A Multidimensional Approach for Modelling and Supporting Adaptive Hypermedia Systems

Transcription

1 A Multidimensional Approach for Modelling and Supporting Adaptive Hypermedia Systems Mario Cannataro, Alfredo Cuzzocrea, Andrea Pugliese ISI-CNR, Via P. Bucci, 41/c Rende, Italy {cannataro, Abstract. In this paper we present an XML-based modular architecture for the modelling and the run-time support of web-based Adaptive Hypermedia Systems. The proposed model and the supporting architecture allow the hypermedia adaptation along three different adaptivity dimensions : user`s behaviour (preferences and browsing activity); technology (network, user`s terminal, etc.); external environment (time, location, language, socio-political issues, etc.). A view over the Application Domain corresponds to each possible position of the user in the adaptation space. For the description of the logical structure of the hypermedia we propose a graph-based layered model: XMLbased models are used to describe metadata about basic data fragments and neutral pages to be adapted (i.e. presented) with respect to the user position. An authoring and simulation tool which allows the design and simulation of the Application Domain is also presented. 1 Introduction The linking mechanism of hypermedia offers users a large amount of navigational freedom so that it becomes necessary to offer support during navigation. The personalisation of web presentations and contents, i.e. their adaptation to user requirements and goals, is becoming a major requirement of modern web-based systems. Different application fields where contents personalization can be useful are on-line advertising, direct web-marketing, electronic commerce, on-line learning and teaching, etc. Main aspects that should be taken into account are: the different classes of users that will use the system; they are becoming more and more heterogeneous due to different interests and goals, world-wide deployment of services, social conditions, etc.; the kind of user terminals and network. User terminals can differ not only at the software level (browsing and elaboration capabilities), but also in terms of ergonomic interfaces (scroll buttons, voice commands, etc.); networks can differ e.g. regarding communication bandwidth (narrow, medium, broad) and dynamic properties (per-user bandwidth, latency, error rate, etc); the spatial and temporal conditions of the particular user, geo-political issues, etc.

2 To face some of these problems, in the last years the concepts of user modelling and adaptive graphical user interface have come together in the Adaptive Hypermedia (AH) research theme [2, 5]; some recent Adaptive Hypermedia Systems (AHS) are outlined in [5, 7]. Basic components of AHS are: the Application Domain Model, used to describe the hypermedia basic contents and their organisation to depict more abstract concepts. The most promising approach in modelling the Application domain is data-centric, and many researches employ well known database modelling techniques [1]; the User Model (profile), which attempts to describe the user s characteristics and preferences and his/her expectations in the browsing of hypermedia. In the following, we will use the term User Model to indicate the behavioural model only; the overall model, comprising non-behavioural characteristics, will be captured by means of the presented multidimensional approach; the techniques to adapt presentations with respect to the user s behaviour and to the content provider s goals. Such techniques can be generally distinguished into adaptive presentation, i.e. a manipulation of information fragments, and adaptive navigation support, i.e. a manipulation of the links presented to the user. More recently, the capability to deliver a given content to different kind of terminals, using e.g. wired or wireless networks, (i.e. the support of multi-channel accessible web systems), is becoming an important requirement of AHS. In this paper we present an XML-based modular architecture for the modelling and the run-time support of web-based Adaptive Hypermedia Systems, supporting in particular multi-channel access. The proposed model, named XML Adaptive Hypermedia Model (XAHM), allows the hypermedia adaptation along three different adaptivity dimensions : user s behaviour, technology and external environment. The construction of a User Model, i.e. the assignment of the user s position over the user s behaviour dimension, is carried out using a probabilistic interpretation of hypermedia structure. Furthermore, XAHM makes use of XML technologies to (i) describe basic multimedia data, (ii) describe neutral pages to be adapted, (iii) offer an advanced multi-channel access support at run-time. The rest of the paper is organised as follows. Section 2 presents XAHM. Section 3 describes the system architecture. Section 4 discusses related works. Finally, Section 5 contains conclusions and outlines future work. 2 Application Domain Modelling This Section presents our approach to the modelling of Adaptive Hypermedia. After a description of the proposed adaptation scheme, we show a graph-based layered model for the Application Domain and the XML-based model used to describe metadata about basic information fragments.

