reviews in addition to

content-based features to generate model based on

predictions for the business-user combinations.

The application of recommender systems in

online learning has become a thriving research field

(Pan et al., 2010). The task of a recommender system

in online learning is to recommend relevant learning

Improve Performance of Recommender System in Collaborative Learning Environment based on Learner Tracks

271

materials to the students and help them in decision

making (Aguilar et al., 2017). Zheng et al. (2015)

proposed a recommendation approach to mitigate

learning issues in online learning communities. Chen

et al. (2014) proposed a recommender system to

recommend learning materials in an online learning

platform and their results demonstrated significant

improvement in performance. Takano and Li (2010)

proposed a recommender system for online learning

by utilizing a feedback method that extracts student’s

preference and web-browsing behavior.

Experimental results of previous works show that

using recommender systems on online learning

community obtain significant achievements (Kardan

and Ebrahimi, 2013).

2.4 Framework

Many learners now choose to join the online learning

platform while completing their social studies. For

most of the current situation, many learners join the

online learning platform with great interest, however

many of them are unable to persist due to the

information burden of a large number of resources

and the lack of precise recommendation assistant. We

aim at helping such learners. From the previous

sections, SoIS can provide massive and diverse

learning resources for collaborative learning

environment, and the recommender system can solve

the information burden problem encountered by

learners in collaborative learning environment.

Therefore, we propose to provide learners with a

complete learning framework, which contains a

collaborative learning platform based on SoIS and a

recommender system associated with the learner's

track. Such a learning framework can not only

integrate diversified resources and enable learners to

learn in a collaborative manner, but also provide

accurate recommendations to maintain learner’s

enthusiasm for learning.

The structure of framework: An online learning

platform based on SoIS, learners are divided into

different learning groups according to their learning

goals, each group has a certain study subject. A

learner can choose to join one or more study groups

according to their learning goals. Every learner has

the authority to upload and share learning resources

on the platform, and these resources can come from

other different resource systems. And learners can

search for the resources they want in the platform;

learners can discuss with other learners in the same

group; learners can give scores to used resources.

https://xapi.com/

3 THE PROPOSED METHOD

This paper focuses on improving the performance of

recommender system of a collaborative learning

environment in SoIS, to help learners make decisions

and reduce their information burden from massive

heterogeneous educational resources by

recommending appropriate educational resources,

and ultimately improves their learning efficiency. It

proposes a method to record and analyze learner’s

historical learning behaviours, calculate their

knowledge level. Then, build a learner model, and use

the vector space model to transform it into a

multidimensional space (Sivaramakrishnan et al.,

2018), and make personalized recommendations by

calculating the cosine similarity of learner vectors

(Geng et al., 2018).

3.1 Tracking System

When learners complete learning actions in

collaborative learning environment, they leave

‘digital tracks’, almost all interactions in the past

represent tracks and can be regarded as learner’s

study experience (Wang, 2016). The information

extracted from these tracks can help in many cases

(e.g., decision making, recommendation, etc.).

Reiner et al. (2001) proposed a complete set of

plans to record and analyze network user behaviour,

and visualize the data for easy analysis. Benevenuto

et al. (2009) used web crawlers to obtain user

behaviour data from websites for analysis.

Alexandros and Georgios (2013) proposed a

framework for recording, monitoring and analyzing

learner behaviour while watching and interacting

with online educational videos. Although these

methods have been proven effective, they are not

efficient enough and some methods are difficult to

achieve. We propose to use xAPI

, which has a

complete and efficient method, to help recommender

system collect and organize the behaviours of learners

in a collaborative learning environment.

xAPI is a standard established by the US

Department of Defense and the White House Office

of National Science and Technology Policy in the

Advanced Decentralized Learning program, which is

a new specification for learning technology that

makes it possible to collect tracks data about the wide

range of experiences a person has (online and offline).

People learn from interactions with other people,

content, and beyond. These actions can happen

anywhere and signal an event where learning could

KMIS 2020 - 12th International Conference on Knowledge Management and Information Systems

272

occur. All of these can be recorded with xAPI. When

an activity needs to be recorded, the application sends

secure statements in the form of 'Noun, verb, object'

or 'I did this' to a Learning Record Store (LRS). LRS

in xAPI has two types of APIs:

 Statement API for inputting and outputting the

statement;

 Document APIs for access richer information

(e.g., strings, word documents, pictures,

videos, etc.).

