Multi-application systems provide opportunities to gather user data from outside of the individual application itself. Aggregated user data may be useful to address the cold- start problem as well as the sparseness of user data. Connecting data from different sources and services from distributed application environments is in line with advancements in multi-core and multi-tasking architectures. While there are theoretical and software frameworks for distributed user modeling, assessments of modeling
23
techniques are almost always reported in terms of single applications. With a better understanding of the user interests, adaptive systems can provide better personalization. Sharing and reusing the user model information between applications can bring the advantage for profile providers as well as profile consumers by enriching the user models.
Current systems that provide personalized services to users are mostly develop their own proprietary application environments in ad-hoc manner as a part of a specific application requirement (Dim and Kuflik 2012). These proprietary user models are of evidence in system developer’s focus on specific characters of their users in order to provide a specific service (e.g., movie recommender system). Over the years, these user models and their application environments are moved from providing complete, monolithic solutions in user modeling servers (Kobsa 2007) to dynamic solutions in the areas of interoperability and interlinking (Leonardi, Abel, Heckmann, Herder, Hidders et al. 2010, Carmagnola, Cena and Gena 2011). User models can be developed by adapting the content consumed or produced by the user, and their specific task, background, history and information needs (Renda and Straccia 2005). These models can bring users’ attention to valuable content via personalized presentations. (Berkovsky, Kuflik and Ricci 2008) presented a definition of mediation to introduce cross-system personalization using the technique to integrate and match user modeling data. Recognizing the user interest based on observed user activity is confounded by idiosyncratic work practices. As a result, systems that aggregate evidence of user interest from a wide variety of sources are more likely to build a robust user interest model.
24
There are two main approaches to user modeling in a component-based architecture. These vary based on the degree of centralization of the user models. Decentralized (or distributed) user modeling had its roots in agent-based architectures; here fragments of user model are kept and maintained by each independent application. Another important distinction among user modeling approaches is whether the model is represented via features or content. Feature-based user models define a set of feature- value pairs representing various aspects of the user, such as interest in a specific category or a level of knowledge in a specific area. Content-based approaches take into account the user's area of interest, as an example, the textual content of documents the user has previously indicated as relevant. These systems generate recommendations by learning user needs with the analysis of available rated content.
In a centralized approach, the integrated user model is stored in a central server and the model is then shared across several user-adaptive applications. Apart from alleviating the applications re-inventing the wheel, centralized user model give an opportunity to share the same user model between several applications. These include generic user modeling servers such as IPM (Bae, Kim, Meintanis, Moore, Zacchi et al. 2010), CUMULATE (Brusilovsky, Sosnovsky and Shcherbinina 2005, Yudelson, Brusilovsky and Zadorozhny 2007), UMS(Kobsa and Fink 2006) and PersonisAD (Assad, Carmichael, Kay and Kummerfeld 2007) as well as framework developed for mashing up profile information (Abel, Baumgartner, Brooks, Enzi, Gottlob et al. 2005, Abel, Henze, Krause and Plappert 2008, Abel, Heckmann, Herder, Hidders, Krause et al. 2009, Houben, Leonardi and Van Der Slujis 2009) to facilitate aggregated user data.
25
PersonisAD is a distributed framework for building ubiquitous computing applications. It defines a user model based on data gathered from different sensors and combines their preferences using resolvers to provide a tailored experience. CUMULATE is a generic modeling server developed for a distributed E-Learning architecture to help students select the most relevant self-assessment quizzes by inferring their knowledge of a predefined set of topics based on authored relationships among activities in the educational applications and topics. UMS is a user modeling server based on the LDAP protocol which allows for the representation of user interests using a predefined taxonomy for the application domain.
Attempts to bridge user models in various systems require conversion of the user models data between various applications, domains and adhering to semantic representations (Martinez-Villaseñor, Gonzalez-Mendoza and Hernandez-Gress 2012). Some of these have been done using mapping techniques of user models (Vassileva, McCalla and Greer 2003, Bennani, Chevalier, Egyed-Zsigmond, Hubert and Viviani 2012) and more recently using machine learning methods (Berkovsky, Kuflik and Ricci 2008). These user modeling systems do not easily comply with a standard format, technique or vocabulary to enable user modeling interoperability (Martinez-Villaseñor, Gonzalez-Mendoza and Hernandez-Gress 2012).
This dissertation is based on the immediate need for approaches to setup user interests and the distinctions between them to be constructed based on the content encountered rather than pre-agreed upon by the contributing applications.
26