Personalising Electronic Books
AbstractThe paper addresses how hyperdocuments, accessible via electronic books (e-books) which are read using the World Wide Web, can be endowed with features that personalise the interaction process that takes place between the reader and the e-book. A novel, abstract approach to modelling the personalisation of hyperdocuments is introduced. This approach aims to make available features that allow readers to interact with these documents in a manner much closer to that with paper-based documents. The research is based on a formal characterisation of personalisable hyperlink-based interaction. This characterisation is unique in formally modelling a rich set of user-initiated personalisation actions that allow users to come closer to satisfying their specific, often dynamic, information retrieval goals.
The first speculative discussion of the concepts underlying the notion of using technology to display books to readers can be found in Bush's (1945) proposal for a system called memex ("Memory Expander''). Bush described the memex as "a sort of mechanised private file and library". After describing the mechanics of memex, Bush stated:
All this is conventional, except for the projection forward of present-day mechanisms and gadgetry. It affords an immediate step, however, to associative indexing, the basic idea of which is a provision whereby any item may be caused at will to select immediately and automatically another. This is the essential feature of memex. The process of tying two items together is the important thing.
This quotation highlights the importance Bush placed on the user in the process of constructing associative relationships between units of information. An important conclusion that can be drawn is that the memex system was envisaged to be personalisable, in that a user could make an associative relationship or link between two units of information. It can be observed that even today the vast majority of hyperdocuments, such as those which are read as e-books using the Web, provide each user with the same associative relationships (retrieval possibilities) but disregard user-specific information goals and histories. Furthermore, all decisions as to the association of information units are decided by the designer of the system, with users being given no opportunity to interfere with these decisions.
This paper defines the content of an e-book to be a network of digital information units which may comprise text, graphics, video, animation and/or sound. When these information units are rendered they provide the user with optional links to other information units. Such links can provide context-based, non-linear navigation between information units. In this paper a rendering unit in an e-book is viewed as a hyperpage, and a collection of hyperpages as a hyperdocument read using a Web browser or handheld device. The core of this definition is content, a digital object containing an electronic representation of a book whose pages are interconnected by hyperlinks. More generally, an e-book is viewed as an electronic version of a paperback or cloth-bound book. In its digital form the content of an e-book may be dynamically generated on the fly. It may be accessible world wide, via the Internet and read using an e-book reader such as a Web browser. An e-book reader is defined to be an electronic device capable of displaying e-books. E-book software operates on an e-book reader providing book display functions. Personal computers, PDAs, pocket PCs and dedicated readers such as Rocket e-book (http://www.ebook-gemstar.com) and Cytale Cybook (http://18.104.22.168/ebookweb/stories/storyReader$84) are e-book readers, while Web browsers such as Internet Explorer (Grauer 1998), MS Reader for the pocket PC (http://www.microsoft.com/reader/default.asp) and Palm Reader Pro (http://www.peanutpress.com) are examples of e-book reader software.
Although there have been many impressive proposals for the development of e-book technologies (Harrison 2000, Press 2000) and much research has been conducted into their deployment (Landoni et al. 2000a) and design (Wilson 2002, Landoni et al. 2000b, Landoni et al. 2001, Marshall et al. 2001), a significant limitation of many e-books read using the Web is the inability to address users' information needs on an individualised basis (Brusilovsky 1996, Brusilovsky et al. 1998, De Bra et al. 2002b). Most e-books provide distinct users with the same information retrieval possibles and present that information in the same manner. If one considers the question of how users interact with a book over extended periods of knowledge acquisition, and the actions they are able to take that would be difficult or impossible if the book were a hyperdocument, it can be argued that users of hyperdocuments are limited in the functionality available to them (Brusilovsky et al. 1998 ). For example, when a user interacts with a paper-based book, they are given the opportunity not only to read its content in a linear manner, but also to annotate it, insert and delete content, bookmark and highlight particular sections and to make, in context, cross references to parts of the book and to other books. Such functionality is rarely found in e-books read using the Web.
