Ontologically-enriched Unified User Modeling for ...

Ontologically-enriched Unified User Modeling for Cross-System Personalization Bhaskar Mehta† , Claudia Niederee† , Avare Stewart† , Marco Degemmis? , Pasquale Lops? , and Giovanni Semeraro? †

?

Fraunhofer IPSI, Darmstadt 64293, Germany Dipartimento di Informatica, Universit` a di Bari, Bari 70126, Italy {mehta,stewart,niederee}@ipsi.fraunhofer.de {degemmis,lops,semeraro}@di.uniba.it

Abstract. Personalization today has wide spread use on many Web sites. Systems and applications store preferences and information about users in order to provide personalized access. However, these systems store user profiles in proprietary formats. Although some of these systems store similar information about the user, exchange or reuse of information is not possible and information is duplicated. Additionally, since user profiles tend to be deeply buried inside such systems, users have little control over them. This paper proposes the use of a common ontologybased user context model as a basis for the exchange of user profiles between multiple systems and, thus, as a foundation for cross-system personalization.1

1

Introduction

Typically, personalization occurs separately within each system that one interacts with. Each system independently builds up user profiles, i.e. information about a user’s likes/dislikes, and uses this information to personalize the system’s content and service offer. Most of the personalization techniques [4] rely on either the implicit collection of information about users by tracking and analyzing their system usage behavior or the users explicitly providing information about themselves or giving feedback to the system. Such approaches have two major drawbacks: 1) investments of users in personalizing a system are not transferable to other systems; 2) users have little or no control over the information that defines their profile. Cross system personalization, i.e. personalization that shares information across different systems in a user-centric way, can overcome the aforementioned problems. Information about users, which is originally scattered across multiple systems, is combined to obtain maximum leverage. In this paper we present the Unified User Context Model(UUCM), an extensible, ontology based user context model, that is the foundation of the approach to cross-system personalization as an application of unified user models. The user becomes a hub and a switch, moving, controlling and synchronizing user profile data as part of a so-called Context Passport [3]. 1

This research was partially funded under the IST-2003-507173 Project VIKEF

2

Related Work

User models have been used in recommender systems for content processing and information filtering. Recommender systems, by observing preferences through interactions with users, keep summaries of their preferences in a user model, and utilize this model to adapt themselves to generate customized information or behavior. Systems incorporating models of users interest [2] and other cognitive patterns have been widely used to selectively filter information on behalf of users. Task models of user are considered important [1] based on the assumption that the goals of users influence their information needs. Along with the aforementioned modeling dimensions, environmental aspects are considered a key issue when modeling the user for improving the interaction between human and computer. Besides these more generic aspects of user modeling, there are also some efforts in standardizing user model related aspects, mostly in application-specific areas. The vCard specification and X.500, known as LDAP, are related standards. The IMS Learner Information Package (LIP) specification offers a data model that describes characteristics of a user needed for the general purpose of recording and managing learning related history, goals and accomplishments. In addition, the IEEE Public And Private Information (PAPI) specification was created to represent student records. The above standards are well known, but suffer from some drawbacks. vCard is suited for light weight user profiles like contact information or directories. While LDAP allows storing user information as entries made up of attributes, the directory schemas place restrictions on the attribute types that must be, or are allowed to be, contained in an entry. IMS and PAPI are more generic and based on standards like XML. However, they are not conceptually extensible. A unified user profiling format needs to take into account the domain knowledge that might be required for various applications. In order to support personalization across multiple systems, a broader understanding of the user is required as is also discussed in [3, 2].

3

The Unified User Context Model

The UUCM is a user context model that is structured along different dimensions and captures the fact that the user interacts with systems in different working contexts by structuring the model accordingly. In order to support cross-system personalization, the model has to be flexible and extensible enough to deal with the variations in personalization approaches and to incorporate the various aspects relevant for capturing the users’ characteristics and his current situation. The main building blocks for the UUCM is an extensible set of facets representing the characteristics of the user and his current context. We use the term facets instead of properties, because we do not only capture attribute value pairs, but also probabilities and qualifiers for facet values, thus giving a richer description as it is typical for frame-based languages. An extensible set of UUCM dimensions

enables the structuring of the facets into groups of user characteristics (e.g. facets related to cognitive pattern). In the context of UUCM, qualification of names as well as values of the facets is a crucial aspect. Both names and values may refer to vocabularies or ontologies, giving the possibility to connect to shared vocabularies, thus simplifying interpretation in a global (cross-system) context. Each UUCM facet is described by the following properties, part of which are optional: – Facet name - name of the UUCM facet to be described; – Facet qualifier - used to bind the facet to a defining vocabulary or ontology; – Facet value - value of the facet, which can be a simple literal as well a reference to a structured resource depending on the domain; – Value qualifier - a qualifier for the value(s) of the facet, i.e. it points to the vocabulary or ontology the value is taken from; – Value probability - a weight reflecting the probability of the facet value; – Facet dimension - each facet is assigned to one of the UUCM dimensions. The UUCM defines a meta-model for the concrete dimensions and facets used in the description of a user context model. For the cross-system personalization approach, that we are aiming for, it is assumed that this user context metamodel is published as a shared ontology and all participating systems rely on this model. In support of the UUCM, other ontologies are required: a facet ontology that defines the different facets, a dimension ontology that defines facet dimensions and, optionally, also ontologies for the facet values. More details are available in [3]. Information for filling profiles based on UUCM context models are gathered by observing the user interactions, such as items bought, rating provided by users on items, keywords searched, etc. An alternative form to collect values for the facets is to ask the user to fill in an initial form, where the user can enter information about his/her characteristics. The main problem of this process is that its validity depends on whether the users are willing to update the information regularly. A possible solution is to integrate/update the explicits interests given by the user with the automatic generation of profiles exploiting supervised learning techniques [5]. An example of collecting information for the Relationship Dimension can be found in [6].

