User Modeling, Personalization and Adaptivity: Potential Keys to Success for FIBAC ProjectPaper
Ludovico Solima, Italy , Cosimo Birtolo, Italy, Simona Acanfora, Italy, Massimiliano Minei, Italy
The Italian research project FIBAC (Fruizione Innovativa dei Beni Artistici e Culturali – Innovative Fruition of Cultural Heritage Assets) is co-funded by the Italian Ministry of Education, University and Research and conducted by Poste Italiane, University of Salerno (Department of Information Engineering, Electrical Engineering and Applied Mathematics and CRMPA – Pure and Applied Mathematics Research Centre) and some small and medium Italian enterprises, i.e., Protom Group, Space, Meta, Nexsoft, and Lit Com. The project aims at making a technological system prototype able to create personalized visit paths in museums and art galleries and to deliver information about art objects specifically tailored to the user’s preferences and needs, following a narrative structure.
In general terms, this goal responds to the need of redefining the relationship between a museum and its visitors, increasingly centered on the user and mindful of his/her prior knowledge and experiences. In the framework of such a wide project, this paper intends to deepen three major aspects: user modeling, personalization and adaptivity.
User modeling. The key to customizing a service is creating a user model. FIBAC intends to do this by offering reliable and meaningful services for the museum field. For this purpose, a deep study was carried out at both at national and international levels to analyse specific characteristics useful to creating a user model that can be applied for customization, without redundancies and computationally manageable; in short, FIBAC wants to offer a user model that can outline cultural visitors’ profiles that are meaningful for the visit, both on the basis of information explicitly provided by the user, and on the basis of implicitly gathered data, by analysing a user’s behaviour or by importing information from external systems (e.g. social networks).
Personalization. The personalization phase started from the study of the methodologies that analyse a user’s requests in natural language that adopt users’ profiles, indoor tracking techniques and automatic recognition of art objects. A further step consisted of improving the procedures for planning custom museum paths and, in close connection with them, for personalizing museum guided tours, according to a pull/push logic, which respectively take into consideration the user’s requests and the information suggested on the basis of the user’s profile and his/her location.
Adaptivity. The whole outlined process is enhanced by repeated use; with factual and direct information referred by the user, these repeated uses make up the basis of an adaptive system that knows how to model itself according to users’ changes, progress, developments and afterthoughts both during the visit and afterwards, and also in future visits to other museums.
FIBAC is currently an ongoing project (it started in 2011 and it is expected to end in 2015); once closed, it will propose several innovative and distinctive aspects with respect to currently available cultural heritage communication solutions. Current solutions are mainly based on the use of ICT technologies (e.g. three-dimensional reconstructions, mobile devices, audio and video guides) and do not have a deep impact on a visitor’s experience as they leave him/her in an essentially passive role. By contrast, FIBAC’s innovative technological solutions aim to enhance the centrality of visitors and of their experiences through the interpretation of their personal needs.
2. The FIBAC Project
The purpose of the project is to define and validate a new model based on cultural re-mediation and, most of all, to define new methodologies, techniques and prototypes for personalised and adaptive fruition experiences of cultural heritage in real and virtual contexts.
On the basis of the models, methodologies and techniques to be defined, two prototypes will be designed and developed to assess the research results both in real and virtual museums. The project results present several innovative aspects with respect to the current market solutions for cultural heritage use. In other words, FIBAC aims at emphasizing the centrality of the user with respect to ICT during the experience. In FIBAC’s vision, the experience of cultural heritage should ensure a “dialogue” between the user and the museum removing the current boundary between action and contemplation.
From a technological point of view, the project objective relates to the definition of a system for dynamic generation of personalised paths in museums and art galleries (both real and virtual) in order to deliver personalized and contextualized information to better understand the artworks following a narrative structure.
Cultural heritage, indeed, has been for several years a privileged domain for personalization systems and recent research recognizes how important it is to offer personalized and individual support for users. In particular, it has been stated (Falk, 2009) that museum visitors represent a particularly heterogeneous group and that their visit is an experience consisting of physical, personal, and socio-cultural aspects. Consequently, to offer support to any user, it is necessary to take into consideration contextual and personal features and, moreover, a visitor’s behavior may also change during the visit and this may require an adjustment of the information offered.
