1 An Agent-based Architecture to Recommend Educational Video Domenico Rosaci and Giuseppe M. L. Sarné Abstract—Agent-based recommender systems are tools able the other agents) into a global trust measure on the basis of to assist users’ choices with suggestions coming closest to their the number of interactions occurred between the two agents. orientations. In this context, it is relevant to identify those users This characteristic appears important being the trustworthiness that are the most similar to the target user in order to require them suitable suggestions. However, particularly when we deal of an agent determinable only after a sufficient number of with video contents for e-Learning, it should be appropriate observations, and this leads to the necessity of dynamically also to consider (i) recommendations coming from those students changing the importance of the reliability vs the reputation. resulted the most effective in suggesting video and (ii) the effects of the device currently exploited. To address such issues in In the ELSA architecture (Figure 1) each device hosts a multimedia scenario, we propose a multi-agent trust based a device agent which monitors the student’s behavior on recommender architecture, called ELSA, appositely conceived to that device to build his/her local profile. An assistant agent this aim. Some preliminary performed simulations permitted to evaluate our proposal with respect to the other considered agent- periodically collects such profiles to build the student’s global based RSs. profile and, based on it, associates the student with one/more partitions. Each partition groups similar students and it is Index Terms—Device adaptivity, e-Learning, Multimedia, Rec- ommender system, Trust system managed by a partition agent that off-line precomputes per- sonalized suggestions for its members by considering the exploited device and effectiveness (based on the adopted trust I. I NTRODUCTION model) in providing opinions in the CF process. Note that An increasing number of video contents (VC) is available assistant and partition agents exploit the cloud technology. on the Web and many recommender systems (RS) support Finally, each VC site is supported by a site agent that builds users by suggesting them those VCs that better meet their personalized site presentations [9]–[12] for each its student orientations on the basis of their past interests (Content- with the VCs suggested by the partition agents which him/her based RSs [1]) and/or those of similar people (Collaborative belongs to. Filtering RSs [2]) or a their combination (Hybrid RSs [3]). We remark as in computing suggestions the partition agents A current trend in collaborative filtering (CF) processes is to exploit information periodically sent them by the assistant consider both the similarity with respect to the characteristics agents to take into account changes occurring in students’ of the device used in accessing VCs [4] (e.g. interface features, interests. This implies that partitions are periodically recom- storage and computational capabilities, bandwidth and con- puted, allowing the system to adapt itself to the evolution nectivity cost) and the effectiveness in providing suggestions. of students’ behaviors, while the VC sites have not the For instance, (i) a suggested video could not be accessible on computational costs for continuously performing onerous com- a mobile device unable to correctly displays it and/or if its putations. Some simulations performed to evaluate ELSA by download is expansive in time or money or (ii) the opinions comparing it with other agent-based RSs shown an improve- provided by a user, apparently similar to the target user, could ment in terms of performances. result totally ineffective or even misleading. To face such issues in a VCs e-Learning scenario, we propose a distributed multi-agent RS architecture, called E- Learning Student Assistant (ELSA), to support students with personalized suggestions about VCs which considers both the device currently exploited by a user and the information clouds derived by a trust system [5]–[7] in order to choose the partition agents most effective users in generating CF suggestions. In par- assistant agent ticular, ELSA adopts the Trust-Reliability-Reputation (TRR) model [8] which dynamically merges reliability (a measure of the trust directly perceived by an agent with respect to device agents site agent another) and reputation (a measure derived by the opinions of D. Rosaci is with