Both types of APIs follow the RESTful

architecture, enabling data in LRS to be processed by

HTTP, including adding, deleting, querying, and

modifying. LRS provides a variety of attributes

choose from, as shown in Table 1.

Table 1: Attributes of LRS.

Property Description Required

id assigned by LRS if not set

by the learning record

provider.

Recommended

actor whom the Statement is about Required

verb action taken by the actor. Required

object

activity, agent, or another

statement

Required

result

representing a measured

outcome.

Optional

context

context that gives the

Statement more meaning.

Optional

timestamp

timestamp of when the

events described within this

statement occurred. Set by

the LRS if not provided.

Optional

stored

timestamp of when this

statement was recorded.

Set by LRS

authority

agent or group who is

asserting this Statement.

Optional

version

the statement’s associated

xAPI version.

Not

Recommended

attachment

headers for attachments to

the statement

Optional

In addition to the three required options (‘actor’,

‘verb’, ‘object’), here are three additional options,

‘id’, ‘timestamp’, and ‘authority’, corresponding to

the elements in Table 1. After the system connected

to the API of LRS and set other parameters well, LRS

will record every activity generated by students in the

collaborative learning environment. Simply put, no

matter any learner, as long as he/she accesses the

platform server with LRS, all his/her behaviours will

be recorded in the form: Track= [id, timestamp, actor,

object, authority], which can be interpreted as: [track

https://github.com/adlnet/xAPI-Spec/

number] [time] [who] [action] [target] [where],

corresponding to the elements in Table 1.

3.2 Learner Model

The function of the learner model is to standardize

learner's information, including the learner's basic

information and the knowledge level extracted from

the chaotic and disorderly behaviour tracks obtained

in Section 3.1.

3.2.1 Knowledge Level

Knowledge level indicates the learner's knowledge

reserve and ability value in a certain field, which is an

important basis for recommendation. According to

the learner tracks collected in Section 3.1, the

necessary information (various actions performed by

learners in designated groups) can be extracted.

Each learner and each group in a collaborative

learning environment has a unique ID. To calculate

the knowledge level of a learner in a designated

group, the necessary elements are shown in Table 2.

Table 2: Elements of knowledge level calculation.

Learner

Group

Time

span

Shared Used Reshared

























Learner ID, the learner's number in the global

collaborative learning environment; Group ID, the

group's number in the global collaborative learning

environment; Time span, the time this learner has

been in designated group; Shared, the total number of

resources shared by this learner in designated group;

Used, the total number of resources (uploaded by this

learner) used by other learners; Reshared, the total

number of resources (uploaded by this learner)

reshared by other learners.





indicate the number of

times.

The calculation rule of the learner's knowledge

level is shown in Equation 1:

=ts*w

+st*w

+ut*w

+rt*w

)

represents the knowledge level of l

in group

. g

represents the j-th learning group, and l

represents the i-th student.ts represents time span of

the learner have been joined in designated group; st

represents shared times completed by learner l

; ut

represents used times, other learners downloaded the

resources l

uploaded, which shows that the quality of

Improve Performance of Recommender System in Collaborative Learning Environment based on Learner Tracks

273

his/her resources has been recognized by others, and

this is also a reflection of high knowledge level; rt

represents reshared times, other learners are willing

to share again, indicating that this resource is indeed

of high quality. w

, w

, and w

are weights,

which use to emphasize the importance of each

element, defined according to the actual situation.

3.2.2 Learner Profile Standard

Ali Ben Ameur et al. (2017) proposed a learner model

contains several attributes (e.g., name, age,

preference, competency, etc.) of learner, however, the

model contains too many dimensions for which

specific precise values cannot be given. Here,

redefine the learner profile standard, each dimension

can reflect the important features of the learner, and

the value of each dimension can be accurately

defined, as shown in the following Table 3.

Table 3: Learner profile standard.

ID Composition Dimension Value

Gender G 1,2

Degree D 1,2,3,4,5

Preference P 1,2,3,4,5

English level E 1,2,3,4,5

Knowledge level K Kl

Define an execution standard for each dimension

to prepare for the next step of learner vectorization:

Gender (Male: 1, Female: 2), Degree (Elementary: 1,

Bachelor: 2, Master: 3, Doctor: 4, Others: 5),

Preference (Scientific papers: 1, Electronic books: 2,

Video courses: 3, Report documents: 4, Others: 5),

EnglishLevel (Expert: 1, Proficient: 2, Competent: 3,

Advanced beginner: 4, Novice: 5), and Knowledge

level is calculated from Equation 2.

