2009 International Conference on Computational Science and Engineering
Let’s Meet: Integrating Social and Learning Worlds Melody Siadaty School of Interactive Arts and Technology Simon Fraser University Canada
[email protected]
Dragan Gasevic School of Computing and Information Systems Athabasca University Canada
[email protected]
consciously/unconsciously create knowledge through their everyday interactions with their peers and friends This embedded information in Social Web platforms can be coupled with the information available in learning systems and leveraged further to provide learners with enhanced personalization or recommendation services. For instance, when visiting their profiles on a social network such as Facebook, learners can be provided with a list of suggested friends who have more academic similarities with them and perhaps will be more suitable collaborative learning peers. Such a service will encourage learners to actively take part in social collaborations, make use of their friends’ knowledge and perhaps follow the same learning path their successful peers have already gone through.
Abstract The widespread welcome to the social web in recent years has led to the emergence of a new source of user information which is a result of users’ everyday activities and contributions on social networks. Leveraging this information within learning environments bears beneficial opportunities that can i) bring the educational (what learners learn in educational settings) and social (what they perform within their social networks) worlds closer, and ii) support further recommendation and personalization services based on more ‘real’ data about learners. This information, however, is neither explicitly available nor formally represented on the social web paradigm, an attribute of the social web regarded as being a “walled garden”. In this paper, we propose a framework that leverages semantic web technologies to locate and extract such data, originating from multiple resources on the social web, to support advanced learning services such as personalized peer recommendation.
One of the major issues with today’s social networks is that they are “walled gardens”; i.e. although a lot of information can be embedded inside them, they are often closed to the outside world [1]. Moreover, each of the social worlds uses its own proprietary format for saving and representing (sometimes semantically identical) data, or covers only some partial data about users e.g. either social aspects such as the information contained in users’ ‘profiles’, which Facebook collects, or more professional attributes like what LinkedIn calls ‘connections’. Such issues make extracting embedded information from social webs and utilizing it in learning environments not to be an easy task, where each piece of users’ information is coming from a different source that in most of the times is neither compatible with other platforms nor portable or centrally accessible.
1. Introduction Nowadays, learning is no more an isolated process restricted to and happening solely in learning environments. The information about learners in traditional learning environments normally can be obtained only via learners’ direct interaction with the system. Today learners, however, are also residents of Facebook1, Flickr2, MySpace3, LinkedIn4 and many other so called Social Web worlds, where they can meet, collaborate, share, gain and
To support formal and sharable representation of knowledge, the Semantic Web technologies in general and ontologies in particular, offer a promising solution [2]. Ontologies shape the building blocks of the
1
www.facebook.com www.flickr.com 3 www.myspace.com 4 www.linkedin.com 2
978-0-7695-3823-5/09 $26.00 © 2009 Crown Copyright DOI 10.1109/CSE.2009.124
Marek Hatala School of Interactive Arts and Technology Simon Fraser University Canada
[email protected]
879
semantic web. They can be shared, merged or extended to fit the requirements of the domain they are representing. Moreover, they allow for reasoning mechanisms over the available relationships, making it possible to infer new knowledge based on the existing instances. Whilst the walled-garden social networking model restrains the portability of social information, ontologies guaranty high levels of expressiveness and flexibility. E-Learning is one of the areas that can greatly benefit from the synergy of social- and semantic-rich, or so labeled ‘Social Semantic Web’[3], applications. In this paper, we propose an ontological framework which leverages a set of different user profile information available on social networks with the goal of bringing learners’ social and learning worlds closer to each other, building seamless learning opportunities. As a step toward this goal, the proposed framework recommends users their most suitable friends (located within their social graph) who can be suitable collaborative learning partners.
networking activities of learners, a prototype was implemented. This prototype was realized in a clientserver architecture and developed using the PHP programming language. The initial plan was to integrate user profile information from two of the leaders of social networks, each focusing on different aspects of user information: one more focused on personal information, friendship and social connections (i.e. Facebook) and the other one more focused on academic and professional links (i.e. LinkedIn). However, at the time of developing this prototype, there was no available client library supported by LinkedIn that is capable of connecting to LinkedIn structure and getting through users' profile information. Facebook on the other hand, provides official support for a few official libraries, i.e. those developed in PHP and JavaScript and other client libraries such as those using ASP.NET, C#, Lisp and Perl, which mainly are supported by the platform developer community of Facebook. This paper reports on a research which is in progress, thus we describe each module of the proposed framework within the context of the current prototype to conform to what that actually is realized so far. However, it should be noted that the ontological basis of the proposed framework easily allows for integrating other features to the framework and extending it with more enhanced classes and relationships based on the needs and requirements of the learning environment, in which the framework is to be deployed.