Personalisable, adaptive hyperlink-based systems (PA-HLBSs), more often referred to as adaptive hypermedia (AH) (Brusilovsky et al. 1998, Brusilovsky 2001) constitute an area of research that aims to enhance the functionality of hypermedia systems, such as hyperlink-based e-books (Landoni et al. 2000b), by making the user interaction process personalisable. The approach taken is to endow systems with personalisation features which may be initiated by the users or by the system itself. AH is assumed to be useful in areas, such as learning, where users have different information-seeking goals, histories and preferences. AH aims to use knowledge provided by (or captured about) specific users to tailor the information and the links presented to each specific user. By building a knowledge of users, AH can support users in navigation by limiting the options for traversal to information units, suggesting relevant links to follow and providing additional information on links and information units. This paper proposes a system-independent approach to personalising hyperdocuments as a value-adding strategy for e-book technology. This work is based on research (Ohene-Djan 2000) that has contributed a formal model of personalisable, adaptive hyperlink-based interaction. The main contributions of that model were a formal definition of core hyperlink functionality and a formal definition of hyperlink-based personalisation through extensions to this core. Such extensions allow for the specification of hyperdocuments that facilitate user-initiated personalisation of their content.
The remainder of this paper is structured as follows. Section 2 provides a brief background to our research. To motivate the contributions of the paper, section 3 contrasts the actions that users of paper-based books can perform with those that users of hyperdocuments cannot, and vice-versa. Section 4 introduces a framework for the personalisation of hyperdocuments. Section 5 uses an example to show how the framework proposed allows the desiderata identified to be met. Section 6 contrasts our results with those of other researchers.
This section briefly outlines some issues, concepts and principles of our research, and opens by describing why hyperlink-based interaction, as a model for information retrieval, is a natural candidate for modelling and deploying e-books. Several fundamental issues relating to the design of personalisable hyperdocuments are then outlined, together with some principles for personalisation, which have emerged from preliminary research.
To understand PA-HLBSs research and its potential importance to the design of hyperdocuments accessible via e-book technologies, it is useful to consider first how users seem to read and write documents. Human intelligence depends on models with which people communicate. Broadly, it is believed that one associates pieces of information with other information to create complex knowledge structures or networks of information. Such a network forms a mental model which may be viewed as a semantic network comprised of units of information represented as nodes, and associations between units represented as links (Rada and Wang 1995). Communicators must be able to persuade the audience that acceptance of their communicated mental model is in the best interests of the audience.
During this process, a typical medium (i.e. paper, video) can only capture a sequential presentation. An author must transform their non-linear communication (mental model) into a linear form. Authors have to linearise their mental model before it can be communicated. As the linear representation is not natural, authors may provide additional information (i.e. tables of contents and indexes). Readers, in turn, transform the linear message into their own non-linear models in their minds. Generally this process involves the reader breaking up the information into smaller parts and then rearranging these parts based on their information goals. This process is shown in Figure 1.
Smith et al. (1987) suggested a writing model which consists of three cognitive phases:
- an exploring phase during which knowledge is formulated, initial drafts are made, and ideas are grouped into different perspectives;
- an organising phase during which initial drafts are organised into sequence and an outline is made;
- an encoding phase during which the document is written.
Although these phases may be undertaken in sequence, generally authors like to move freely from one phase to another. Delisle and Schwartz (1989) describe the following activities:
- Recording relevant ideas, where initial drafts of ideas are made;
- Exploring relationships among ideas, which is the associative linking process;
- Structuring ideas, which is the process of organising drafts and creating an outline;
- Preparing the document, during which the document is written, linearised, formatted and printed.
Research into PA-HLBSs is motivated by their potential to enable authors partially to mimic the processes of writing described above. The logical model of HLBSs is largely that of a semantic network based on the associative linking of units of information attached to the nodes of the network. The analogy of a semantic network to a hypernetwork is straightforward and has long been recognised (Conklin 1987, Rada, 1990). PA-HLBSs allow authors and users directly to create networks that communicate their mental models through the use of computer-supported links and annotations. Authors and users may not only recreate their mental models as non-linear information structures (hypernetworks) but also tailor these structures, over time, to satisfy their information retrieval needs on an individualised basis, as shown in Figure 2.