3.3 Learner Vectorization

This Section uses the method of Vector Space Model

(VSM

). VSM can help to project the learner's

features that the computer cannot recognize into the

multi-dimensional vector space, making the learner

recognizable by the computer. C.Peterson et al.

(2020) proposed to use a vector space model to

predict human relational similarity. Based on VSM,

W.Sholikah et al. (2020) proposed method was

designed to construct a general vector space and

semantic relation identification, and the results show

that the use of multi-task learning with a general

https://www.sciencedirect.com/topics/computer-

science/vector-space-models

vector space can overcome the problem of cross-

lingual semantic relation identification.

According to the implementation standard in

Section 3.2.2, learner can be projected into a 5-

dimensional vector space, and the dimension is

expressed as Equation 2:

=(G, D, P, E, K)

)

is the vector of l

in group g

. g

represents the

j-th learning group, and l

represents the i-th student.

The components (G, D, P, E, K) correspond to the

dimension code in Table 3.

All learners in designated group can be expressed

as a vector set after vectorization, shown as Equation

= v

,…,v



)

Since the state of the learner is changing at each

time point, for example, after the learner has

completed the learning behaviour, their knowledge

level will improve. So, it is necessary to specify the

time point t when building the vector set.

3.4 Similarity and Recommendation

This part is mainly about similarity calculation and

how to recommend to target learners based on

similarity results.

3.4.1 Similarity Calculation

Similarity is used to measure the common

characteristics between two instances, and distance is

adopted to indicate the differences between them.

Many tasks, such as classification and clustering, can

be accomplished perfectly when a similarity metric is

well-defined (Xia et al., 2015). Many similar

measures have been proposed by researchers, such as

Pearson Similarity (Lü and Zhou, 2011), Jaccard

Similarity (Jalili et al., 2018) and Euclidean Distance-

based Similarity (Hawashin et al., 2019). Cosine

similarity is a widely used metric that is both simple

and effective (Xia et al., 2015). After transforming in

section 3.3 to obtain the learner feature vector set, the

cosine similarity algorithm can efficiently and

conveniently calculate the similarity between any two

learners.

Su et al. (2020) define a plan to calculate vector

similarity as indicators of similarity between users,

KMIS 2020 - 12th International Conference on Knowledge Management and Information Systems

274

revealing and measuring the difference between

users’ general preferences in different scenarios.

Cosine similarity can intuitively show the similarity

between two vectors with same dimensions.

When recommending resource to the first learner

in group g

at time t, recommender system needs

to calculate the cosine similarity between l

with other

members l

. Here use (A, B) to represent the two

learner vectors (l

, l

), 5 dimensions in two vectors

correspond to the dimension code in formula 2. The

calculation is shown as Equation 4:

Similarity



A,B



∑

×B

d=1



∑



d=1



∑



d=1

(4)

3.4.2 Recommendation

The similarity between feature vectors is an important

basis for recommender system to measure the

similarity of learners, because the feature vectors are

transformed from the features of the learners, and the

features of the learners are an important manifestation

of the current state of the learners. If two learners

have a high degree of similarity, it proves that their

learning situations are very similar, so the resources

they currently need are very similar. Recommender

system can recommend resources based on this

principle.

According to the cosine similarity calculated in

the previous section, recommender system can

recommend resources to learners who study in the

same group. The processes are:

 If Learner A and Learner B have a highest

similarity, from the resources Learner B has

shared recently, select the latest resource and

recommend that resource to Learner A.

 If none of the resources Learner B has shared

recently, then, from the other Learner C in the

group with the second highest similarity with

Learner A, choose the resources that Learner C

has shared recently, select the latest resource

and recommend to Learner A.

 If there is still no result, continue from the

second step, until find a qualified resource.