Starting from user modeling and personalization, FIBAC wants to answer to the needs of redefining the relationships between the museum and the user, centered on the user and her/his past experiences and background. On the basis of the models, methodologies and techniques to be defined, two prototypes will be designed and developed to assess the research results both in real and virtual museums.
Moreover, the project investigates Genetic Algorithms, which are suitable to analyze optimization problems where the target function is not easily derivable, as it is in FIBAC’s case. The project also studies text mining and concept extraction methods and algorithms, in order to allow the interpretation of explicit requests made by the user in natural language. A further characterizing technological aspect is the definition of methodologies and techniques to create an automatic recognition system of artworks in order to allow both the link of information about a specific artwork and the identification of the place where the visitor is, through a map of the artworks available in the museum. The project investigates and prototypes several techniques of image recognition (giving a particular attention to the automatic recognition of artworks), among which are classification techniques such as Naïve Bayes and Artificial Neural Networks.
3. User modeling, personalization and adaptivity
The general aim of FIBAC is a technological system for the creation of personalized tours in museums and art galleries and the delivery of customized information for the interpretation of art works in a narrative structure.
This technological system has a double implementation: a Web portal and an application for smartphones and tablets. Although it is one of the main efforts in the project, the way in which the informative narrative structure will be delivered in FIBAC is out of scope here and will not be treated.
Generally speaking, the delivery of content in a museum can take place before, during or after the visit or, from another point of view, it can be offered either offline or online. More specifically, in the case of FIBAC, we focus on the information given during the visit, on the one hand, and the information given online on the other.
It is worth recalling that FIBAC will tackle the two phases “before” and “after” the visit mainly through the Web portal.
Again, in the cultural heritage field it can be generally observed that the task of content customization is usually interpreted in a very simple way, thanks to the design of different tours for different categories of visitors (for example, for children, art history beginners, and so on). Things are very different indeed among museums and among countries: in some cases (in Italy, for example) (Xanthoudaki, 2000), tours for specific categories of visitors are indeed still quite rare.
Considering all these aspects, the main question that our research group tried to answer was how to make the information proposed to visitors appropriate to their needs, expectations, and interests.
First of all, we focused on the visiting process as a function of three items: personal background, physical context, and relational dimension (Falk & Dierking, 1992).
As for the first item (personal background), we designed an accurate user model able to gather information about FIBAC users in two ways: explicitly and tacitly.
The major aim of the visitor modeling in FIBAC is gathering as much information about the user as possible in a non-intrusive way, in order to customize both the use of content and its presentation on the basis of a visitor’s preferences and knowledge. Therefore, user modeling has a core task within FIBAC, because it has to be able to select the most suitable art works for the visitor’s profile. In the literature, several branches of the research have experimented with user modeling both for recommendation and adaptive systems (Reinecke, Reif & Bernstein, 2007; Bohnert & Zukerman, 2009; Blanchard & Allard, 2010; Reinecke, Schenkel & Bernstein, 2010; Antoniou & Lepouras, 2010). Among explicit information about users, we plan to have a registration procedure during which basic data will be collected; besides, we will have more detailed information that users can edit in a “my profile” section. For example, in the “my profile” section, users will be allowed to specify a range of interests or a self-evaluation about expertise in different subjects.
We will gather tacit information via social media accounts that could be used to enter FIBAC. Moreover, the use of FIBAC will allow us to track and analyze visiting behaviours and indoor localization – two important aspects of the whole project. We won’t go more in depth about localization systems here; suffice it to say that they will be based either on Wi-Fi triangulation or on image recognition, or QR-code, according to the specificities of each museum.
The second item with an important impact on the visiting process is physical context; in this case, we refer to the physical setting, art works on exhibition, and information available in the museums. There are several studies dealing with this topic that try to identify the correlations between the features of context and behaviors of use (Goulding, 2000; Solima, 2013).
Lastly, we have the relational dimension of the visit (and of the visitor, of course), which can be considered through the analysis of group behaviors: as a matter of fact, there are significant differences if a visit experience is made in a family context or in a school context, or in an informal context of friends and relatives. It’s well known how social interaction during a visit can change its nature in a very deep way (Cone & Kendall, 1978; Bitgood, 1993; Dierking, 1994; Blud, 1990; Galani, 2003; vom Lehn, 2006).
All these aspects produce an inter-individual variety (that’s to say, among different users) and an individual variability (the same individual during the time) in visiting behaviors.