the Dept. DIIES, University of Reggio Calabria, Loc. Feo di Vito, 89122 Reggio Calabria, Italy, e-mail: domenico.rosaci@unirc.it G.M.L. Sarné is with the Dept. DICEAM, University of Reggio Calabria, Loc. Feo di Vito, 89122 Reggio Calabria, Italy, e-mail: sarne@unirc.it Fig. 1. The architecture of ELSA 2 device assistant site partition agent agent agent agent 1 local profile global profile 0.8 site catalogue 0.6 δ pre-computed 0.4 recommendations student Web access working profile working profile 0.2 recommendations 0 personalized Web site presentation 0 20 40 60 80 100 120 140 160 180 [for each partition which the user belongs to] DAYS Fig. 2. The behaviour of ELSA Fig. 3. The coefficient δ II. T HE ELSA A RCHITECTURE The Local Profile LP stores the student’s sessions history Each ELSA Web site is an XML site which presents in accessing VCs on D and includes the Accessed Video Set unvisited VCs potentially interesting to each visiting student. (AV Set) and the Interest Weight Set (IW Set). AV Set is a To this aim, each VC on the platform is associated with a set of tuples ⟨id, LVid , LAVid , ϕid ⟩, each one associated with a unique identifier code and with a category c (e.g. Java, XML, video v, where id is the identifier code (unique in ELSA) of v, etc.) belonging to a catalogue C, common to the ELSA agents, LVid is the number of accesses to v performed by s on D with realized as an XML-Schema document where each element LAVid (i.e. Last Access Video) its last access date and ϕid ∈ represents a category c ∈ C. Besides, for each student a global [0, 1] a score provided by s to measure his/her appreciation profile is built and updated to represent videos and categories about v, where 1/0 means the maximum/minimum satisfaction. of interest and a measure of their relevance for him/her. After the first access of s to v, his/her device agent computes In the ELSA architecture (Figure 2) each device agent is as- ϕ as the ratio between the time spent by s on v with respect to sociated with a student’s device, assistant and partition agents its whole time length. It will be the measure of the satisfaction live on the cloud and each site agent exploits computational degree of s in absence of his/her explicit evaluation, always in and storing resources of a Web site server. When the student the range [0, 1]. For each further access of s to v then he/she s uses the device D to visit a Web site W , the associated could (i) confirm the current ϕ value, (ii) provide a new ϕ device agent d: (i) monitors his/her Web activities to build value or (iii) accept a new device agent evaluation of ϕ. and update his/her Local Profile; (ii) sends its Working Profile, IW Setd is a set of tuples ⟨c, LCc , LACc , LIWc ⟩, each one storing constraints in accessing VCs on D, to the site agent associated with a category c ∈ C, where LCid is the number w associated with W . Besides, each device agent periodically of accesses a video belonging to c performed by s on D sends its Local Profile to the student’s assistant agent that with LACc (i.e. Last Access Category) its last access date builds and updates his/her Global Profile to compute a global and LIWc (i.e. Local Interest Weight) a measure of his/her measure of relevance for videos and categories visited by s interest in c on D, that we suppose to be null after 180 days. in order to associate s with one or more partitions, each one In other words, the first time in a day that c is visited, LIW managed by a partition agent which in turn: (i) stores all the is updated as LIWc = δ · LIWc + 1 based on its past value Global Profiles of the affiliated students and of the ELSA weighted by δ and a contribute for the current access set to 1. sites; (ii) for its affiliate students determines similarities and The coefficient δ (Figure 3) decreases the past LIWc based on trust levels, in terms of effectiveness in providing suggestions, its age (measured in days) and ψ is a coefficient experimentally to pre-compute content-based (CB) and CF suggestions about set to 0.445. More formally: potentially attractive unvisited VCs: (iii) sends the precom- ( ) ψ1 180−(current date−LACc ) puted suggestions, compatible with the Working Profile of the δ= 180 device exploited by s in visiting W , to a site agent which arranges them on-the-fly in a presentation for s. The device agent periodically sends its profile to the assis- tant agent of s and its profile W P to the site agents of the A. The Device Agent visited ELSA Web sites (see below). Periodically, the