-E-Learning students can benefit from proper class formation process based on the student needs. In particular, Online Social Networks make available data concerning users' interactions, as skills and trust relationships, that are behind the dynamics of thematic social network groups, and can be explouted to form e-Learning classes. To this aim, we propose a model based on such information, which are properly combined to support the dynamics of e-Learning classes on Online Social Networks. The approach provide a way to give suggestions to users about the best classes to join with and to class adminastrors the best students to accept. The proposed approach has been tested by simulating an e-Learning scenario within a large social network by showing its capability to satisfy all the actors.
E-Learning (EL) represents a good solution for courses, as
it provides time and location flexibility, low costs and
information sharing [
In addition, software agents can support EL class formation
processes by suggesting to students (classes) about the best
classes (students) to join with (to accept) [
It is obvious that, due to the huge amount of data of OSNs
and the huge number of thematics groups, examining the
entire space of data to suggest suitable solutions for learners’
needs is impracticable. Therefore, based on previous research
experiences [
The experimental trials have shown that running the CF algorithm allows students and class administrators to improve the average value of the convenience within classes.
The rest of the paper is organized as follows: Section II introduces the context and the Expertize, Trust and Advantage measures. The proposed architecture is described in Section III, while Section IV discusses the GF algorithm. Section V presents the experiments we carried out, Section VI examines related literature and, finally, in Section VII we draw our conclusions.
Let be N the set of OSN members, (|N | = N ), C the set
of classes (|C| = C), with each class c ∈ C consisting of a
number of learners and at least a teacher. We also suppose that
each user ui ∈ N is associated with a software agent [
We also define a behavioral measure, which is related to the interactions carried out by a learner (see Section II-A) and a trust measure, which considers the level of mutual trust among OSN members (see Section II-B).
The principle behind the definition of the behavioral measures is that, in order to form classes, a balance between required and/or offered skills should be desirable, as each learner is interested to improve his/her knowledge by joining with classes where the other members have suitable capabilities and managers of classes are interested to include users holding skills and attitude to interact.
Classes. Lets’ define a class c as a tuple hS, W, Vc, oi where:
(i) S = { s1, s2, . . . , sm} is the skill set required by the class
manager of c; (ii) W = { w1, w2, . . . , wm} is the weight set
used to evaluate the students’ skills; ( iii) Vc is the minimum
overall skill grade, computed over the specific skill set S,
required to join with c; (iv) o is the reference topic or subject or
goal of c. More formally, for an OSN user uk and a skill set S,
m
V (k, S) = P wi · g(k, si), where g(k, si) ∈ [
The User attitude H. The attitude of the user uk to require and/or offer interactions for his/her skills is computed as:
H(k) = α · H(k) + (1 − α) · H(k)
where the new value of H ∈ [
Class Behavior. The class behavior for the class cj , denoted
as B(j) ∈ [
The second measure is based on the concept of trust [
An interaction between two generic OSN learners up and ur consists of a process where up starts with one or more learning tasks with ur. Consequently, their software agents ap and ar observe the interactions of their owners to register the interaction features (type, topic, duration) and collect feedbacks about other OSN users to compute their respective reputations. Such feedbacks refer to the quality of these interactions (remember that users’ skills are evaluated by the Behavioral measures).
Let ηp,r and ρp,r be respectively the measures of reliability
and reputation that the OSN user up (i.e. agent ap) computes
for the OSN user ur (i.e. agent ar). The trust measure τp,r is
obtained by combining the reliability (ηp,r) and the reputation
(ρp,r) weighted by means a coefficient βp,r ∈ [
βp,r = M ax(β1, β2) where β1 = min
IImp,arx , 1 and β2 = 1 − Ω(pt,)r. The parameter Ω(pt,)r is the average confidence at time t for the current set of recommenders that provided at least a recommendation to ap about ar computed as Ω(pt,)r = 1 PiR=p1,r | σp(t,r−1) −τq,r| and where Rp,r is the set of agents | pRrop,vri|ded an opinion about ar. It minimizes the effect of untrustworthy opinions by giving more relevance to those mentors evaluated by ap as the most similar to it. Ip,r is the number of interaction that is incremented at each step and when it is greater than the threshold Imax then the “knowledge” between two users is considered maximum. As a result, the contribute of the reputation in computing trust decreases as much as the number of the interactions occurred between the two involved learners constantly increases.