It is now generally accepted (Brusilovsky 1996) that users of HLBSs may differ in their information goals insofar as they may have preferences as to what information is provided and which links are used to navigate the information space. Users may also differ in their histories insofar as they are likely to have different knowledge of the information contained within the HLBS, of the information space and how it may be navigated.
Most of the interaction a user might experience with a hyperdocument is determined by the design decisions that shaped the hyperdocument in terms of its content, rendering and navigation possibilities (i.e. its links). These decisions are unilateral and irreversible (i.e. cannot be overridden by users as they interact with the hyperdocument), so it can be said that the designers own the hyperdocument.
This paper proposes one approach to overcoming this impediment by extending HLBSs with formally defined personalisation actions that effect a transfer of ownership from former designers of the hyperdocument to each of its users, thereby enabling the latter to redesign the hyperdocument according to their specific information goals and histories.
The challenge is, therefore, how to model, at a suitable level of abstraction, the space of possibilities for personalisation actions that could be made available to beneficiaries of PA-HLBSs. Such a model should characterise the notion of "transfer of ownership" and should avoid being technology-driven. Furthermore, the choice of personalisation actions should fall out from this abstract model of interaction and should ultimately be subject to empirical tests for effectiveness gains, although this last desideratum is not tackled in this paper.
In answering this challenge, the following derived problems present themselves:
- How to model interaction with a hyperdocument?
- How to transfer ownership of the process of interaction with a hyperdocument from the designers to its users?
- How to make design decisions more explicit?
- How to distinguish, model and implement personalisable hyperdocuments?
- How to distinguish, model and implement personalisation and adaptivity?
A further goal is to incorporate the following principles for personalisation, which emerged from preliminary research:
- Personalisation should, to some degree, represent the measured transfer of ownership of the process of interaction with a hyperdocument from designer to user;
- All hyperpage design decisions should, in principle, be able to be the subject of personalisation requests;
- A model of hyperlink-based personalisation should aim to accommodate all recognised personalisation actions (see Brusilovsky (1996) for a comprehensive review);
- Personalisation actions should be clearly defined, explicit and capable of being formally defined;
- All personalisation actions should be consistent, repeatable and revisable;
- Any choice of personalisation actions should ultimately be subject to empirical tests for effectiveness gains.
This section discusses the motivation for ownership transfer via personalisation as a value-adding strategy. It does so by contrasting the kind of personalisation actions that owners of traditional learning materials (such as printed books) and users of HLBSs can carry out.
Assume users are computer students. As part of their degree, students are required to take a course on e-commerce, for which the textbook is Electronic Commerce by Schneider and Perry (2000). A small section of the book is depicted in Figure 6.
Consider how students interact with a textbook over extended periods of learning. Furthermore, consider interactions which would be difficult or impossible if the book were a hyperdocument. Some of the personalisation actions that fall into this category include:
- Selective reading. To open and read some page(s) only (possibly consulting the table of contents or index) and perhaps only partially. In current hyperdocuments, options for traversal are defined by the links the designer of the hyperdocument has embedded in it. Users are not provided with facilities to make links between pages and therefore define their own paths through the material. Also, when a user does want to traverse a link defined by the designer there is no metadata about the target, thereby making it impossible not to commit to the traversal. Since traversing links can incur high computational costs, frustration sometimes ensues.
- Annotation. To add handwritten material to the original printed text, which for the student enriches the meaning of the text. Examples include underlining or highlighting a passage, writing comments in the margin, cross-referring to some other page of the book or some other information resource (e.g. a Web site, another book, etc.). At present such features are rare.