Simply put, the basis of this recommendation

method is that learners with similar features will

select similar resources.

https://www.coursera.org/

https://dzone.com/

4 DISCUSSION

In this article, our theoretical basis is building a

precise recommender system in collaborative online-

learning environment of SoIS based on learner’s

tracks, to reduce the burden of learners to find

suitable resources and improve their learning

efficiency. Unlike many existing online learning

environments, such as Coursera

, DZone

, the

environment we designed has a personalized

recommender system is based on learners internal

(knowledge level) and external (gender, degree,

preference, and language) information, which means

that the recommender system is personalized. The

system uses the xAPI standard protocol to collect and

record learner’s historical learning behaviours in

collaborative online-learning environment. At this

step, for the recommender system, knowing the

knowledge level of the learner is very important for

recommending suitable resources. The system will

extract information from these historical learning

behaviours to calculate learner’s knowledge level and

form learner models. Then the learner model will be

projected to the multi-dimensional vector space

through the vector space model, so that the learner

becomes comparable. Finally, the similarity between

learners is judged by calculating the similarity

between the vectors, so as to make resource

recommendations to each other learners in the same

group.

From the perspective of calculation method, we

chose the method of calculating the similarity of

learners, which considers the features of learners and

the connections between learners, but it didn’t take

the resource features and the result feedbacks

(attitudes of learners after using resources) in to

consideration. In many cases, the features of

resources are also important factors that affect the

choice of learners, and the interactive results

feedbacks between learners and resources are

important indicators that reflect whether the resources

meet the learners’ requirements. If we collect the

result feedbacks generated between different learners

and different resources, separate the positive result

feedbacks from the negative result feedbacks, and

form a data set, the direction can be transformed into

supervised machine learning. Thus, it will be possible

to study learners with ‘some features’ who used

resources with ‘some features’ and come out the

conclusion of ‘positive’ or ‘negative’. These data

could be recorded and used as training data set for

Improve Performance of Recommender System in Collaborative Learning Environment based on Learner Tracks

275

supervised learning in machine learning to train a

machine learning model, this model is familiar with

the possible results of different learners and different

resources. For the target learner, the model can judge

which resources combined with him/her will produce

positive results, and then recommend these resources

to the him/her.

5 CONCLUSIONS AND FUTURE

WORK

This paper proposed a method to improve the

performance of recommender system in collaborative

learning environment based on learner tracks.

Although similar research has already existed, for

example, Kanoje et al. (2016) used Data Mining and

Information Retrieval technologies to obtain user

information and use this to build a recommender

system; Pan et al. (2020) proposed a dynamic user

profile called User profile refactoring (builds a

dynamic user profile and determines the weights of

the extended tags in the profile.), combined with Tag

cloud generation (discovers potentially relevant tags

in an application domain) and Tag expansion (finds a

sufficient set of tags upon original tags) to build a

recommender system. However, none of these

methods mentioned the concept of extracting

information from the learner's track, because a lot of

valuable information is often reflected in the

behaviour. Therefore, this article proposes to obtain

internal information (knowledge level) from track

and combine the basic information (e.g., gender,

degree, language skills, etc.) to make the

recommendations has good interpretability and

personalization.

To test our work, we chose the MEMORAe

collaboration platform as our experiment

environment. This web platform already has the

expected functions: It has the theoretical structure of

the collaborative learning environment we mentioned

before; the semantic collaboration model makes it

possible to track learner activities (e.g., sharing,

voting, etc.). It was tested as collaborative learning

platform (Abel, 2015). Install and set xAPI on the

web server, it will provide the learner's track

information to the recommender system. Then the

recommender system calculates the resources that

each student may need at each moment and feeds it

back to the learning platform. And as we have

mentioned in the discussion section, in the later work,

this article can be considered as a priori step of

http://memorae.hds.utc.fr/demo/labo/

supervised machine learning, and it is very promising

to train the machine learning model as the decision-

making core of the recommender system.

REFERENCES

Wang, N., 2016. Towards a competency recommender

system from collaborative traces. Other [cs.OH].

Université de Technologie de Compiègne.

Saleh, M. and Abel, M-H., 2018. System of Information

Systems to support learners (a case study at the

University of Technology of Compiègne). Behaviour &

Information Technology.

Chughtai, M.W., Selamat, A., Ghani, I. and Jung, J.J., 2014.

Retracted: E-Learning recommender systems based on

goal-based hybrid filtering. International Journal of

Distributed Sensor Networks. Vol. 10 (7), pp. 912130.