1) Computation of Reliability: The reliability measure,
ηp,r ∈ [
The parameter ϑp,r considers the relevance assigned to the interaction between up and ur, let it be Ψp,r. In principle, malicious behaviors aimed to gain good reputation with low value interactions (Ψ ≪ 0.5) but high reliability (σ ≫ 0.5) can start on interactions of high relevance (Ψ ∼ 1) (due to a good reputation) but providing poor performance (σ ∼ 0). Therefore, the closer the ratio Ψ/σ to 1, the higher the value of ϑ; the farther the value Ψ/σ from 1, the lower the value of ϑ. A possible choice for ϑ is represented by the adoption of the Gaussian centered in 1, as ϑ = e−(Ψ/σ−1)2/v2 . ϑ acts as a “filter” for those values of σ which, for the correspondent values of Ψ, may reflect a malicious behavior, while large values of v will select only those values of σ for which σ/ϑ is close to 0 by ensuring that almost the whole history of feedbacks σ is considered in computing η.
2) Computation of Reputation: The reputation measure
ρp,r ∈ [
X q=1 τq,r Through the usual meaning of these indexes, 0/1 means that ur is totally unreliable/reliable.
Behavioral and trust measures are combined to measure the convenience, for a user, to join with the class cj . The asymmetric nature of the trust measure implies also the asymmetry of the convenience. In particular, let φ be a parameter computed as φ = (1−| H(k)−B(j)| ) , where | c| is the number of users | cj| (i.e. agents) affiliated with c. Then the convenience (γu,c) for the user u to join with the class c, and that (ηc,u) of the class c to accept the affiliation request of a user u are computed as: γk,j = φ X
τk,i ai∈cj ηj,k = φ X
τi,k ai∈cj
Both measures increase with the difference between the
behaviors of ak and cj . As a consequence of the asymmetric
nature of trust, the procedure described in Section IV is
distributed among the agents assisting learners and those assisting
class managers. As it will be discussed in the experimental
Section, the aim of the distributed procedure is to let the
system to reach a balance in terms of convenience among all
the considered actors of the proposed OSN EL scenario. [
The Learner Agent Behavior. The behavior of a learner agent consists of several tasks periodically executed to maintain data useful to run the CF algorithm (see Section IV). Let uk be the generic learner and ak his/her agent, the following tasks are triggered by the learner and executed by the agent as: (i) Any interaction of learner uk with one or more peers will trigger agent ak to update Behavioral measures; (ii) Any reliability change of a user uj that interacted with a peer, will trigger ak to update the Reliability measure; (iii) The Convenience measure will be updated for any change in the Reliability measure; (iv) Behavioral and Trust measures are periodically sent to the class agent once and if they have been recalculated; (v) The generic software agent ak will assist user uk to take decision about joining with or leaving classes.
The Class Agent Behavior. The behavior of a Class Agent consists of several tasks executed periodically to maintain data useful to run the CF algorithm (see Section IV). Let cj be a class and Aj the associated software agent, the following tasks are triggered by the interactions among software learner agents ak and the class agent Aj as: (i) Any message of ak containing updated Behavioral and/or Convenience measure will trigger agent Ak to update Behavioral and/or Convenience measures for the whole class; (ii) Whenever the Behavioral measure of the class cj has changed, Aj will send the updated measure to all the learner agents of the class cj ; (iii) Ak will assist the class manager of cj to take decision about the requests coming from agents ak to join with or leave classes.
FORMATION (CF)
In our approach, each Learner Agent has: (i) to update all the proposed measures whenever one or more interaction occurred, (ii) to send the new values to its class agents and (iii) to assist its own user to take decision about joining with or leaving classes by executing the CF algorithm (For this aim, it will receive behavioral and trust measures from its own class agents).
Each class agent has (i) to wait for learner agents messages in order to update the proposed measures of the entire class, (ii) to send updated behavioral measure for allowing learner agents to update their own convenience measures and (iii) to assist its own class manager to take decision about the requests coming from learner agents to join with or leave classes.
Let T be the time between two consecutive steps of the CF procedure executed by the generic learner agent, in order to join with a set of classes of the same topic. We also suppose that agents can query a distributed database named CR (Class Repository) on which the list of the classes is stored.