- Content insertion and deletion. To add content (e.g. copied from another book, or a Web site) at specific points and to delete (or cross out, not necessarily physically) parts of the text. Features to enable the users of a hyperdocument to tailor its content have not been implemented widely. In particular, users are not often given the ability to remove parts of a hyperdocument which they deem inappropriate for study at a point in time.
- Bookmarking. To leave markers at particular pages (e.g. to indicate that these pages are more important in some respect than others). Note that marked pages are easily identifiable for selective reading. Although Web browsers allow users of hyperlink-based learning systems to bookmark pages, such bookmarks are often limited to recording only the Internet address (location) of a page. Users are not normally given the opportunity of indicating the importance of some bookmarked page over another or why a particular page is bookmarked at all.
Each of the above actions are admissible on paper-based materials that the student owns, but are markedly difficult on hyperdocuments because the student has no means of claiming ownership of the material nor of personalising it (even if ownership could be transferred from the designer to the student, because the student may not have the designer knowledge required for personalisation to be carried out).
Now consider how students would interact with a textbook implemented as a hyperdocument. Furthermore, consider interactions which would be difficult or impossible if the book were not a hyperdocument. Some actions that fall into this category include:
- Context-based nonlinear navigation. To benefit from the availability of contextual links at points deemed appropriate by the designer of the hyperdocument and traverse the material nonlinearly.
- Dynamic information management. To make use of computational resources to react and respond to requests dynamically (e.g. to benefit from a web of information comprising many, possibly remote, independently generated and maintained interrelated hyperdocuments).
Each of the above actions are admissible on electronic materials that the designer prepares, but markedly difficult on paper-based documents even if the student owns them.
In summary, although users of hyperdocuments benefit from the concept of linking and from the fact that electronic media typically exist in a computing environment, there are many useful actions that the owner of paper-based documents can take that are not open to users of hyperdocuments.
This section describes a framework for the personalisation of hyperdocuments that aims to make available to users of, e.g. electronic learning materials, features that allow these users to interact with the hyperdocuments in a manner much closer to that of owners of paper-based ones, while still benefitting from the ability to traverse contextual links and to exploit the computational environment in which the material is embedded.
A general, open architecture for HLBSs, of the kind shown in Figure 3 as a simplified data flow diagram, is assumed. Here we propose one view of the interior of the shaded oval in Figure 3. It is assumed that hyperlink functionality is provided as a client technology dependent on loose couplings to (at least) a user-interface server (UIS) and a database server (DBS). The classical example of a UIS is a Web browser (e.g. Netscape Communicator). Among other functions, browsers broker requests and have rendering capabilities. Browsers can render formal texts (e.g. rendering expressions authored in HTML or XML). Examples of DBSs are database management systems (DBMSs) which support client server architectures (e.g. Oracle, or MySQL).
Broadly, the dynamics associated with Figure 3 are as follows. The UISs capture requests for desired hyperpages. The UISs channel requests for hyperpages into the HLBS proper. If a request is for a hyperpage which resides in a remote HLBS, then the core of hyperlink functionality interacts with it to obtain the requested hyperpage in the form of a rendering expression that the core can pass back for the UISs to render. If the request is for a local hyperpage (e.g. one which is known to the core) then the latter responds by returning a rendering expression to the UISs, possibly after querying one or more DBSs to fetch some or all of the content specified for the requested hyperpage.
Implicit in Figure 3 is the assumption that personalisation actions in HLBSs should not, and need not, be compounded with personalisation actions that might be provided by user-interface and database components in HLBS architectures. The shaded oval in Figure 3 is responsible for what users experience as hyperlink-based information retrieval. Notwithstanding the fact that users may well want to personalise database and user-interface features, clearly it can be argued that whatever is in scope for personalisation actions in HLBSs resides in the shaded oval.
To model adaptive, personalisable hyperlink-based interaction, a framework is proposed in which the shaded oval in Figure 3 is partitioned into regions. Non-adaptive, non-personalisable HLBSs are modelled by a group of functions referred to as the H-region. Personalisable HLBSs require the addition to the H-region of the functions provided by the P-region. Adaptivity, seen as user-specific, system-initiated personalisation, comprises another region of functionality but is not within the scope of this paper (see Ohene-Djan (2000) for details).