3 2.1 Adaptation Space The goal of AHSs is to adapt contents and presentations to satisfy the user s goals and/or requirements. In XAHM, the Application Domain is modelled along three different orthogonal adaptivity dimensions (Fig. 1): User s behaviour (preferences and browsing activity); External environment (location, language, socio-political issues, external web sites content, etc.); Technology (network bandwidth, Quality of Service, user s terminal, etc). E External Environment T U AH(U, E, T) User s Behaviour Technology Fig. 1. Adaptivity dimensions and adaptation space Each position of the user in the adaptation space is a tuple [U, E, T] indicating that the user belongs to the group identified by the U value, is located into the environment identified by the E value, and is using the technology identified by the T value. The Adaptive Hypermedia is a sub-space AH(U, E, T) of the overall adaptation space, specified by the author by specialising the description of the hypermedia for some particular positions of the user. A personalised view over the Application Domain corresponds to each of these possible positions, and is obtained instantiating the neutral XML descriptions of pages with respect to the actual position of the user. The AHS monitors the different possible sources that can affect the position of the user in the adaptation space, collecting a set of values, called User, Technological and Environment Variables. The User and Environment variable mainly drive the generation of page content, while the technology one mainly drives the adaptation of page layout. The current position of the user along a generic dimension is obtained using a mapping function from the corresponding variables to a mono-dimensional value. For example, if we consider n Technological Variables, each of which has an associated domain V i, i=1,..., n, then there will exist a simple mapping function f: V 1V 2...V n T, and T will have V 1 * V 2 *...* V n possible values. Without loss of generality, the domain T can be N. A classification algorithm, whose description can be found in [7], carries out the mapping from the User Variables to the User Dimension (i.e. user profile).

4 2.2 A Layered Model for the Logical Structure of the Hypermedia The proposed Application Domain Model uses a layered data model; it extends the Adaptive Data Model described in [9] and comprises the following levels: 0. Information Fragments (IF) or atomic concepts, like texts, sounds, images, videos, etc. at the lowest level. Data can be structured, semi-structured or unstructured and can be provided by different sources (e.g. external or local databases, XML and HTML documents, texts, files and so on). They are described by metadata represented by XML documents; the structure of such documents will be shown in Section Presentation Descriptions (PD), which describe Page Concepts constituted of XML documents. A Presentation Description is a neutral page whose elements are be associated to a portion of the adaptation space so it is transformed and delivered to the user considering his/her position in that space. Each link in a page is annotated by weight, that represents its importance with respect to each other links, and that in our model is a probability (see Section 2.2.1). Final pages composed of actual fragments, called Presentation Units (PU) or pages, are dynamically generated at run time in a target language (XML, HTML, WML, etc.) and delivered. 2. Elementary Abstract Concepts (EAC) representing larger units of information. One or more Presentation Descriptions organised in a weighted digraph [11], compose Elementary Abstract Concepts. Arcs represent relationships between elementary concepts or navigation requirements. 3. Application Domain. Finally, an Application Domain (i.e. AH) is composed by a set of Elementary Abstract Concepts organised in a digraph. Arcs represent relationships between EACs. It should be noted that although the structures of the hypermedia data at the levels 2 and 3 of our data model are essentially the same, the semantic interpretation of the inter-eac browsing is different with respect to the intra-eac browsing. The modelling of an adaptive hypermedia comprises the following phases: 1. definition of M stereotype user profiles, representing users groups; 2. definition of the overall Application Domain, as a directed graph of EACs, differentiating its link structure with respect to user profiles; 3. definition of the structure of each EAC, differentiating its link structure with respect to user profiles; 4. construction of the Presentation Descriptions, differentiated with respect to all the adaptivity dimensions. The Application Domain and the EACs are differentiated with respect to the user s profile only (User s Behaviour dimension) because they need to be directly described by the author and it could not be suitable to build and maintain a large variety of weighted digraphs. However, future extensions of the system could support the personalization of the link structure of the hypermedia with respect to the Technology dimension, so allowing lighter (i.e. with shorter paths) versions of the hypermedia, to be browsed in a more agile way.