device agent prunes its profile LP from aged information. Each device agent d builds and updates its Device Profile consisting of Working (W P ) and Local (LP ) Profiles. The Working Profile stores the following data and data structures: B. The Assistant agent • m, n : numbers of categories and videos for category that The assistant agent a of s runs on the cloud to build a global s desires to be suggested in a Web presentation on D; representation of activities and interests of s (by using the • Constraint List (CL) : list of restrictions in accessing information received from his/her device agents) to associate videos due to device hardware/software limitations; s with one/more partitions (Figure 1). The assistant agent pro- • Agent List (AL) : list of the student’s assistant agent and files consists of the Assistant Working (AW P ) and the Global of all his/her partition agents identifiers. (GP ) Profiles. The first stores the following information: 3 / • N D, N P : number of device agents of s and number of as sims,t = |GIW Ss ∩ GIW St | |GIW Ss ∪ GIW St |. In partitions which s desires to be affiliated; computing sims,t we consider a student as interested in c • k : a real coefficient, arbitrarily set by s, ranging to [0..1], only if its GIWc value is greater than a system threshold and weighting similarity vs trustworthiness; J. Therefore,∑ the average/ similarity sims,P is computed as • z : number of agents that s desires to exploit in computing sims,P = ( t∈P sims,t ) |NP |, where NP is the number of collaborative filtering recommendations. students of P , and sent to as to consider the affiliation of s, The Global Profile (GP ) includes the Global Accessed (this implies that s will send its Global Profile GPs to p). Video Set (GAV Set) and the Global Interest Weight Set When s visits the site W by using the device D, the (GIW Set). GAW Set represents each video v accessed by associated device agent d sends its W P to the site agent s with a tuple ⟨id, GVid , GLAVid , Φid ⟩, where: (i) id is the w that forwards it to the partition agents of s. To generate identifier code (unique in ELSA) of v; (ii) GVid is the overall CB suggestions, each partition agent of s builds a list (CBs ) number of accesses to v performed by s; (iii) GLAVid (i.e. storing the ns,D most popular VCs of W , unvisited from s Global Last Access Video) is the most recent date among the and compatible with W P , for each one of the ms,D most accesses to v performed by s; (iv) Φid is the average rating interesting categories into GPs . To compute CF suggestions among all the ϕid values stored in his/her Local Profiles. In each partition agent uses opinions coming from those students GIW Set each category c ∈ C is associated with a tuple resulting mostly similar to s (also for the device exploited in ⟨c, GCc , GLACc , GIWc , ⟩, where: (i) GCc is the overall num- visiting W ) and trusted in P for providing fruitful suggestions. ber of accesses to c performed by s; (ii) GLACc (i.e Global To this aim, a partition agent compares the profile of s stored Last Access Category) is the most recent date among the in SISS [s, D] ∈ SISW with each profile DSISW [t, D] of accesses to c performed by s; (iii) GIW each its student t that visited W by using the same type of ∑cNisD the Global Interest Weight computed as GIWc = GC 1 · i=1 LCc · LIWc,i . device in order to compute a similarity measure SIM (s, t, D), on all the shared categories, c ∈ C as SIM (s, t, D) = c Note that GIWc implicitly considers device and connection based ∑ c∈SISW [s,t] SISW [s,D] |SIWc,s − SIWc,t |. All these mea- ∩ cost in the choices of s. In particular, the connection cost is an indirect indicator for the relevance of c [4]. Periodically a sures are normalized among them to assign to each agent updates the global profile GP of s by using updated local t a score ηt by taking into account similarity and trust as profiles and sends its GIW Set ∈ GP to each partition ηt = k · SIM (s, t, d) + (1 − k) · τs,t , where τs,t is the trust agent in ELSA to obtain a similarity measure of s with of t perceived by s (see Section III). Then the partition agent their affiliated. In turn a will affiliate s with the first k inserts in the list CFs the most popular ns,D videos stored in partitions having the higher similarity measures, where their the site W , unvisited from s and compatible with W P , for partition agents will generate personalized suggestions for s. each one of the ms,D most interesting categories