This brought us to face content customization issue through an innovative and dynamic approach: we are trying to make dynamic content as a function of Genetic algorithms and user ratings, for example about authors, subjects, ages, materials, techniques, etc.
Genetic Algorithms will work first on the user’s information (the aspects we have illustrated before: registration, profile, social network, visiting behaviors); second, on other users’ information (based on the behavior and needs of similar visitors, as Amazon does); and third, on relations and connections among art works in terms of authors, subjects, ages, and so on.
In the following pages, these three topics (user modeling, customization and adaptivity), now presented in a theoretical framework, will be analyzed in terms of their technical aspects, as faced in FIBAC project.
4. User modeling
As noted by Kobsa (1994) and Brusilovsky (1996), the study of user modeling focuses on several research areas such as Natural-Language Dialogue Systems, Knowledge Representation and HCI (Human Computer Interaction). User modeling has been used in adaptive and interactive systems. The user’s profile can be considered “a description of users or user groups. This description includes user goals, tasks, preferences and/or background knowledge related to the problem domains” (de Koch, 2001).
A user model is a representation of the knowledge and personal characteristics which the system believes that a user possesses. It is different from both the actual knowledge possessed by a user and the knowledge employed by system designers (de Koch, 2001).
FIBAC’s user model has an overlay approach aimed at expressing the implicit characteristics of each user through the use of knowledge of cultural heritage. In detail, the concepts describing the artworks in the museum are the same as the user’s cultural preferences (implicitly inferred or explicitly expressed by the user himself) and this allows there to be a personalized path within the museum’s rooms. Brusilovsky considers overlay models as a powerful and flexible tool that can measure independently user knowledge on different topics (de Koch, 2001). According to this view, the weighted overlay (Brusilovsky, 2007) within FIBAC’s model takes into account the weighted user’s preferences for each topic.
Extending this approach, the user’s model in FIBAC expresses how strong cultural influence derives from the place where the visitor lives and his previous knowledge of specific themes (for instance, “User Mario has a poor knowledge of Roman Architecture”). In FIBAC, the user modeling of the visitor will gather information about the interaction between the visitor and the system during the tour (e.g., requests of more information about a specific artwork or the skipped information). All this information will be used after the visit in order to suggest further contents directly or indirectly connected to those included in the museum.
Adaptive systems suggest the implementation of an overlay-based model, as an ontology which is connected with a model of knowledge (Dicheva, 2010). In detail, FIBAC intends to use ontologies for a user’s profile, enabling the possibility of inferring new information starting from the acquired data. This avoids the cold-start problem which characterizes those profiles with little information or some software systems at the initial stage (Cena, 2011).
In FIBAC, a further advantage offered by Web Semantic Technologies, and ontologies in detail, is the enabling of interoperability of the cultural visitor’s model. In other words, it will be possible to share or port some information included in the user’s model.
Recommendation Systems are a way to improve personalization by giving personalized suggestions and by providing users with a customized search of items. In other words, recommendation systems are aimed at helping users in search of interesting items among a large set of items within a specific domain by using knowledge about the user’s preferences in the domain. These were introduced in 1992 by means of the Tapestry project (Goldberg, 1992) but, recently, have largely been adopted in different domains.
Among recommendation techniques, Collaborative Filtering (CF) has gained great success in the most of the application and has been proven to be one of the most successful techniques used (Adomavicius & Tuzhilin, 2005). Recently model-based CF, such as Clustering CF, which uses the user-item database to infer a model which is then applied for predictions, are investigated and discussed in order to outperform traditional memory-based CF. Different works (Birtolo et al., 2011, Huang & Yin, 2010) prove that these algorithms improve the quality of predictions, in particular when sparse data is considered.
In FIBAC, recommenders are adopted when a thematic path is generated and when a tourist plans his visit remotely or on-site by means of a Web Application or a Mobile App respectively.
FIBAC detects two customization phases: (i) Macro-Customization, and (ii) Micro-customization. In a first step, a “macro-customization” of the use is made, that is to say that, on the basis of some features of the user’s profile and of the aim of the visit (explicit request, time of the visit and map of the museum), a set of cultural artifacts is defined and a visit path is suggested. In a second phase, a “micro-customization” of the use is put into practice: that is to say that, according to some features of the user’s profile, specific content is defined to be associated to each artwork on the path.