The CF procedure performed by the learner agents (See
Fig. 1(a)). Let Xn be the set of the classes the agent an is
affiliated to, and NMAX the maximum number of classes an
agent can analyze at time t, with NMAX ≥ | Xn| . Besides,
suppose that an stores into a cache the class profile of each
class contacted in the past and the timestamp d of the last
run of the CF procedure for that class. Let the timestamp
ξn and χn ∈ [
The CF procedure performed by the class agent - Fig. 1(b). Let Kc be the set of the agents affiliated to the class c, and KMAX the maximum number of learners allowed to be within the class c 1 with | Kc| ≤ KMAX . Suppose that the class agent Ac stores into a cache the profile P of each user u 1For convenience it is assumed the same for all the classes and topics Input:
Xn, NMAX , ξn, χn; Y = { c ∈ C} a random class set : | Y | ≤ NMAX , Xn T Y = { 0} , Z = (Xn S Y ) 1: for c ∈ Z : dc > ξn do 2: Send a message to Ac to retrieve the profile Pc. 3: Compute γun,c 4: end for 5: Let be Lgood = { ci ∈ Z : i ≤ j → γun,ci ≥ χn} , with | Lgood| = NMAX 6: j → 0 7: for c ∈ Lgood ∧ c 6∈ Xn do 8: send a join request to Ac 9: if Ac accepts the request then 10: j → j + 1 11: end if 12: end for 13: for c ∈ { Xn − Lgood} ∧ j > 0 do 14: Sends a leave message to c 15: j → j − 1 16: end for Input:
Kc, KMAX , ωc, πn, ar, Z = Kc S{ ar} ;
1: if (V (r, Sc) < Vc ∨ | Kc| ≥ KMAX ) then
2: Send a reject message to ar
3: else
4: for a ∈ Kc do
5: if du ≥ ωc then
6: ask to a its updated profile
7: end if
8: end for
9: for a ∈ Z do
10: compute ηc,a
11: end for
12: Let be Kgood = { a ∈ Z : γc,a ≥ πc}
13: for a ∈ Kc − Kgood do
14: send a leave message to a.
15: end for
16: if ar ∈ Kgood then
17: the request of ar is accepted
18: end if
19: end if
Fig. 1. CF algorithm. Top (a): Learner Agent. Bottom (b): Class Agent
managed by his/her learner agent a ∈ Kc and the timestamp
du of its acquisition. The procedure run by Ac is triggered
whenever a join request by a learner agent ar (in the interest of
ur) is received by Ac (with the profile Pr). Let the timestamp
ωc and πn ∈ [
In order to evaluate the described approach, we performed some experiments to investigate on the convergence of the CF algorithm described in Section IV. As a measure of the internal convenience for a class cj , we introduced the concept of Average Convenience (AC), computed as the average of all the measures of convenience ηj,i computed by cj ∈ C for all its students ui ∈ cj . To measure the global convenience of all the classes of N , we computed the mean M AC = Pcj∈C ACj /|C| and the standard deviation DAC = qPcj∈C (ACj − M AC)2/|C| .
A first test involved three scenarios consisting of 50, 100,
and 200 e-Learning classes, as summarized in Table I. To
compute the convenience, we assumed that 20% of OSN
members is unreliable. Behavioral coefficients hreq and hoff
and the values of trust (τ ), have been sampled from a normal
distribution [
For the second set of experiments we assumed a variable
v0a.1luteo o0f.9r. A=vKaNlmmueaaxxr·|·| UC<|| ,1assayshuoswtnhaitnusTearbsl,einIIoI,verraanlgl,incganfrjooimn
more places (Nmax · | U | ), than the total allowed (Kmax · | C| ).
In particular, the best improvement, in terms of M AC, is for
r = 0.4 (+20%), r = 0.5 (+16%) and r = 0.6 (+20%).
It means that, from one hand, finding a class to improve the
personal convenience γ is a bit more difficult for the user
when r < 1, therefore the CF algorithm helps to improve
the MAC with respect to the random composition of classes.
Nevertheless, the algorithm clearly needs a certain degree of
freedom to give some benefits. Therefore, when r is very
small, the improvements, in terms of M AC are comparable to
those given for values of r close to 1. In overall, these results
point out that the CF algorithm gives, on average, a relevant
improvement of the convenience for the classes. [
In order to test the effectiveness of the trust model we have verified, by simulations, that the class formation algorithm will lead to high and stable values of average convenience. Simulations have shown that the CF algorithm will lead significant benefits in terms of average quality of interactions.
Group/class formation is an important task to promote EL
activities and obtain effective results [
A recent survey [
In [
Class formation in e-Learning is a critical task for the quality of such activities. In this work we focused on a distributed algorithm supported by a trust model and some behavioral measures based on information coming from the OSN (i.e. users trust relationships, interaction quality, historical attitude to interact with peers) to improve the metrics for dynamic class composition in OSNs. This flexibility is aimed at improving the quality of learning experiences and it is obtained by combining information about trust and previous interactions in a unique measure named “convenience”. In this work we have shown a first set of experimental results obtained by simulating an artificial scenario with a variable number of users and groups. The results have shown that the class formation algorithm will lead to high and stable values of average convenience.
As future work, we will perform a further experimental campaign in order to verify that the convergence to high values of convenience leads to significant benefits in terms of average quality of interactions. Moreover, a further set of experiments is needed to verify the effectiveness of the trust model to limit malicious behaviors, in order to give trust values which reflect the actual behavior, in terms of overall reliability, of the students.