The H-region models a core of hyperlink-based functionality. Conceptually, the H-region behaves as a composer of hyperdocuments from specifications, i.e. what the designer of a hyperpage designs is not a hyperpage, but rather a specification of how to build the hyperpage upon request. Hyperpages are modelled as formal specifications and a formal language has been defined for this purpose (Ohene-Djan 2000). The functionality of the H-region is shown in Figure 4.
Within the H-region, users can only request for hyperpages to be rendered. The decisions that the designers of a hyperdocument have made with respect to content, navigation and rendering cannot therefore be overridden. Upon a request arriving from a UIS, a composition function parses the hyperpage specification into a series of actions that, when executed, convert the specification into renderable text that is sent to the UIS as the response to the original request.
The semantics of a hyperpage specification (of which an example is shown in Figure 7) have been formalised as a program which, when interpreted, typically fetches content from a DBS, composes the content into a renderable text (making use of template variables as a binding mechanism) and finally responds to the original request with renderable text. Figure 8 indicates how the hyperpage specified in Figure 7 might be rendered.
The P-region comprises a group of functions that are non-disruptively added to the H-region in order to model personalisable hyperlink-based interaction. Personalisation is viewed as the process of handing over to the user the ability to annotate specifications or rewrite them or both, thereby allowing the user to override, in principle, each and every designer specification. Within the P-region users can not only request a hyperpage, but also annotate or rewrite it, thereby creating their own version of it. The decisions that the designers of that hyperpage have made with regard to content, navigation and rendering of the hyperpage can therefore be overridden by users and this kind of event characterises ownership transfer. The functionality of the P-region is shown in Figure 5.
When personalisation functionality is layered over the core, a designer can annotate a hyperpage in preparation for differences in users' goals and histories. A user can personalise not only such annotations (of which an example is shown in Figure 9), but the hyperpage specifications as well. Personalisation requests (of which examples are shown in Figure 10) allow users to specify which hyperpages are to be personalised and how they should be transformed.
The kinds of personalisation actions modelled are based on annotating and rewriting the hyperpage specifications. Annotation pairs a hyperpage specification with notes of interest to the user and, by doing so, assumes that versioning takes place. Such notes take one of the following forms. First, a note can assign user-specific values to user-generic attributes of interest (e.g. that the level of difficulty of a given page or component part is high, or that 'planets' is a keyword of relevance to a given page). Second, a note can specify a rewriting action over the renderable text after it has been composed by the H-region, i.e. after content has been fetched and made ready for display (e.g. to map American into British spelling forms). This form of post-composition rewriting can also be conditional on the environment (e.g. replace images with captions if the display unit is text-only).
The existence of annotations on hyperpages allows for:
- personalisation of a specified hyperpage;
- the specification of alternatives to a specified hyperpage;
- the specification of comparable hyperpages to a specified one
- the recording of information about a hyperpage (i.e. the current values of attributes set by previous annotations).
Annotations are not operations on hyperpage specifications, i.e. they do not alter the latter. Rather, they give rise to a user-specific pairing with a hyperpage specification. Rewriting also causes versioning and can be characterised simply as the editing of hyperpage specifications. If a user edits the hyperpage specification as conceived by its designers, ownership is thereby transferred. If, subsequently, the same user edits that hyperpage specification again, adjustment takes place.
A formal language for annotations has been defined, as has a formal language for the personalisation requests used to maintain annotations and hyperpage specifications (Ohene-Djan 2000). Personalisation requests allow users to take the following actions to enhance their interaction process:
- Content tailoring: insertion or deletion of hyperpages and, recursively, their page parts;
- Selective content: hiding from users parts of the content of hyperpages which they wish not to see;
- Content rewriting: rewriting of specified phrases in hyperpages;
- Prerequisite content selection: fetching and displaying of prerequisite content for a particular hyperpage or page part;
- Linking: insertion of hyperlinks between hyperpages and their page parts;
- Link annotation: annotation of links made between hyperpages and their page parts;
- Link hiding and blocking: hiding or blocking of links on a hyperpage.