2. The proposed approach Our proposed framework has ontological basis, making use of the extendable and flexible nature of ontologies. In Figure 1, a schema of the framework is depicted. The main component of the framework is its ontological user model, which aims at modeling users by leveraging users’ information available within social networks. By leveraging this ontological user model, users’ profile information on each of these websites can be extracted and combined together, giving a thorough representation of a user without requiring the learning environment to detect and store all this information during its interaction with the user. The learning environment can later consult the user model to recommend suitable peers of the learners to them, who can also be members of various social worlds. Peer recommendation is based on criteria such as similar research affiliations, learning styles and courses currently being taken. To bring educational and social worlds closer to each other, the recommendation service can be integrated with and presented to learners on their social websites such as Facebook or LinkedIn, where users will then have access to an updated list of their suitable learning partners while performing their every day social activities or, it can be modeled as a service of a learning management system and offered to learners within their educational settings.
LOCO
GUMO
FOAF Specification
User Model Ontology
Repository of User Models
Figure 1. framework
A
schema
of
the
proposed
2.1. Ontological user model Most of current ontological-based learning systems use their own locally designed ontologies. The major problem with local ontologies is that they are usable only within the boundaries of the applications they are developed for, and hardly (re-) usable for knowledge sharing with different systems that cover the same knowledge areas or support similar learning services. To address these issues, we utilize the shareable nature of ontologies in the proposed framework and employ
To investigate the effectiveness of the proposed framework in integrating learning and social
880
some of the features of existing user model ontologies, which are generic enough to be used independent of domain applications, and thus reusable in different learning systems.
addition to the above information which can be instantiated from user's social profiles, there are also some relevant factors such as learning styles, which are not usually reflected on the social web sites, unless there are some external applications built on top of the social site that capture them. There is a wide body of research on adapting different types of learning services to learners’ learning styles [5][6]. We utilized a separate application that obtains users’ learning styles via a questionnaire. The applied learning style questionnaire is Index of Learning Styles (ILS), an online 44-item questionnaire for assessing preferences on four dimensions of a learning style model formulated by Felder and Silverman [7]. This questionnaire is a common and well-established means to identify students' learning styles in different domains, based on which also a notable range of descriptions and validation studies are available.
The major part of the user model of the proposed framework is built upon the Friend of a Friend Ontology (FOAF) (Figure 2). FOAF5 and SIOC are the two important semantic technologies which can be utilized to model social networks, i.e. personal information and relationships between different users in a social network. The Friend of a Friend (FOAF) specification is a powerful and practical ontology that provides basic expressions to represent personal user information and relationships between users, groups and communities. Socially Interlinked Online Communities6 (SIOC) is another ontology, built on top of FOAF, which tries to integrate and link different available online communities and exploit the knowledge available within them. In the user model, each user is defined by using the foaf:Person class, containing personal attributes such as name, gender, and image. The foaf:holdsAccount property, is used to represent details of the user’s account on Facebook. Within this concept, the foaf:accountName property is used to indicate the user unique identification number, which is the user identification number assigned by Facebook to each user. Note that when logging into a Facebook account, users should enter their email address that was used to register for that account. Facebook, however, does not allow the exporting of email addresses and thus, they are not accessible via the related API. To address this issue, we utilized the above mentioned identification number instead.