Carlock, P.G., and Fenton, R.E., 2001. System of Systems

(SoS) Enterprise Systems Engineering for Information-

Intensive Organizations. Systems Engineering. Vol. 4

(4), pp. 242–261.

Jamshidi, M., 2011. System of Systems Engineering:

Innovations for the Twenty-first Century. New Jersey:

John Wiley & Sons. Vol. 58.

Carlsson, B. and Stankiewicz, R., 1991. On the Nature,

Function and Composition of Technological Systems.

Journal of Evolutionary Economics. Vol. 1(2), pp. 93–

118.

Breschi, S. and Malerba, F., 1996. Sectoral Innovation

Systems: Technological Regimes, Schumpeterian

Dynamics and Spatial Boundaries. milano: Centro studi

sui processi di internazionalizzazione. Università

commerciale’Luigi Bocconi’.

Vijayalakshmi, D. and Kanchana, J., 2020. Evaluating the

effectiveness of the collaborative learning in fashion

studies. Procedia Computer Science. Vol. 172, pp. 991-

1000.

Riyahi, M. and Sohrabi, M.K., 2020. Providing effective

recommendations in discussion groups using a new

hybrid recommender system based on implicit ratings

and semantic similarity. Electronic Commerce Research

and Applications. Vol. 40, pp. 100938.

Gütl, C. and Chang, V., 2008. Ecosystem-based theoretical

models for learning in environments of the 21st century.

International Journal of Emerging Technologies in

Learning (iJET). Vol. 3.

Álvarez-Arregui, E., Rodríguez-Martín, A., Madrigal-

Maldonado, R., Grossi-Sampedro, B-Á., and Arreguit,

X., 2017. Ecosystems of Media Training and

Competence: International Assessment of its

Implementation in Higher Education/Ecosistemas de

formación y competencia mediática: Valoración

internacional sobre su implementación en la educación

superior. Comunicar. Vol. 25 (51), pp. 105–114.

O’Connell, J., 2016. Networked Participatory Online

Learning Design and Challenges for Academic Integrity

KMIS 2020 - 12th International Conference on Knowledge Management and Information Systems

276

in Higher Education. International Journal for

Educational Integrity. Vol. 12 (1): 4.

Ouf, S., Abd Ellatif, M., Salama, S.E., and Helmy, Y., 2017.

A Proposed Paradigm for Smart Learning Environment

Based on Semantic Web. Computers in Human

Behavior. Vol. 72, pp. 796–818.

Perkins, D.N., 1995. L’individu-plus Une Vision Distribuée

de la Pensée et de l’Apprentissage. Revue française de

pédagogie. Vol. 111, pp. 57–71.

Negre E., 2015. Information and Recommender Systems.

Wiley Online Library. Vol 4.

FanaeeTork, H. and Yazdi, M., 2013. A semantic VSM-

based recommender system. Int. J. Computer Theory

Eng. Vol. 5 (2), pp. 331-336.

Yoldar, M.T. and Özcan, U., 2019. Collaborative targeting:

Biclustering-based online ad recommendation Electron.

Commer. Res. Appl. Vol. 35, Article 100857.

Paradarami, T.K., Bastian, N.D., and Wightman, J.L., 2017.

A hybrid recommender system using artificial neural

networks. Expert Syst. Appl. Vol. 83, pp. 300-313.

Pan, P.Y., Wang, C.H., Horng, G.J. and Cheng, S.T., 2010.

The development of an ontology based adaptive

personalized recommender system. In: ICEIE 2010–

2010 Int. Conf. Electron. Inf. Eng. Proc. Kyoto, Japan,

pp. 76–80.

Aguilar, J., Valdiviezo-Díaz, P. and Riofrio, G., 2017. A

general framework for intelligent recommender

systems. Applied Computing and Informatics. Vol. 13

(2), pp. 147-160.

Zheng, X.L., Chen, C.C., Hung, J.L., He, W., Hong F.X.

and Lin, Z., 2015. A hybrid trust-based recommender

system for online communities of practice. IEEE Trans.

Learn. Technol. Vol. 8, pp. 345–356.

Chen, W., Niu, Z., Zhao, X. and Li, Y., 2014. A hybrid

recommendation algorithm adapted in e-learning

environments. World Wide Web. Vol. 17, pp. 271–284.

Takano, K. and Li, K.F., 2010. An adaptive e-learning

recommender based on user’s webbrowsing behavior.