This section uses an example to show how the desiderata identified in section 3 are met by the functionality arising from the framework described in section 4. Consider Figure 6, where an excerpt from Schneider and Perry (2000) is transcribed.
To specify a hyperpage to correspond to Figure 6, the approach taken is to divide a page into a sequence of segments called chunks. Each chunk comprises a content specification (C-spec) (i.e. either hardwired text or data returned as the result of queries sent to databases, figures, etc.) and a specification of how this content should be rendered (R-spec) for presentation (e.g. via HTML tags). A chunk may contain template variables. Conceptually, a template variable is associated with bits of text or else acts as a placeholder for content which becomes available after evaluation of the query associated with it. The rendering part is specified in a language which the UIS can render, except that this renderable text may be interspersed with template variables. After content is retrieved and the template variables are assigned, textual replacement takes place substituting values (i.e. retrieved content) for references (i.e. occurrences of template variables). The overall result is renderable text (e.g. HTML).
In addition, a chunk may be associated with two sets of identifiers. The first is the set of entry points to the chunk and the second is the set of its exit points. An entry point enables the chunk (as an anchor in the page) to be referenced in a request. An exit point enables a chunk to establish navigable links to the anchor denoted by the exit point. In Web terms, an entry point can be thought of as a URL and an exit point as a hyperlink. A detailed formal descriptions of chunks and the component parts can be found in Ohene-Djan (2000).
Figure 7. Figure 6 as a hyperpage specification
Figure 7 shows one way in which Figure 6 might be written as a hyperpage specification. It is assumed that database queries for retrieving content are written in SQL and that HTML is used as a markup language in the construction of a renderable text. For more detail on how the evaluation of Figure 6 into renderable text takes place, see Ohene-Djan (2000). Figure 8 shows the hyperpage after it has been rendered by a UIS (in this case, an HTML browser).
A hyperpage annotation pairs a hyperpage specification with notes of interest. An annotation is, roughly, a collection of contextualised attribute-value pairs. An example of an annotation for the text specified in Figure 7 is given in Figure 9. The attributes used are description, which lets the student or designer provide a summary or abstract; keyword, which lets the student or designer tag content and thereby clarify the information contained; level, which lets the student or designer attach a measure (e.g. of difficulty) to the text; and see-as-well, which lets the student or designer specify hyperpages that describe comparable information. Again, Ohene-Djan (2000) provides more details.
The annotation in Figure 9 will be associated with the hyperpage specification in Figure 7. Conceptually, this versions the page specifically for the annotator.
Figure 9. Annotating Figure 7
Personalisation requests allow a user (e.g. a student) to generate annotations, to update annotations (e.g. those provided at source by the designer), and to update (by versioning) the hyperpage specifications themselves. A few examples of personalisation requests (in a language, and with a semantics, formally defined in Ohene-Djan (2000)) that a student might issue (to the hyperdocument of which Figure 7 is a page) are shown in Figure 10.
Figure 10. Personalising Figure 7
Example 1 in Figure 10 is a request to tailor the content found in the book. The request applies to all hyperpages (because the selection condition is vacuously true). Its effect is to insert the string "These pages belong to student X" into the rendering specification of the first chunk of each page. Example 2 is a request to add content. The request is applied to all hyperpages that contain the string "electronic mail" in its fifth chunk. The effect on each selected hyperpage is the insertion of a new first chunk. Example 3 in Figure 10 shows a request to rewrite a specified string found in the hyperdocument. The request applies to all hyperpages and has the effect of inserting into the annotation of each hyperpage a request (as denoted by the arrow '->') that all occurrences of the string "http://www.ebook.com" in the composed text be rewritten as "http://www.ebook.co.uk" in the rendering text.
In spite of space limitations the example in this section illustrates how hyperpage specifications that can be annotated and maintained via personalisation requests enable users of hyperdocuments to claim ownership by actions that version pages by means of annotations and rewriting requests.
