Building Blocks for User Modeling with data from the Social Web

Abstract

In this paper we present the so-called Grapple User Modeling Framework (GUMF) which provides user modeling functionality via the Web to applications that would like to o er personalization and adaptation features to their customers. GUMF introduces the notion dataspaces which create a logical view on (possibly distributed) user data and provide advanced contextualization and reasoning mechanisms. We showcase the Mypes service that exploits GUMF dataspaces to connect, aggregate, align and enrich user pro le information form the Social Web.

Introduction

ask (query): SPARQL, SeRQL, Grapple Query tell (store): RDF, Grapple statements, XML

Client (Application) lltr e ro Store Module n o C F UM Query Engine G Authentication

Access Control GUMF Grappple User Modeling Framework

Core

Reasoning Logic Reasoner Plug-ins

Plug-ins & Rule Repository

Repository

Repository query/store Repository Web

GUMF Admin

Interface Configuring plug-ins, dataspaces, reasoning

rules Administrator pro le information form social networking services (Facebook, LinkedIn), social media services (Flickr, Delicious, StumbleUpon, Twitter, Blogspot) and others (Google). Mypes enables users to overview their distributed pro les and via GUMF it allows application developers to integrate aggregated, semantically enriched pro le data in RDF or vCard format via a lightweight REST interface into their applications to provide personalization and adaptation functionalities. 2

Grapple User Modeling Framework 5 cf. GUMF help pages: http://pcwin530.win.tue.nl:8080/grapple-umf/help/ 1. get other accounts of user

2. aggregate public profile data

Account Mapping Social Web Aggregator

3. Map profiles to target user model

Profile URI

Aggregated,

enriched profile (e.g., in RDF or vCard) Mypes dataspace Profile Alignment Semantic Enhancement Cache GUMF

4. enrich data with semantics

FOAF vCard

WordNet® Blog posts: Bookmarks: Other media:

Social networking profiles: ins directly in uences the behavior of dataspaces. Further, administrators can adjust the plug-ins and reasoning rules to their needs. In the next section we will highlight GUMF dataspaces and describe plug-ins we implemented that support aggregation of user data from the Social Web. These plug-ins and the corresponding dataspace enable Mypes to visualize the distributed user data traces. Moreover, as GUMF dataspaces can be shared across di erent client applications, we also enable other client applications to bene t from the pro le aggregation. 3

Mypes: Enriching User Pro les via GUMF To illustrate the functionality of GUMF dataspaces we present the Mypes service which is a GUMF client application that particularly highlights GUMF's ability to support the task of gathering information about users for user adaptive systems [ 1 ]. In this section we rst present the GUMF components we implemented to allow for pro le aggregation and enrichment before we describe Mypes features in more detail. 3.1

Linkage, Aggregation, Alignment and Enrichment of User Data The Grapple User Modeling Framework aims to provide a uniform interface to user data that might be distributed on the Web. To feature access to distributed user data and to align and enhance the data, GUMF and the corresponding GUMF components depicted in Figure 2 respectively perform the following steps. 1. Account Mapping Given a user's URI of an online account, the account mapping plug-in gathers other online accounts of the same user by exploiting the Google Social Graph API6, which provides such mappings for all users who linked their accounts via their Google pro le, for example: 6 http://code.google.com/apis/socialgraph/ x x x x x x x Facebook LinkedIn Twitter Blogspot Flickr Delicious Stumble Last.fm Google

Upon x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x traditional

pro le attributes nickname first name last name full name profile photo about email (hash) homepage blog/feed location locale settings interests education affiliations industry tag-based pro le posts

The four plug-ins can be plugged into dataspaces. We applied them to the Mypes dataspace which forms the basis for the Mypes service. 7 http://wordnet.princeton.edu/ (a) traditional pro le

(b) ltering tag-based pro le Mypes exploits GUMF to provide an interface that is valuable for casual users, who would like to overview their distributed pro le data. Further, it makes the enriched public pro le data that is available via the Mypes dataspace (see Figure 2) also available via a lightweight RESTful interface (in FOAF and vCard format). Hence, thanks to GUMF and the plug-ins presented above, Mypes just has to focus on representing the public pro le data { either for end-users or for systems that require additional information about their users.