presents into This approach appears as a reasonable compromise, although their profiles from the zs agents having the highest η score. it could not consider the last orientations of s. D. The Site Agent C. The Partition Agent The site agent w is associated with the Web site W and A partition P , associated with a partition agent p, affiliates its profile only consists of the catalogue of the categories students similar for interests and all the ELSA site agents. present on the site. When the student s is visiting W by using The profile of p consists of the Site Catalogue Set (SCS), the the device D, the associated device agent d sends its current Global Profile Set (GP S) and the Site Interest Set (SIS). The Working Profile W P to the agent w that in turn forwards it to first includes the catalogues of all the ELSA Web sites. The all the partition agents of s for receiving their CB and CF Global Profile Set stores the global profiles of all the students suggestions lists, conform to W P , they pre-computed for s. of P . The Interest Site Set stores for each site W and in a data From such lists the site agent selects the first most relevant n section SISW , a list SISW [s, D] for each student s visited W VCs for each one of the m categories. The site agent uses the by using D. The elements of SISW [s, D] are pairs (c, SIWc ), selected recommended VCs to build on-the-fly a personalized where c is a category present in W and considered interesting site presentation for s When a student’s Web session ends, by s, and SIWc is its Site Interest Weight, an interest measure the site agent informs the partition agents of s about his/her of s in c computed on the site side. These information are sent choices in order to update the students’ trust values. by the site agent w to p at the end of each Web session of s. The partition agent p computes the similarity measures each III. T HE T RUST C OMPUTATION time it is required by an assistant agent to evaluate the similar- This trust model, derived by [8], provides each student ity of its student s with those affiliated with P based on their of P with the trustworthiness of the other students of P in Global Interest Weight Sets (in order to consider the affiliation suggesting VCs associated with a category c ∈ C. In particular, of s with P ). The average similarity measure sims,P ∈ [0..1] for each category c and for each student s, to know the trust between s and each other student t ∈ P is the mean of all values of each other student t ∈ P (i.e. τs,t c ) on c, a linear the Jaccard similarity measures sims,t [13], a real number system of N S equations in N S variables (where N S is the ranging in [0..1] computed as the ratio between the number Number of Students of P ), that admits only one solution, has of their common categories which s and t are interested to to be solved. With respect to the category c and the students and their total number (data stored into the s and t profiles) s and t then each equation [8], [14] has the form: 4 TABLE I 1 S ETTING OF THE ELSA PARAMETERS OF THE DEVICE AGENTS . 0,9 0,8 0,7 Average Precision device NP m n z k 0,6 0,5 RS2 PC 3 3 4 3 0.5 0,4 MWSuggest tablet 3 3 4 3 0.5 0,3 ELSA cellphone 3 3 4 3 0.5 0,2 0,1 0 S200 S400 S800 Set ∑ j∈P −{s,t} τj,t · τs,j c c c τs,t c = αs,t · ρcs,t + (1 − αs,t c )· ∑ c (1) Fig. 4. Average precision of RSs for different set of simulated users j∈P −{s,t} τs,j 0,9 which combines reliability and reputation measures weighted 0,8 c by αs,t ∈ [0..1], that depends on the number of interactions 0,7 occurring between s and t with respect to c (i.e. ics,t ) and 0,6 Average Recall c = ics,t /N , if ics,t < H or 1 otherwise, 0,5 is computed as αs,t 0,4 RS2 c MWSuggest where the integer H is a system threshold. In other words, αs,t 0,3 ELSA increases with the direct knowledge that s has of t about c. The 0,2 first contribution in Eq. 1 is the reliability (i.e. ρcs,t ∈ [0..1]) 0,1 0 based on the level of knowledge that s has of t about the S200 S400 S800 Set category c, where 0/1 means that t is unreliable/reliable. It is computed as the number of suggestions that s accepted from t, normalized by the total number of recommendations provided Fig. 5. Average recall of RSs for different set of simulated users by t to s. The second contribution, ranging in [0..1], is the reputation of t that s computes by requiring an “opinion” on standard measures Precision πs (πs = |T Ps | |T Ps ∪ F Ps |) the capability of t to provide good suggestions in the category and Recall