This section compares and contrasts work that is related to the framework proposed in this paper (the origins of which lie in the formal model of personalisable, adaptive hyperlink-based interaction proposed in Ohene-Djan (2000)).
The HAM (Campbell and Goodman 1987) was the first attempt to define an abstract model within which HLBSs could be expressed. The model describes a transaction-based storage system. The definition of the HAM consists of a description of HAM objects and the operations that can be applied to them. The HAM describes a lower level machine tied closely to the storage (file) system while having a looser connection to applications and user interfaces. Although the approach taken in devising the HAM is related to the work reported in this paper, it differs from it in several important respects.
The HAM views a hypernetwork as a database of objects and the operations described are primarily those to manipulate this database (e.g. create, delete, get, filter). This approach can be contrasted with the model proposed here, which attempts to identify and characterise those operations which are unique to HLBSs as opposed to those operations present in HLBSs qua database applications.
Although the HAM views a core of hyperlink functionality (the HAM itself) as being loosely tied to applications and user interfaces, it views the host file system's structure and contents as being tightly coupled to the HAM. This view can be contrasted with the work reported here, in which both user interface and database functionality are viewed as being loosely coupled to possibly remote servers internal structure of which may be unknown. How such coupling takes place is viewed as a side-effect of the query specification languages and markup languages used to specify the content and presentation of hyperpages.
It can be seen from the specification of the HAM that its authors realised the need for features to allow personalisation of hypernetworks. The filtering mechanism described in the HAM allows subsets of HAM objects to be extracted from hypernetworks (HAM graphs), thereby allowing a user to specify a subset of objects to be presented. However, this approach to personalisation is primarily data-oriented and could be broadly understood as the specification of logical views over an underlying database. In contrast, the work reported here models personalisation of a core of hyperlink functionality, as opposed to specifying particular (i.e. personalised) database queries.
The Dexter model (Halasz and Schwartz 1990, Halasz and Schwartz 1994) has become the foundation for much research into HLBSs modelling and is a reference against which many HLBSs are compared. The model is divided into three layers: the storage layer, the within-component layer and the run-time layer. The main focus of the model is the storage layer, which describes a database that is composed of a hierarchy of data containing components, called nodes, interconnected by links. The storage layer, therefore, models the basic node/link network structure of a hypernetwork.
The Dexter model is similar to the work reported in this paper insofar as both attempt to characterise the concepts and operations that make HLBSs distinctive from other information systems. For example, the Dexter model describes an interface mechanism for addressing (referring to) locations or items within the content of an individual component. This mechanism, known as anchoring, provides a clean separation between the storage and within-component layers. Anchoring provides the functionality to allow links to be made between documents and also parts of documents.
In the framework proposed in this paper a content specification plays a comparable role to the Dexter model's anchoring mechanism. A C-spec defines content which is to appear in a hyperpage. A C-spec takes the form of data values or requests to DBSs (i.e. query expressions which DBSs can evaluate into data values which are served back). A C-spec may be as simple as a number or a string and as complex as a sequence of complex queries which are to be sent to a variety of DBSs.
In the Dexter model, presentation specifications provide an interface between the storage layer and the run-time layer (i.e. UIS). Such specifications are the mechanism by which information about components presented to the user can be encoded into the hypernetwork (at the storage layer).
In the framework proposed here rendering specifications (R-specs) define how content is to be rendered by a UIS. An R-spec takes the form of formal text (e.g. markups) in a language which the intended UIS can render. This renderable text may be interspersed with template variables that act as placeholders for content as defined by the C-spec with which the R-spec is paired. An R-spec can therefore be used to specify how content is to be presented in a manner similar to that of presentation specifications in Halasz and Schwartz (1990) and Halasz and Schwartz (1994).
The Dexter model does not address the issue of tailoring the interaction process that takes place between users of HLBSs and the system itself. No aspects of personalising the interaction process are explicitly modelled and no consideration is given to users of HLBSs that may have differing information retrieval needs, user goals and backgrounds.