The analysis of how to manage the adaptivity of an itinerary in the FIBAC project has taken into account recommendation systems as useful tools to select artworks on the basis of both explicit and implicit preferences of the user. These systems represent a useful tool for user modeling and the customization of the visit. Still, recommendation systems give a too simple vision of the museum visit, ignoring the role that the sequence and the path followed play for full enjoyment of suggested artworks. Even though solutions have been proposed that consider also the order in which suggestions are given, they do not adapt themselves to the complexity and variety of information necessary to generate a museum path. Moreover, these approaches do not fit well with the adaptivity features necessary to FIBAC. The elements and their composition in a path are driven by rational criteria which consider constraints, preferences, and contextual information. In particular, these are used to change the path in response to the information gathered during the visit.
During their use, it is possible to assess the quality of the tour, analyzing for example if the sequence and the set of artworks previously proposed has been adopted.
Different adaptation strategies have been studied in the project in order to suggest a different path or different items within the planned path or extremely the system can regenerate a new path with the lesson learned by user interaction.
7. Conclusions: a new starting point
We learned several concepts during the research and development of the prototype FIBAC. These concepts can be summarized in ten points, which represent the major guidelines of the project and can be important starting points for future research:
- Communication in museums is addressed to a varying public and therefore FIBAC’s major objective is personalization;
- Acquiring information about interests and preferences of users, as well as about what they like less, is a central issue for good personalization both of the itinerary and of the information given;
- The adaptivity of the services offered is meant as the capacity to respond to the events during the tour (change of the itinerary, particularly long stops before specific artworks, artworks skipped even if suggested, requests for further information, ratings given to artworks and/or to the information received); all this is a further qualifying aspect of project FIBAC;
- FIBAC speaks the user’s language, and not viceversa, meaning that the system will try to use the language already used and known by the visitor, so that he will not have to acquire new expertise to enjoy the artwork;
- As far as possible, the artworks presented in FIBAC will be told by “stories” and not by a list of facts and information; it is an effort aiming at giving a museum tour the dimension of a more complex and complete experience.
- The same narrative approach will allow, where possible, to create personalized itineraries;
- The multimedia component of the service must have an explanatory role, not diverting the attention of the visitor and not becoming a mere technological showing off;
- Experts in this field should take a leading role in validating content in order to preserve the scientific accuracy of the information given;
- During and after the visit, the suggestion of new concepts and ideas can arouse new interest in a visitor; in other words, if it is generally true that a visitor looks for confirmation of his/her knowledge, it is also true that it is important to give new information and create new interests;
- As far as possible, during the visit the system must push the visitor to watch artworks and, at the same time, to watch and enjoy their context, avoiding being captivated by the device.
This is an ambitious approach, that the currently available technologies nevertheless make possible to realize. Therefore, now is a good opportunity to use these technologies in the cultural heritage field.
This work was partially supported by Italian Ministry of University and Research (MIUR) under the FIBAC Project PON01-02705.
Adomavicius, G. & A. Tuzhilin (2005). “Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions”. In IEEE Transactions on Knowledge and Data Engineering 6, 734-749. doi: 10.1109/TKDE.2005.99.
Antoniou, A. & G. Lepouras (2010). “Modeling visitors’ profiles: A study to investigate adaptation aspects for museum learning technologies”. Journal on Computing and Cultural Heritage 2, article 7, 19 pages. doi: 10.1145/1841317.1841322.
Birtolo, C., Ronca, D., Armenise, R. & M. Ascione (2011). “Personalized suggestions by means of Collaborative Filtering: A comparison of two different model-based techniques”. In Nature and Biologically Inspired Computing (NaBIC), 2011 Third World Congress, 444-450,
Bitgood, S. (1993). “Social Influences on the VisitorMuseum Experience”. Visitor Behavior 3, 4-5.
Blanchard, E.G. & D. Allard (Eds) (2010). The Handbook of Research in Culturally-Aware Information Technology: Perspectives and Models. Hershey, PA: IGI Global. doi:10.4018/978-1-61520-883-8.
Blud, L.M. (1990). “Social Interaction and Learning Among Family Groups Visiting a Museum”. Museum Management and Curatorship 9, 43-51.