ρs (ρs = |T Ps | |T Ps ∪ F Ns |) for the set of c to each student j ̸= t ∈ P . This opinion is the “trust” that j produced suggestions for s that can be interpreted as the c has in t, weighted by the trust measure τs,j that s has of j In probabilities that a suggested link is considered as relevant this way, the reputation is the weighted mean of all the trust by s and that a relevant link is recommended, respectively. measures τj,tc that each student j (with j ̸= s, t) assigns to t. The Average Precision π and the Average Recall ρ of each RS, defined as the average of the π and ρ values, have been IV. E XPERIMENTS computed for the all tested RSs and student sets and show We performed some simulations to compare ELSA with as ELSA is better than the other tested RSs on both such RS2 [15] and MWSuggest [4] by using 20 XML Web sites, measures that increase when the size of the agent popula- each one including in average 50 VCs associated with cat- tion is larger. The maximum advantage in term of precision egories belonging to C. The RSs have been implemented (resp. recall) with respect to the second best performer (MW- in JADE [16] and JADE/LEAP [17] for the devices having Suggest) is equal to 10,25 % (resp. 12,5 %) for a size of limited resources. The ELSA and MWSuggest device agents 200 users, and becomes equal to about 16 % (resp. 19 %) for (associated with a desktop PC, a tablet and a cellphone) have a size of 800 users. If the advantage of ELSA with respect set their parameters as in Table I. Note that k is set to 0.5 and to RS2 might be attributed to consider the exploited device gives the same relevance to similarity and trust. in generating the suggestions, the better performances with We simulated students visiting the Web sites and stored in respect to MW-Suggest are surely due to the use of the trust a file (exploited as test-set) the first simulated 200 choices model in computing the CF suggestions (indeed in terms of of each one on the first 10 Web site for different VCs as a performances these two systems perform in an identical way list of tuple (consisting of source (s) and destination (d) links with respect to the CB component of the suggestions). and the timestamp (t) of this choice) and other 200 tuple for the other 10 sites used to evaluate the RSs. The experiment V. R ELATED W ORK involved three sets of 200, 400 and 800 simulated students. e-Learning RSs [18], [19] guide students by suggesting For each set, when a student s is visiting a Web page, each educational resources [20] based on their profiles [21], [22]. partition agent (for each one of the 3 RSs) provides a set R(p) In this context, CB suggestions are in line with users’ past of suggestions. Each element r ∈ R(p) is a link to a VC and interests [22], but suffer for attribute selection [23], over- (i) if r is accepted then it is considered as a true positive and specialization and inability to consider unknown items, while inserted in a set T Ps of all the true positives generated for s, CF [24] approaches have high computational costs due to (ii) if it is unaccepted then it is a false positive and inserted high data dimensionality and sparsity [25]. By combining in a set F Ps , otherwise (iii) if a choice of s not belong to CB and CF techniques most effective (hybrid) suggestions R(p) then it is considered as a false negative and inserted in can be computed [3], [26]. For instance, hybrid e-Learning a set F Ns . To measure RSs performances we computed the systems are presented in [27], which explores the impact of 5 using a massive repository of educational indexed resources, relationships among peers allow to select which ones of them personal data derived from students’ actions and more com- can be considered more authoritative in answering a query binations of CB and CF recommenders and in [28] where within a given topic, whereas trust about learning resources group collaboration is supported, independently of time and allows the most reliable resources to be selected. space distance, and hybrid suggestions consider students and Learning Networks are open infrastructure to provide teach- learning resources profiles, metadata, structural and semantic ers and learning objects. However, due to the available number filtering criteria. of learning resources teachers are supported in finding the most Nowadays, students access Web resources every time and suitable for them by RSs. In this context, to make accurate everywhere by using different type of device