Given the Google pro le URI of a user, Mypes queries GUMF and the Mypes dataspace particularly to obtain the corresponding aggregated, enriched pro le data. It then enables users to overview their public pro les, for example, users can inspect particular pro le values, analyze the completes of their pro les on a bar chart and understand to what kind of information about themselves is publicly available. Mypes thus raises the users' awareness of their pro le data distributed across di erent services on the Web.

Figure 3(a) shows an example of an aggregated Mypes pro le, namely the traditional pro le attributes gathered from the diverse services (see Table 1). When accessing http://mypes.groupme.org/mypes/user/116033/rdf the FOAF prole in RDF/XML syntax is returned. Mypes exports all available values for a pro le attribute, e.g., if a user speci es her name di erently at the di erent services then all these di erent values are provided.

GUMF also connects the tagging activities users perform in the di erent tagging systems. As the semantic enrichment plug-in (see Fig. 2) extends tag assignments with meta-information that states to which WordNet category the corresponding tag belongs to, it is possible to lter the aggregated tag cloud of a user according to WordNet categories. For example, Figure 3(b) shows the aggregated tag cloud ltered so that only tags related to locations are displayed. For this kind of tag cloud, Mypes provides an alternative visualization: tags related to locations are mapped to country codes (using the GeoNames Web service8), which are sent to Google's visualization API to draw a geographical intensity map that highlights those countries that are frequently (possibly indirectly) referenced by tags in the pro le (see bottom in Figure 3(b)). Mypes also features RDF export for these (speci c facets of) tag-based pro les using the Tag Ontology9 and SCOT10 vocabulary. 4

Conclusion

In this paper we described how the Grapple User Modeling Framework (GUMF) enables to enrich user pro les with user data gathered from the Social Web. We presented the Mypes service that exploits this functionality to enable casual users to overview their distributed pro les and allows adaptive applications to re-use the aggregated and enhanced pro le information for their own purposes. In our future work we plan to evaluate the actual bene t for end-users of adaptive applications which make their adaptation decisions based on the enriched pro les produced by GUMF and Mypes respectively.

Acknowledgments This work is partially sponsored by the EU FP7 project GRAPPLE (http://www.grapple-project.org/). 8 http://www.geonames.org/ 9 http://www.holygoat.co.uk/projects/tags/ 10 http://scot-project.org/scot/

References

1. Jameson , A. : Adaptive interfaces and agents. The HCI handbook: fundamentals, evolving technologies and emerging applications ( 2003 ) 305 { 330

2. Abel , F. , Henze , N. , Herder , E. , Krause , D. : Interweaving public user pro les on the web . In Kobsa, A. , Bra , P.D., eds.: International Conference on User Modeling, Adaptation and Personalization (UMAP), Hawaii , USA, Springer ( June 2010 )

3. Bizer , C. , Heath , T. , Berners-Lee , T. : Linked data - the story so far . International Journal on Semantic Web and Information Systems (IJSWIS) 5 ( 3 ) ( 2009 ) 1 { 22

4. Heckmann , D. , Schwartz , T. , Brandherm , B. , Schmitz , M. , von WilamowitzMoellendor , M.: Gumo - the general user model ontology . In Ardissono, L., Brna , P. , Mitrovic , A., eds.: User Modeling. Volume 3538 of Lecture Notes in Computer Science., Springer ( 2005 ) 428 { 432

5. Yudelson , M. , Brusilovsky , P. , Zadorozhny , V.: A user modeling server for contemporary adaptive hypermedia: An evaluation of the push approach to evidence propagation . In Conati, C. , McCoy , K.F. , Paliouras , G., eds.: User Modeling. Volume 4511 of Lecture Notes in Computer Science., Springer ( 2007 ) 27 { 36

6. Assad , M. , Carmichael , D. , Kay , J. , Kummerfeld , B. : PersonisAD: Distributed, active, scrutable model framework for context-aware services . In LaMarca, A., Langheinrich , M. , Truong , K.N., eds.: Pervasive Computing. Volume 4480 of Lecture Notes in Computer Science., Springer ( 2007 ) 55 { 72

7. Abel , F. , Heckmann , D. , Herder , E. , Hidders , J. , Houben , G.J. , Krause , D. , Leonardi , E., van der Slujis, K.: A framework for exible user pro le mashups . In Dattolo, A. , Tasso , C. , Farzan , R. , Kleanthous , S. , Vallejo , D.B. , Vassileva , J., eds. : Int. Workshop on Adaptation and Personalization for Web 2 .0 at UMAP '09, CEUR Workshop Proceedings ( 2009 ) 1 { 10