GUMO:Interest GUMO:hasInterest
foaf:img LOCO:hasLearningStyle
foaf:Person
foaf: Organization
LOCO: LearningStyle
LOCO:hasCategory
foaf:knows foaf:member foaf:holdsaccount
LOCO: LearningStyle Category foaf: OnlineAccount foaf:accountName
String (user_ID)
foaf:Group LOCO: LS_Visual_ Verbal
LOCO: LS_Sequential _Global LOCO: LS_Sensing _Intuitive
LOCO: LS_Inductive _Deductive
Figure 2. Ontological User Model of the Prototype The learning styles of each user are modeled via the um:LearningStyle class of the user model ontology within the LOCO framework7. LOCO (Learning Object Context Ontology) is an integrated set of ontologies (e.g. learning design ontology, domain ontology, user model ontology and learning object ontology) aimed at capturing the information about a specific context of use of a learning object in a specific learning design [8]. All the above described concepts are those that are related to a single user. To model the available relationships and connections between different users, the foaf:knows property is used. To build the social graph of each user, this property is used for each of the user's friends, along with the foaf:Person class to indicate the existence of a new instance of a person and the rest of the classes in the user model to indicate user properties of each friend (Figure 2). In the
Users’ general interest is the next concept in the user model ontology. To represent the users’ interests the GUMO:hasinterest attribute is utilized. GUMO (General User Model Ontology [4]) is a top-level ontology for user modeling that allows for the uniform interpretation of decentralized user models. Finally, in 6
foaf:name
string (pic_URL)
foaf:member
Group and Affiliation are other information which can be extracted from social profile data and are modeled in the user model ontology via foaf:Group and foaf:organization classes respectively. Within the context of the current prototype, the concept of ‘group’ more indicates membership in social groups such as reading groups or fans of a music band, while ‘affiliation’ more represents educational and professional-based memberships such as being a member of a specific graduate school or a company.
5
String (user_name)
http://xmlns.com/foaf/spec/ http://sioc-project.org 7
881
http://jelenajovanovic.net/ontologies/loco/user-model.rdf
following section, we describe in details how each of the above pieces of the user data is inferred and populated in the user model ontology.
(answer b). By answering the 11 related questions per scale, each learner would receive a personal preference for that dimension between 11a (+11) to 11b (-11). If the final score of a scale is 1 or 3 (either negative or positive) the learner is said to be fairly well balanced on the two dimensions of that scale. If they gained 5 or 7, they are diagnosed with a moderate preference for one dimension of the scale and finally, if the score is 9 or 11, they have a very strong preference for that side of the dimension.
2.2. Exporting social web information In the current prototype of the proposed framework, exporting personal information from Facebook accounts is performed through mapping the existing Facebook XML data to concepts in the user model ontology. At the very first step, the user is authenticated through the Facebook’s API. Upon authentication of the user, the process can begin by querying the API to retrieve the user’s personal information and her social graph (i.e. his/her friends and their profile information). The information exported from Facebook profiles include users’ name, interests, affiliations, groups and the list of their friends. Interests include whatever users might list as their interest. The input format for this field in Facebook is open-ended text so before mapping it into the corresponding class of the user model ontology, it should be processed, trimmed and mapped to a unified format for all the users. Affiliations can be selected from a pre-defined list of affiliations provided by Facebook, where each affiliation has its unique identifier number. Groups are defined in a similar way to affiliations. For each of the user’s friends, the same set of information from their Facebook profiles is instantiated in the user model ontology. To retrieve the learning styles for each user, the ILS questionnaire is used. This questionnaire consists of four scales, each with 11 items: sensing-intuitive, visual-verbal, activereflective, and sequential-global (Figure 3).