and, therefore, recommendations by solving the problems due to sparsity RSs should suggest resources (i) natively compatible or (ii) of educational datasets, in [50] it is proposed to adopt trust adaptable to the device. Moreover, suggestions can be based information obtained by monitoring the teachers’ activities. on (i) a unique global student’s profile which takes into Finally, EVA [51] is a framework of learning recommender account learner’s activities performed on each his/her device agents migrating among users based on a cloning mechanism. or (ii) different learner’s profiles, one for each his/her device. Each agent stores in its profile the knowledge learned from ELENA [29] is a full distributed RS where a personal agent all its past and current owners. A reputation system, inspired recommends resources to a student based on the information to genealogical criterion, helps the system to select the most stored in his/her profile, only referred to the exploited device, trusted agents in the community to be cloned and migrated and in the those of other students sharing the same interests among users’ to provide suggestions. and device. An interface based on a traffic light metaphor brings out the recommended resources. In ISABEL [30], a VI. C ONCLUSIONS distributed multiagent learning system, each student’s device We presented ELSA, a fully distributed hybrid RS agent is monitored by an agent when the student accesses e-Learning architecture appositely designed to provide students with per- Web sites, each one associated with a teacher agent. Each sonalized suggestions on potentially interesting VCs by also student is associated with one or more tutor agents providing taking into account the device currently exploited. Besides, in personalized suggestions for him/her, also considering the the generation of CF suggestions, ELSA considers not only device, that the teacher agent shows in a personalized pre- those users that are the most similar to the current user but sentation compatible with the student’s devices. Also in [31] also those that in the past are resulted as the most effective in are considered the opportunities provided by mobile devices provide suggestions on the basis of a trust system integrated to delivery personalized contents (adapted by algorithms de- in ELSA. Some simulated evaluations have shown promising signed to this purpose) compatible with learner’s preference on performances of the ELSA platform with respect to some other that device, device capabilities and contextual environment. tested RSs proposed in the literature. As our ongoing research Human interactions widely exploit the concept of trust to we are planning to test ELSA with real users in the next future. know the trustworthiness of own counterparts with respect to same held skills or in order to avoid deceptions [32]– R EFERENCES [35]. Therefore trust plays a role similar to a social control [1] P. Lops, M. Gemmis, and G. Semeraro, “Content-based recommender to determine the best subjects to interact with, particularly systems: State of the art and trends,” in Recommender Systems Hand- important in virtual environments which encourage possible book. Springer, 2011, pp. 73–105. malicious behaviors [36], [37]. In particular, information de- [2] J. Breese, D. Heckerman, and C. Kadie, “Empirical analysis of predictive algorithms for collaborative filtering,” in Proceedings 14th International rived by direct experiences (i.e., reliability) can be used to trust Conference on Uncertainty in Artificial Intelligence. Morgan Kauf- others, but they usually only exist for a narrow set of users mann, 1998, pp. 43–52. and/or for a small number of times. As a consequence, a direct [3] R. Burke., “Hybrid recommender systems: Survey and experiments,” User Modeling and User-Adaptivity Interaction, vol. 12, no. 4, pp. 331– and reliable opinion about someone could be impossible to 370, 2002. have, therefore to trust potential partners the opinions provided [4] D. Rosaci and G. M. L. Sarné, “Recommending multimedia web services by other users (i.e. reputation) have to be considered and in a multi-device environment,” Information Systems, vol. 38, no. 2, pp. 198–212, 2013. the reputation accuracy increases as much as their number [5] J. Sabater and C. Sierra, “Review on computational trust and reputation increases [36]. Reliability and reputation information are often models,” Artificial Intelligence Review, vol. 24, no. 1, pp. 33–60, 2005. combined together in order to obtain a single