Bohnert, F. & I. Zukerman (2009). “Non-Intrusive Personalisation of the Museum Experience”. In UMAP-09. Proceedings of the 17th International Conference on User Modeling, Adaptation, and Personalization: formerly UM and AH, 197-209.
Brusilovsky, P. & E. Millán (2007). “User Models for Adaptive Hypermedia and Adaptive Educational Systems”. In The Adaptive Web. Lecture Notes in Computer Science. Springer-VerlagBerlinHeidelberg, Vol. 4321, 3-53. doi: 10.1007/978-3-540-72079-9_1.
Cena, F., Likavec, S. & F. Osborne (2011). “Propagating User Interests in Ontology-Based User Model”. In Artificial Intelligence Around Man and Beyond. Lecture Notes in Computer Science, Springer BerlinHeidelberg, Vol. 6934, 299-311. doi: 10.1007/978-3-642-23954-0_28.
Cone, C.A. & K. Kendall (1978). “Space, Time, and Family Interaction: Visitor Behavior at the ScienceMuseum of Minnesota”. Curator 3, 245-258.
de Koch, N.P. (2001). Software Engineering for Adaptive Hypermedia Systems – Reference Model, Modeling Techniques and Development Process. PhD thesis. Ludwig-Maximilians University of Munich.
Dierking, L. (1994). “Rôle de l’interaction sociale dans l’expérience museale”. Publics & Musées 5, 19-43.
Falk, J. & L. Dierking (1992). The museum experience. Washington, D.C.: Whalesback books.
Falk, J. H. (2009). Identity and the Museum Visitor Experience. Walnut Creek, CA: LeftCoast Press.
Galani, A. (2003). “Mixed RealityMuseum Visits: Using new technologies to support co-visiting for local and remote visitors”. Museological Review Extra Special Issue 10, 1-15.
Goldberg, D., Nichols, D., Oki, B. M. & D. Terry (1992). “Using collaborative filtering to weave an information tapestry”. Commun. ACM 35, 12, 61-70. doi: 10.1145/138859.138867.
Goulding, C. (2000). “The museum environment and the visitor experience”. European Journal of Marketing 3/4, 261-278.
Huang, C. & J. Yin (2010). “Effective association clusters filtering to cold-start recommendations”. In 2010 Seventh Int. Conf. on Fuzzy Systems and Knowledge Discovery 5, 2461–2464.
Kobsa, A. (1994). “User modeling and user-adapted interaction”. In C. Plaisant (Ed.) Conference Companion on Human Factors in Computing Systems (CHI ’94). New York, NY: ACM, 415-416.
Reinecke, K., Reif, G. & A. Bernstein (2007). “Cultural User Modeling With CUMO: An Approach to Overcome the Personalization Bootstrapping Problem”. In Workshop on Cultural Heritage on the Semantic Web. ISWC.
Reinecke, K., Schenkel, S. & A. Bernstein (2010). “Modeling a User’s Culture”. In E.G. Blanchard & D. Allard (Eds) The Handbook of Research in Culturally-Aware Information Technology: Perspectives and Models. Hershey, PA: IGI Global. doi: 10.4018/978-1-61520-883-8.ch011.
Solima, L. (2013). “Fatti, più che parole. L’indagine osservante e l’analisi del comportamento di fruizione del visitatore museale”. Mercati e competitività 4, 103-125.
Sosnovsky, S. & D. Dicheva (2010). “Ontological technologies for user modelling”. Int. J. Metadata, Semantics and Ontologies 1, 32-71.
vom Lehn, D. (2006). “Embodying experience. A video-based examination of visitors’ conduct and interaction in museums”. European Journal of Marketing 11/12, 1340-1359.
Xanthoudaki, M. (2000). “La visita guidata nei musei: da monologo a metodologia di apprendimento”. Nuova Museologia 2, 10-13.
L. Solima: Introduction, § 2, Conclusions; C. Birtolo: § 1; S. Acanfora & C. Birtolo: § 4; M. Minei: § 3; S. Acanfora § 5. Proofreading and translation into English by Simona Losito (Poste Italiane).
L. Solima, C. Birtolo, S. Acanfora and M. Minei, User Modeling, Personalization and Adaptivity: Potential Keys to Success for FIBAC Project. In Museums and the Web 2013, N. Proctor & R. Cherry (eds). Silver Spring, MD: Museums and the Web. Published May 30, 2014. Consulted .