4

UUCM in a Real World Cross System Scenario

There are three objectives which cross-system personalization needs to address: 1) broader user models that can cope with the variety in user modeling, 2) handling heterogeneous personalization systems and approaches, and 3) giving more control to the user. In line with these objectives, we claim that user profiles should be stored closer to the user. However, maintaining user profiles on the user’s side presents some challenges. Interacting with multiple information systems may lead to a large amount of interaction data. Since the individual system best understands the local interactions this should be done within the individual personalization engine and only higher level descriptions of users should be

Fig. 1. User Profile Ontology for MSN and Yahoo Messenger

Fig. 2. (a)John’s Profile for MSN and Yahoo(b)John’s Yahoo Profile in UUCM

exchanged between the information system and the unified user profile, which we call the Context Passport. The exchange of such information requires a negotiation between activities that an information system can perform and those activities that the user context outlines. The Cross System Communication Protocol (CSCP) provides a platform for such negotiations. 4.1

Example Scenario for Unified Profiling in Instant Messaging

In an examle scenario, two Instant Messaging (IM) applications: the MSN and Yahoo Messenger are considered. Essentially the user profiles of these IM networks are lists of contacts that a person is explicitly connected with. While the profiles of these applications focus on modeling relationship aspects, the profiles have a different structure to model this user information. MSN Messenger allows user to have custom names, but has predefined categories for classifying contacts (Friends, Family, etc). Yahoo Messenger does not allow custom names, but allows creation of new categories. Furthermore, with MSN Messenger, it is possible to have a person on your list, yet to block them from contacting you. Yahoo has different structure, and to block or ignore a contact, this person has to be deleted from the list and then added to a separate ignore list. Most user profile formats like PAPI and IMS will fail to store a unified profile completely. As an concrete example, we take up a simple user profile for a fictitious user John’s MSN and Yahoo (Figure 2) profiles. Both these models use common concepts, which can be represented by a common vocabulary consisting of concepts im:Contact,

im:Contact-List, im:Group related by relationships discussed in Section 3, as shown in Figure 1(a),(b). Analyzing the domain for IM user profiles, we reach a common model shown in Figure 1(c), composed of only two facets. The same common model can be used for AOL, ICQ and other IM applications with minor modifications to the common vocabulary. Thus this model represents the domain model for the IM domain. Further additions to the model are possible, but for this example, we assume the profile to be composed of only a categorized list of contacts. Using this understanding, we can represent both the profiles in the UUCM format. Figure 2(b) shows how the Yahoo profile can be represented. The MSN profile can be similarly represented. We note that the profiles for these applications lie completely in the Relationship dimension. By using a common format for representing users, these two applications can more easily interoperate, and one application can connect to both networks while maintaining a common profile. Similar models can be constructed for eCommerce websites and personalized content providers.

5

Conclusions

This paper proposes the use of a unified user profile format, which can be extended for use with multiple applications, and potentially be used to exchange common information between multiple systems. The UUCM provides a basis for the realization of cross-system personalization approaches that enable the exchange and reuse of user profiles across different systems. UUCM components refer to common vocabularies or ontologies in order to give the possibility to interpret the user models in the different contexts.

References 1. C. Kaplan, J. Fenwick, and J. Chen. Adaptive hypertext navigation based on user goals and context. User Modeling and User-Adapted Interaction Journal, pages 193–220, 1993. 2. A. Kobsa. Generic user modeling systems. User Modeling and User-Adapted Interaction Journal, 11:49–63, 2001. 3. C. Nieder´ee, A. Stewart, B. Mehta, and M. Hemmje. A multi-dimensional, unified user model for cross-system personalization. In Liliana Ardissono and Giovanni Semeraro, editors, Proceedings of the AVI Workshop on Environments for Personalized Information Access, Italy, pages 34–54, May 2004. 4. A. Pretschner and S. Gauch. Personalization on the web. Technical Report ITTC-FY2000-TR-13591-01, Information and Telecommunication Technology Center (ITTC), The University of Kansas, Lawrence, KS, December 1999. 5. G. Semeraro, M. Degemmis, and P. Lops. User profiling to support internet customers: What do you want to buy today? Informatica, 26(4):407–418, 2002. 6. A. Stewart, C. Nieder´ee, B. Mehta, M. Hemmje, and E. Neuhold. Extending your neighborhood-relationship-based recommendations for your personal web context. In Z. Chen, H. Chen, Q. Miao, Y. Fu, E. Fox, and E Lim, editors, Proceedings of the 7th International Conference on Asian Digital Libraries, volume 3334 of LNCS, pages 523–532, 2004.