synthetic trust [6] D. Sarvapali, S. Ramchurn, and N. Jennings, “Trust in multi-agent systems,” The Knowledge Engineering Review, vol. 19, pp. 1–25, 2004. measure [38]–[43] also by considering a multidimensional [7] S. Na, K. Choi, and D. Shin, “Reputation-based service discovery in approach [40], [44]–[47]. multi-agents systems,” in Proceedings IEEE International Workshop on With respect to a RS, trust can be assumed as the perception Semantic Computing and Applications. Springer-Verlag, 2010, pp. 326– 339. that the source is competent or, conversely, that a learning [8] D. Rosaci, G. Sarnè, and S. Garruzzo, “Integrating trust measures resource is valid and interesting. In other words, it is the per- in multiagent systems,” International Journal of Intelligent Systems, ceived skill of the recommender to offer the right suggestions. vol. 27, no. 1, pp. 1–15, 2012. [9] C. Anderson, P. Domingos, and D. Weld, “Adaptive web navigation In [48] video on demand are recommended by adopting CB for wireless devices,” in Proceedings 17th International Joint Con. on and CF techniques, but this later exploits only expert users Artificial Intelligence. Morgan Kaufmann, 2001, pp. 879 – 884. selected by a trust system. In [49] the idea of trustworthiness [10] P. De Meo, D. Rosaci, G. M. L. Sarné, G. Terracina, and D. Ursino, “An xml-based adaptive multi-agent system for handling e-commerce is associated to both learning resources (described by common activities,” in Proceedings 1st International Conference ICWS-Europe ontologies) and peers in a P2P e-learning scenario. Trust 2003, ser. LNCS, vol. 2853. Springer-Verlag, 2003, pp. 152–166. 6 [11] S. Macskassy, A. Dayanik, and H. Hirsh, “Information valets for [36] A. Birk, “Boosting cooperation by evolving trust,” Applied Artificial intelligent information access,” in Proceedings AAAI Spring Symposium Intelligence, vol. 14, no. 8, pp. 769–784, 2000. Series on Adaptive User Interfaces. AAAI, 2000. [37] G. Lax and G. M. L. Sarné, “Celltrust: a reputation model for c2c [12] D. Rosaci and G. M. L. Sarné, “A Multi-Agent Recommender System commerce,” Electronic Commerce Research, vol. 8, no. 4, pp. 193–216, for Supporting Device Adaptivity in e-Commerce,” Journal of Intelligent 2006. Information System, vol. 38, no. 2, pp. 393–418, 2012. [38] Aberer K. and Despetovic Z., “Managing Trust in peer-2-peer Infor- [13] G. Greenstette, Explorations in Authomatic Thesaurus Construction. mation Systems,” in Proceedings 10th International Conference on Kluwer Academic Pub., 1994. Information and Knowledge Management. ACM, 2001, pp. 310–317. [14] F. Buccafurri, D. Rosaci, G. M. L. Sarné, and L. Palopoli, “Modeling [39] M. Gómez, J. Carbó, and C. Benac-Earle, “An anticipatory trust model cooperation in multi-agent communities,” Cognitive Systems Research, for open distributed systems,” in Anticipatory Behavior in Adaptive vol. 5, no. 3, pp. 171–190, 2004. Learning Systems, ser. LNAI, vol. 4250. Springer-Verlag, 2007, pp. [15] K. Che, Y. Tsung-Hsien, and L. Wei-Po, “Personalized multimedia 307–324. recommendation with social tags and context awareness,” in Proceedings [40] T. Huynh, N. Jennings, and N. Shadbolt, “An integrated trust and World Congress on Engineering 2011, vol. 2. Springer, 2011, pp. 6–11. reputation model for open multi-agent system,” Autonmous Agent and [16] JADE URL, “http://www.jade.tilab.org,” 2012. Multi Agent Systems, vol. 13, 2006. [17] Caire G., LEAP 3.0: User Guide, TLAB. [41] S. Kamvar, M. Schlosser, and H. Garcia-Molina, “The eigentrust algo- [18] H. Drachsler, H. Hummel, and R. Koper, “Personal recommender rithm for reputation management in P2P networks,” in Proceedings 12th systems for learners in lifelong learning networks: the requirements, Interantional Conference on World Wide Web. ACM Press, 2003, pp. techniques and model,” Internatinal Journal of Learning Technology, 640–651. vol. 3, no. 4, pp. 404–423, 2008. [42] P. Melville, R. Mooney, and R. Nagarajan, “Content-boosted collab- [19] O. Santos and J. Boticario, Educational Recommender Systems and orative filtering for improved recommendations,” in Proceedings 18th Technologies: Practices and Challenges. Information Sc. Ref., 2012. National Conference on AI. AAAI, 2002, pp. 187–192. [20] P. Brusilovsky and C. Peylo, “Adaptive and intelligent web-based [43] L. Xiong and L. Liu, “Supporting reputation-based trust for peer-to-peer educational systems,” International Journal of Artificial Intelligence in electronic communities,” IEEE Trans. on Knowledge and Data Eng., Education, vol. 13, no. 2, pp. 159–172, 2003. vol. 16, no. 7, pp. 843–857, 2004. [21] H. Liu and V. Keselj, “Combined mining of web server logs and web [44] J. Sabater and C. Sierra, “Reputation in gregarious societies,” in Pro- contents for classifying user navigation patterns and predicting users’ ceedings 5th International Conference on Autonomous Agents. ACM, future requests,” Data Knowledge Engineering, vol. 61, no. 2, pp. 304– 2001, pp. 194–195. 