5 2.2.1 Probabilistic Graph-Based Scheme and User Classification. The proposed graph-based model supports a probabilistic interpretation of the arcs weight the mapping from the collected variables about user s behaviour (User Variables) to the corresponding dimension of the adaptation space (which, as said before, is organised identifying groups of users) is based on the evaluation of values related to such probabilistic interpretation. The mapping is carried out also considering some other values, related to intrinsic properties of the hypermedia structure. An AD with M different user profiles can be seen as a flat set of XML documents where the generic document contains, for each profile, a set of annotated outgoing links. The AD can be mapped in a weighted digraph G where each node corresponds to a XML document and each directed arc to an outgoing link; in turn, the weighted digraph G can be referred to as the set of the weighted digraphs G k, k=1,,m, obtained extracting from G the nodes and arcs corresponding to each profile. Each G k is named Logical Navigation Graph [6]. The proposed probabilistic approach assumes that the weight of the arc (i, j) in G k is the conditional probability P(j k,i), namely the probability that a user belonging to the group k follows the link to the j node having already reached the i node. The User Variables contain the recently followed path R = {R 1,, R r-1, R r} composed by the last visited nodes (where R r-1 is the current node and R r is the next node) and the time spent on recent nodes, t(r 1),..., t(r r-1), so identifying a sort of sliding temporal window of length r. On this basis, the system evaluates, for each profile k, the following values: the probability of having followed the R path through arcs belonging to the profile k; the reachability of the next node R r starting from the first node R 1, through arcs belonging to the profile k (i.e. the maximum probability path in G k); the distribution with respect to the profile k of the visited nodes from R 1 to R r-1, weighted with the time spent on each of them. For example, let { n 1, n 2, n 3 } be the recently visited nodes and { t 1, t 2, t 3 } the time units spent on each of them: if node n 1 belongs to profiles k 1 and k 2, node n 2 belongs to k 2 and k 3 and node n 3 belongs to k 1 and k 4, the distribution is evaluated as [ (k 1, t 1+t 3), (k 2, t 1+t 2), (k 3, t 2), (k 4, t 3) ]. The visiting times can be measured using a software running on the user terminal. The intrinsic properties of the hypermedia are expressed, for each profile k, by: the average value of the probability of the minimum paths in G k; high values of this term indicate the existence of highly natural paths in the hypermedia; the average value of the length of the minimum paths in G k; high values of this term mean longer natural paths in the hypermedia, which could be an advantage in the overall personalization process; the number of nodes belonging to profile k. The algorithm for the evaluation of the user s profile computes a belonging probability density function considering initial user choices (e.g. expressed in a questionnaire) and the aforementioned values; a detailed description of the algorithm is beyond the scope of this paper and can be found in [7].

6 2.3 XML Metadata about Basic Information Fragments In XAHM, each data source is wrapped by an XML meta-description. The use of metadata is a key aspect for the support of multidimensional adaptation; for example, an image could be represented using different levels of detail, formats or points of view (shots), whereas a text could be organised as a hierarchy of fragments, represented using different languages. In the construction of pages, the author refers only to such metadata, avoiding too low-level access functions to real data fragments. A number of Document Type Definitions [15] for the XML meta-descriptions have been designed; as an example, in Figure 2a is shown the meta-description of a relational table (or view) stored in a relational database. <!ELEMENT table (column+)> <video alias= movie-trailer > <!ATTLIST table <video alias= movie-trailer > alias CDATA #REQUIRED <instance alias= 1st IP-address CDATA #REQUIRED description= first version name CDATA #REQUIRED mime-type= video/mpeg database-name CDATA #REQUIRED size-x= 320 > size-y= 256 bit-rate= 1.5 compression-type= mpeg2 <!ELEMENT column EMPTY> resolution= 150 <!ATTLIST column duration= 3.5 > name CDATA #REQUIRED <query IP-address=... type CDATA #REQUIRED port=... number CDATA #IMPLIED database-name=... > > select video from trailers where key= 15 </query> </instance> </video> (a) (b) Fig. 2. XML metadescriptions The meta-description of a relational database table comprises attributes concerning the location of the DBMS (with an IP address and a database name), the name of the table itself and an alias used to reference it. Sub-elements are the columns of the table, identified by a name (optionally a number) and a type, which can be a predefined SQL type or a complex type described by another meta-description. As an example, if a column of an object-relational table contains objects of a complex type representing a MPEG video (see e.g. Oracle s ORDVideo type [13]), its specific meta-description can have the valid structure of Figure 2b. 3 System Architecture The overall architecture of the system comprises the run-time system and a set of the authoring tools; they have been designed and are now under implementation.