2.3. Identifying Suitable Peers
and
Recommendation
Once different pieces of the user model are located, exported and populated in the user model, the framework can initiate the recommendation service. The goal is to recommend those friends of users who can be better collaborative-learning partners for them. To be a better (collaborative) learning partner, two users should have similarities. The similarity level between two users in this prototype is defined as a combination of their social and academic similarities. Social similarity happens when two users have shared (general) interests or are members of the same group. Academic similarities include having similar learning styles or being members of the same academic affiliations. The adaptation process starts with calculating the similarity level for each of the above criteria (Ai) for each of user's friends. When all of these similarity levels are calculated, in the next step, these levels are merged (ҏҏAi) into the overall similarity level, which is the weighted sum of these criterion-specific similarities. The final suitability rank of each friend (RFi) to be a potential learning peer would then be equal to the overall similarity level (1). RFi = ҏҏ Ai * Wiҏҏ
(1)
To measure social similarities, levels of similarity for ‘interests’ and ‘group-membership’ are calculated. To calculate the interest similarity level between a user and his/her friends, a built-in function of MySQL (‘match-against’) for full-text search is utilized. MySQL uses ranking with vector spaces for ordinary full-text queries9. This function matches two texts with each other and returns a relevancy rank. Academic similarity can be calculated by considering the similarity levels for learning styles and affiliation attributes of a user. Since groups and affiliations have their own unique identity numbers n the Facebook structure, distinguishing whether two users are members of the same group or have the same
Figure 3. Scales of Index of Learning Style Questionnaire (Adapted from ILS Report Form]8) The description of these dimensions is out of the scope of this paper and can be found in the literature, such as [7]. Each question of the questionnaire is answered either with a value of +1 (answer a) or -1 8 http://www.engr.ncsu.edu/learningstyles/ilsweb.html April 09)
of
(Retrieved 9
882
MySQL online documentations: http://dev.mysql.com/doc/
affiliation is performed by a simple search query and the number of similar group/affiliation memberships is returned as the similarity value for these two criteria. For the learning style criterion, three relevancy levels are considered: the two users either i) have the same preference on a dimension and thus are highly similar, or ii) if not the same, but they have close preferences, and thus somewhat similar, or iii) their learning styles are at the two opposite ends of a scale and thus the similarity level is set to zero. Changes in the weights (Wi) reflect the importance of each of these criteria. These weights (scale 0 to 1) can be set using default values (.25), or to allow for learner preferences, individually by learners. The higher the rank, the more suitable a friend would be as a potential learning collaborator. Finally, the potential learning partners are presented to the user in an ordered list based on their descending similarity values (Figure 4).
framework in i) motivating the users to inquire/investigate further information about recommendations made by the system and ii) providing the rational of the recommendations so the users would be motivated to perform collaborative learning activities in real situations.
Figure 4. A sample of the recommendation list provided to the users
10. Evaluation
At the end of the evaluation session, the users were asked to share their general impression of and comments on the study. Three of the users indicated that they would add only those friends who they personally know well and think that would be suitable learning partners for them. Also, three of the subjects stated that the provided information was not detailed enough to help them make a decision whether to accept a friend as a potential learning collaborator or not. Two of these users wanted to have more information on the similarity levels and one indicated that she would actually rather to receive recommendations based on general interests of her friends instead of a combination of social and academic factors.
An evaluation study was performed to evaluate the effectiveness and efficiency of the proposed framework within the context of the implemented prototype. Nine participants, three males and six females, with the average age of 27 participated in the study. All of the subjects were holding an active Facebook account for at least one year and were quite familiar with the Facebook interface and its features. The average number of friends the users had was 246. The experimental procedure was as follow: First, the participants were given a brief demonstration of the functionality of the system and the goal of the framework was explained. In the second step, they were asked to fill out the ILS questionnaire and then were redirected to the recommendation application asking them to log into their Facebook accounts. The users were assured that this information is neither observed nor saved by the system. Finally, each participant was asked to observe their recommendations (Figure 4). As can be seen in this figure, in addition to overall and detailed similarity values, each recommendation is also accompanied by a confirmation label where users can indicate if they are willing to accept a suggestion. Table 1 shows the summary of this evaluation. Number of investigated suggestions indicates the number of suggestions that users further explored, either by clicking on the name of the recommended friend or on the similarity level label to investigate the underlying reasoning. These two features were added to the interface in order to investigate the effectiveness of the proposed
Table 1. Results of the evaluation Subject # of friends in the # of investigated # of accepted social graph recommendations recommendations/ # of total recommendation 1 346 3 0/14 2 409 0 1/1 3 311 0 5/50 4 170 1 6/18 5 107 0 12/35 6 235 4 10/88 7 212 0 0/68 8 302 3 0/73 9 123 5 0/39
4. Related work Ontology-based user modeling is especially important for designing systems that reason over multiple profiles, i.e. social adaptive systems.