330, 2007. [45] D. Jia, F. Zhang, and S. Liu, “A robust collaborative filtering recom- mendation algorithm based on multidimensional trust model,” Journal [22] F. Wang and H. Shao, “Effective personalized recommendation based of Software, vol. 8, no. 1, pp. 11–18, 2013. on time-framed navigation clustering and association mining,” Expert [46] D. Rosaci and G. M. L. Sarné, “Rebecca: A trust-based filtering to im- Systems with Applications, vol. 27, pp. 365–377, 2004. prove recommendations for b2c e-commerce,” in Intelligent Distributed [23] K. Cheung, J. Kwok, M. Law, and K. Tsui, “Mining customer product Computing VII. Springer, 2014, pp. 31–36. ratings for personalized marketing,” Decision Support Systems, vol. 35, [47] G. Wang and J. Wu, “Multi-dimensional evidence-based trust manage- pp. 231–243, May 2003. ment with multi-trusted paths,” Future Generation Computer Systems, [24] U. Shardanand and P. Maes, “Social information filtering: Algorithms vol. 27, no. 5, pp. 529–538, 2011. for automating ”word of mouth”,” in Proceedings Conference on Human [48] J. Cho, K. Kwon, and Y. Park, “Collaborative filtering using dual Factors in Computing Systems. ACM/Addison-Wesley, 1995, pp. 210– information sources,” IEEE Intell. Sys., vol. 22, no. 3, pp. 30–38, 2007. 217. [49] V. Carchiolo, D. Correnti, A. Longheu, M. Malgeri, and G. Mangioni, [25] S. Weng and M. Liu, “Feature-based recommendations for one-to-one “Exploiting trust into e-learning: adding reliability to learning paths,” marketing,” Expert Systems with Applications, vol. 26, pp. 493–508, International Journal of Technology Enhanced Learning, vol. 1, no. 4, 2004. pp. 253–265, 2009. [26] J. Herlocker, J. Konstan, and J. Riedl, “Explaining collaborative filtering [50] S. Fazeli, H. Drachsler, F. Brouns, and P. Sloep, “A trust-based social recommendations,” in ACM Conference on Computer Supported Coop- recommender for teachers,” from: http://hdl.handle.net/1820/4428, 2012. erative Work. ACM, 12 2000, pp. 241–250. [51] D. Rosaci and G. M. L. Sarné, “Cloning mechanisms to improve agent [27] M. Khribi, M. Jemni, and O. Nasraoui, “Automatic recommendations performances,” Journal of Network and Computer Applications, vol. 36, for e-learning personalization based on web usage mining techniques no. 1, pp. 402–408, 2012. and information retrieval,” in Advanced Learning Technologies, 2008. ICALT’08. 8th IEEE International Conference on. IEEE, 2008, pp. 241–245. [28] A. Carbonaro, R. Ferrini, and M. Zamboni, “Considering semantic abilities to improve a web-based distance learning system,” in ACM Interantional Workshop on Combining Intelligent and Adaptive Hyper- media Methods/Techniques in Web-based Education Systems, 2005. [29] P. Dolog, N. Henze, W. Nejdl, and M. Sintek, “Personalization in distributed e-learning environments,” in Proceedings of the 13th Inter- national World Wide Web Conference on Alternate track P & P. ACM, 2004, pp. 170–179. [30] D. Rosaci and G. M. L. Sarné, “Efficient Personalization of e-Learning Activities Using a Multi-Device Decentralized Recommender System,” Computational Intelligence, vol. 26, no. 2, pp. 121–141, 2010. [31] X. Zhao, F. Anma, T. Ninomiya, and T. Okamoto, “Personalized adaptive content system for context-aware mobile learning,” International Journal of Computer Science and Network Security, vol. 8, no. 8, pp. 153–161, 2008. [32] C. Corritore, B. Kracher, and S. Wiedenbeck, “On-line trust: concepts, evolving themes, a model,” International Journal of Human-Computer Studies, vol. 58, no. 6, pp. 737–758, 2003. [33] J. Sabater-Mir and M. Paolucci, “On representation and aggregation of social evaluations in computational trust and reputation models,” International Journal of Approximate Reasoning, vol. 46, no. 3, pp. 458–483, 2007. [34] S. Ramchurn, D. Huynh, and N. Jennings, “Trust in multi-agent sys- tems,” Knowledge Engeenering Review, vol. 19, no. 1, pp. 1–25, 2004. [35] P. Resnick, R. Zeckhauser, E. Friedman, and K. Kuwabara, “Reputation systems,” Commununication of ACM, vol. 43, no. 12, pp. 45–48, 2000.