7 Repository Level Data Sources Level 3.1 The Run-Time System The run-time system supporting the XAHM model has a three-tier structure (Fig. 3), comprising the Presentation, the Application and the Data Layers. The Presentation Layer receives final pages to be presented and eventually scripts or applets to be executed (to detect e.g. local time, location, available bandwidth or the time spent on pages). The terminal User Agent (browser) usually communicates the kind of user s terminal and its software details (OS, browser, etc.). Data Repository XML Repository Object Repository DTD Repository User Modelling Component Data Access Module Wrapper Wrapper Adaptive Hypermedia Application Server User s Terminal Web Server Terminal Software S 1 S n Data Layer Application Layer Presentation Layer Fig. 3. Run-time system architecture At the Application Layer there are two main modules: the Adaptive Hypermedia Application Server (AHAS) and the User Modelling Component (UMC) [4]; they run together with a Web Server. The UMC maintains the most recent actions of the user and executes the algorithm for the evaluation of the user s profile. The AHAS performs the Adaptation Process, executing the following steps: 1. extraction of the Presentation Description to be adapted; 2. receipt from the UMC of the user position into the adaptation space; 3. composition of the final page, composing data fragments extracted from the data sources on the basis of the user position; 4. transformation of the final page into the target (terminal) language using a terminal-dependent XSL style sheet. Finally, the Data Layer stores persistent data and offers efficient access primitives. Each of the data sources S i is accessed by a Wrapper software component that also generates, in a semi-automatic way, the XML metadata describing the data fragments stored in S i. The Data Access Module implements an abstract interface for accessing the Data Sources and the Data Repository.

8 3.2 The Authoring Tools The Authoring Tools have been designed to efficiently support the design, validation and testing of the hypermedia before its deployment in a running system. A Author Module allows to define in a graphic way the logical structure of the hypermedia, to browse the data sources, to access the associated metadata, to compose fragments to build the Presentation Descriptions and to validate them. A Hypermedia Simulator allows simulating the behaviour of the AHS (i.e. the produced probabilistic density functions and the user profile assignments) on the basis of different kinds of users supposed to interact with the system (i.e. moving through its probabilistic structure). In particular, the author can: 1. analyse the intrinsic properties of the hypermedia calculated from its structure; 2. define a set of User Classes that describe the behaviour of typical users; many different User Masks can be assigned to each class, so the behaviour of each user can change during the same interaction with the system; 3. analyse the response of the AHS with respect to the User Classes. 4 Related Work Some interesting Data Models have been used recently to describe and develop (Adaptive) Hypermedia Systems. In the Hypermedia Design Model (HDM) [12] the main design phases are: in-the-large, where the overall hypermedia architecture and behaviour are designed; in-the-small, where the organisation of the overall hypermedia is detailed; structure, where the contents organisation is constructed; dynamics, where the dynamics of the hypermedia are designed. The Object-Oriented Hypermedia Design Model (OOHDM) [14] represents the object-oriented evolution of HDM; it introduces the classical aspects of this paradigm and uses the abstraction and composition mechanisms. The main design phases are: conceptual design, where a conceptual model of the Application Domain is built; navigational design, where the navigational structure is designed; abstract interface design, where an abstract model for the object definition is constructed in terms of interface classes; implementation, where the previously-designed software objects are created and the architecture of the hypermedia (client-server, three-tier, etc.) is defined. WebML [8] defines a notation for the specification and the design of a hypermedia on the basis of some constructs, formally defined in XML and intended for describing the application contexts at a high level of abstraction, disregarding implementation details. The design phase in WebML is carried out composing a set of orthogonal abstraction models: structural model, for the conceptual organisation of data;