883
GroupMe[9], is an interesting project in this regard which uses a semantic learner model based on the FOAF ontology. This work aims at automating the process of grouping students while also fulfilling individual’s personal needs and interests. The European project APOSDLE10 is a context-aware application aiming at discovering collaboration partners and adequate experts in a workplaceembedded e-learning environment. In terms of the recommendation process, this work contains a lot of commonalities with our proposed approach for finding suitable collaborative learners. However, the advantage of our approach is that it allows for integration of users’ data from a diverse set of social worlds, while all of the above mentioned systems deploy local ontologies to capture users’ data. There are also a few extractor applications capable of exporting users’ community and personal information into ontology-based formats. For instance, The ‘Flickr Exporter’ is a web application that exports users’ Flickr information using both FOAF and SIOC specifications[10]. Facebook FOAF Generator11 is another extractor that exports users Facebook data using FOAF specification. Although these applications allow for making social data portable and available, they are not yet integrated with learning services. In our proposed framework, we focus on utilizing such extracted data within learning environments to diminish the gaps between learning settings and everyday social activities.
application which can be deployed in educational settings. This would result in more users willing to try the application, which in turn would encourage more users to have their integrated ontological user model created, getting closer to opening the walls of social networks. Also, employment of other social data (such as microblog posts on Twitter) in the user model is another direction which requires further investigation. Physical location of the users is another determining factor in the suitability rank of a potential learning peer. Foreseeing such data in user profiles would allow for more in-time recommendations.
6. References [1]
[2]
[3]
[4]
5. Discussion and future work
[5]
It should be noted that the results of this evaluation are preliminary and obtained from the first round of user evaluation studies. The existing data in the database was very sparse, and less than 1% of friends' data for each of the participants was available in the database. In order to have a better understanding of the performance of the proposed framework, a thorough user evaluation with a more populated database should be performed. Overall, most of the users admitted that they can not accept a friend as a learning partner only due to the recommendation of the system. They indicated that other factors, such as personal interaction and knowledge of their friends also play an important role when choosing among the recommended list of friends for performing learning activities. Our future work deals firstly with the implementation of the prototype as a stand alone
[6]
[7]
[8]
[9]
[10]
Ch. A. Yeung, L. Liccardi, L. Kanghao, S. Oshani, T. Berners-Lee, “Decentralization: The Future of Online Social Networking,” W3C Workshop on the Future of Social Networking, Barcelona, 2009. M. Siadaty., C. Torniai, D. Gasevic, J. Jovanovic, T. Eap, and M. Hatala, “m-LOCO: An Ontology-based Framework for Context-Aware Mobile Learning,”In: Proc. of the 6th International Workshop on Ontologies and Semantic Web for Intelligent Educational Systems at 9th Intl. Conf. on ITS, Montreal, Canada, 2008. J. Jovanoviü, C. Torniai, , D. Gaševiü, S. Bateman, and M. Hatala, “Leveraging the Social Semantic Web in Intelligent Tutoring Systems,” In: Proc. of the 9th international conference on Intelligent Tutoring Systems, Montreal, Canada, pp. 563-572, 2008. D. Heckmann, T. Schwartz, B. Brandherm, M. Schmitz, and M. von Wilamowitz-Moellendorff, “Gumo – The General User Model Ontology,” In: Proc. of the 10th international User Modeling Conference, pp. 432, 428, 2005. M. Siadaty, F. Taghiyareh, “PALS2: Pedagogically Adaptive Learning System based on Learning Styles, ” In: Proc. of the 7th IEEE Intl. Conference on Advanced Learning Technologies, Niigata, Japan., 2007. S. Graf and Kinshuk, “Analysing the Behaviour of Students in Learning Management Systems with respect to Learning Styles,” Studies in Computational Intelligence, Vol. 93, pp. 53- 74, Springer, 2008. S. R. Viola, S. Graf, Kinshuk, T. Leo “Analysis of Felder-Silverman Index of Learning Styles by a DataDriven Statistical Approach, ” In: Proc. of the 8th IEEE International Symposium on Multimedia, IEEE Computer Society, pp 959 - 964., 2006. J. Jovanoviü, D. Gaševiü, C. Knight, G. Richards, “Ontologies for Effective Use of Context in e-Learning Settings,” Educational Technology and Society, Vol. 10, pp. 47-59, 2007. A. Ounnas, H. Davis, and D. Millard, “Semantic Modeling for Group Formation,” In: Conati, C., McCoy, K., Paliouras, G. (eds.) UM. LNCS (LNAI), Vol. 4511, Springer, Heidelberg, 2007. M. Rowe and F. Ciravegna, “Getting to Me - Exporting Semantic Social Network Information from Facebook,”. In: Proc. of Social Data on the Web Workshop, ISWC 2008, Karlsruhe, Germany, 2008.
10 11
http://www.aposdle.tugraz.at/ http://ext.dcs.shef.ac.uk/~u0057/FoafGenerator
884