In: Proc. - Int. Conf. P2P, Parallel, Grid, Cloud Internet

Comput. pp. 123–131.

Kardan, A.A. and Ebrahimi, M., 2013. A novel approach to

hybrid recommendation systems based on association

rules mining for content recommendation in

asynchronous discussion groups. Inf. Sci. Vol. 219, pp.

93–110.

Sivaramakrishnan, N., Subramaniyaswamy, V.,

Arunkumar, S. and Soundarya Rathna, P., 2018.

VALIDATING EFFECTIVE RESUME BASED ON

EMPLOYER’S INTEREST WITH

RECOMMENDATION SYSTEM. International

Journal of Pure and Applied Mathematics, Academic

Publishing Ltd. Vol. 119 (12e), pp.13261-13272.

Geng, Z.Q., Li, Y.N., Han, Y.M., Zhua, Q.X., 2018. A

novel self-organizing cosine similarity learning

network: An application to production prediction of

petrochemical systems. Energy, Vol. 142, pp. 400-410.

Reiner, D., Tan, M., Ventikos, P., Richard, E., 2001.

System and method for logical view analysis and

visualization of user behavior in a distributed computer

network. United States Patent.

Benevenuto, F., Rodrigues, T., Cha, M., Almeida, V., 2009.

Characterizing user behavior in online social networks.

Proceedings of the 9th ACM SIGCOMM conference on

Internet measurement. pp. 49–62.

Alexandros, K., and Georgios, E., 2013. A Framework for

Recording, Monitoring and Analyzing Learner

Behavior while Watching and Interacting with Online

Educational Videos, 2013 IEEE 13th International

Conference on Advanced Learning Technologies,

Beijing. pp. 20-22.

Ali Ben Ameur, M., Saleh, M., Abel, M-H. and Negre, E.,

2017. Recommendation of Pedagogical Resources

within a Learning Ecosystem. 9th International

Conference on Management of Digital EcoSystems

(MEDES ’17). Nov 2017, Bangkok, Thailand. pp.14-21.

C.Peterson, J., Chen, D., and L.Griffiths, T., 2020.

Parallelograms revisited: Exploring the limitations of

vector space models for simple analogies. Cognition.

Vol. 205, December 2020, 104440.

W.Sholikah, R., Z. Arifin, A., Fatichah, C., and

Purwarianti, A., 2020. Multi task learning with general

vector space for cross-lingual semantic relation

detection. Journal of King Saud University - Computer

and Information Sciences.

Xia, P.P., Zhang, L., Li, F.Z., 2015. Learning similarity

with cosine similarity ensemble. Information Sciences.

Vol. 307, pp. 39–52.

Lü, L., Zhou, T., 2011. Link prediction in complex

networks: A survey. Physica A. Vol. 390 (6), pp. 1150–

1170.

Jalili, M., Ahmadian, S., Izadi, M., Moradi, P., and Salehi,

M., 2018. Evaluating collaborative filtering

recommender algorithms: A survey. IEEE Access. Vol.

6, pp. 74003–74024.

Hawashin, B., Lafi, M., Kanan, T., and Mansour, A.,

2019. An efficient hybrid similarity measure based on

user interests for recommender systems. Expert Syst.

e12471.

Su, Z., Zheng, X.L., Ai, J., Shen, Y.M., and Zhang, X.X.,

2020. Link prediction in recommender systems based on

vector similarity. Physica A: Statistical Mechanics and

its Applications. Vol. 560, 15 December 2020, 125154.

Kanoje, S., Mukhopadhyay, D., and Girase, S., 2016. User

Profiling for University Recommender System Using

Automatic Information Retrieval. Procedia Computer

Science. Vol. 78, 2016, pp. 5-12.

Pan, Y.H., Huo, Y.F., Tang, J., Zeng Y.F., and Chen, B.L.,

2020. Exploiting Relational Tag Expansion for

Dynamic User Profile in a Tag-aware Ranking

Recommender System. Information Sciences. Available

online 18 September 2020.

Abel, M.H., 2015. Knowledge map-based web platform to

facilitate organizational learning return of experiences.

Comput. Hum. Behav. Vol. 51, pp. 960-966.

Improve Performance of Recommender System in Collaborative Learning Environment based on Learner Tracks

277