9 derivation model, for the extension of the structural model by deriving new information from the existing; hypertext model, for the description of some different hypertext views of the hypermedia; composition model, for the organisation of each of the hypertexts in pages and each page in information units; navigational model, for the description of how relationships between data can be transformed in navigational options; presentation model, for the definition of the layout of pages, in a terminalindependent way; user model, for the description of users and their organisation in groups; personalization model, for the definition of Event-Condition-Action rules for the hypermedia personalization. In XAHM, the design of an adaptive hypermedia comprises the following phases: logic design, where the logic structure of the hypermedia is designed at different abstraction levels; data model design, where basic data fragments and the Presentation Descriptions are parameterised on the basis of the three adaptivity dimensions, using XML metadata; presentation design, where the XSL models containing the layouts for different user terminals and software browsers are defined. A qualitative comparison between the four models is shown in Table 1. The considered characteristics are expressiveness, modularity, extensibility (referred to the model itself), and adaptive hypermedia support. Table 1. Hypermedia modelling language comparison Expressiveness Modularity Extensibility Adaptivity support HDM low low low low OOHDM very good medium medium medium WebML good very good very good good XAHM medium good good very good It is possible to note some similarities between XAHM and WebML. Both use XML as basic language; the WebML structural model corresponds to the fragments metadata definition in XAHM; the WebML hypertext and presentation models correspond to the XAHM presentation design; the WebML composition, user and personalization models correspond to the XAHM logic design. Currently, XAHM does not support data derivation as in WebML, but it is designed to take into account the behaviour of many users to dynamically redesign the overall hypermedia structure (e.g. adding or pruning profiles). Similarly to HDM, XAHM employs a hierarchic data model, but it does not support the object-oriented paradigm as in OOHDM. This explains its extension and derivation lacks. Although WebML offers many useful characteristics to describe AHs, we chose to develop XAHM to model with a more systematic approach the adaptation process.

10 5 Conclusions and Future Works In this paper we presented a XML-based modular architecture for the modelling and the run-time support of web-based Adaptive Hypermedia Systems also supporting multi-channel access. The Adaptive Hypermedia is described as a three-dimensional space along the user s behaviour, technology, and external environment adaptivity dimensions. So, the adaptation process is implemented finding the proper position of the user in that space, loading and adapting the corresponding XML page, applying to it the constraints bound to that point. We are currently implementing a prototype of the system, that uses Java and XML to be completely cross-platform and extensible, and is based on the Apache [3] server suite, comprising the Apache HTTP Server, the Cocoon Application Server, the Tomcat Java Servlet engine, the Xerces XML parser and the Xalan XSLT Processor. The data centric approach in the data fragments representation and the use of orthogonal concepts in the modelling phase is currently used to implement terminaland network-independent hypermedia systems. In future works we intend to use such concepts as a basis to implement environment-aware web-based systems, such as the location-dependent services of mobile systems. References 1.Abiteboul, S., Amann, B., Cluet, S., Eyal, A., Mignet, L., Milo, T., Active views for electronic commerce, in Proceedings of the 25 th VLDB Conference, Adaptive Hypertext and Hypermedia Group, 3.Apache Software Foundation, The Apache XML Project, 4.Ardissono, L., et al., A configurable system for the construction of adaptive virtual stores, World Wide Web Journal, Baltzer Science Publisher, Brusilovsky, P., Methods and techniques of adaptive hypermedia, in User Modeling and User Adapted Interaction, v.6, n.2-3, Cannataro, M., Pugliese, A., An XML-based architecture for adaptive web hypermedia systems using a probabilistic user model, in Proceedings of the IEEE IDEAS Conference. IEEE Computer Society Press, Cannataro, M., Cuzzocrea, A., Pugliese, A., A probabilistic approach to model adaptive hypermedia systems, in Proceedings of the Int. Workshop for Web Dynamics, Ceri, S., Fraternali, P., Bongio, A., Web Modelling Language (WebML): a modelling language for designing web sites, WWW9 Conference, P. De Bra, P., Brusilovsky, P., Houben, G.J., Adaptive Hypermedia: from systems to framework, ACM Comp. Surveys, Symposium Issue on Hypertext and Hypermedia, De Bra, P., Design issues in adaptive web-site development, in Proceedings of the second workshop on adaptive systems and user modeling on the WWW, Diestel, R., Graph Theory, Springer-Verlag, New York, Garzotto, F., Paolini, D., Schwabe, D., HDM A model-based approach to hypermedia application design, ACM Transactions on Information Systems, Jan., Oracle Corporation, Oracle intermedia Audio, Image, and Video Reference, Documentation Library, Schwabe, D., Rossi, G., The Object-Oriented Hypermedia Design Model, CACM 38(8): 45-46, World Wide Web Consortium, Extensible Markup Language, Recommendation, 2000.