In the Dexter model the storage layer models a hypernetwork as a known database of nodes and links. As such, the size, scope and form of the hypernetwork can be said to be determined at design time, thereby making the HLBS closed. In contrast, the work reported here views a hypernetwork as a collection of specifications to be evaluated at runtime, i.e. the size, scope and form of the hypernetwork is not known in advance, thereby making the HLBS open.
The Dexter model sees HLBSs as being layered and including interconnected user interface and data storage layers. It is implicitly assumed that these layers are aware of and are tightly bound to each other. In contrast, the proposed model does not view the functionality provided by these layers as being tightly bound. Instead this functionality is provided by a loosely-coupled core of hyperlink functionality served by an open collection of possibly remote user interface and content servers. As such the proposed architecture reflects more accurately those of many HLBSs currently being deployed on the Web.
The Dexter model implicitly assumes that applications making up the within-component and run-time layers are known. This monolithic view cannot always be assumed to be true. The array of different presentation, link anchoring and data storage mechanisms that could potentially be employed by HLBSs implies that their behaviour and computation should not be modelled as an integral part of a HLBS. The architecture for HLBSs proposed in this paper may be extended to support such an array of presentation, link anchoring and data storage mechanisms.
In summary, although the Dexter model has provided a solid foundation for research into closed HLBSs, it is less appropriate for representing dynamic interaction patterns across different categories of user. Furthermore, it does not explicitly address issues of personalising the user interaction.
The ISO standard HyTime (Hypermedia/Time-based Structuring Language) specifies the representation of hyperdocuments in a presentation independent format. HyTime allows for the representation and synchronisation of static and time-based hyperlinked information (Newcomb et al. 1991, Newcomb and Newcomb 1992, De Rose and Durand 1994). HyTime is defined as a subset of the Standard Generalised Mark-up Language (SGML) (Heimburger 1994). HyTime supports addressing by name, by position in the document and by semantic construct. Links can be established to documents that conform to HyTime as well as those that do not. HyTime also provides a document query language.
Although the aims of HyTime and of the work reported in this paper differ, several similarities can be drawn. HyTime and the work reported here model hyperpages as specifications in which external content may be referenced. The level of openness that HyTime allows, in terms of the kinds of media and technologies that can be referenced by hyperpage specifications, is comparable to that which is allowed by the model proposed in this paper.
In both models, addressing of hyperpages and their component parts can be specified using semantic constructs. Furthermore, the ability to incorporate links which are specified using an external formal text is also shared.
Work on XML (Bray et al. 1998) bears some resemblance with ours. Both approaches seek to divide a hyperdocument into units which either specify data or delineate regions where data may be obtained. The use in XML of arbitrary-value pairs associated with a document's structure can be compared to our notion of note. The functionality of an XML-based HLBS is not expected to be dissimilar from that of the H-region. However, we propose an abstract model of hyperlink-based interaction and not just of specification. Also, the specification elements in our framework may, without loss of expressiveness or generality, be recast in SGML.
This paper argues that the hyperlink paradigm is one that is well suited to the design of e-book material. This view is supported by the proliferation of hyperdocuments formatted using the Open eBook Publication Structure (OEBPS) (Open eBook Forum Publication Structure Working Group 2001), an XML-based specification for the content, structure and presentation of electronic books. However, it is still rare to find e-books authored in a manner that enables them to be annotated and subsequently personalised. Our future work will aim to address these issues using the results first reported in Ohene-Djan (2000).
Little research has been conducted into formalising personalisation in HLBSs. Most work has concentrated on providing personalisation features via data querying mechanisms (e.g. Hockemeyer et al. 1998). Although these contributions show how personalisation may be experienced, it can be argued that they merely exploit DBMS functionality rather than accounting for how hyperlink-based interaction can be tailored in a principled manner.
Since we have formalised (in Ohene-Djan 2000) a complete space of possibilities for personalisation actions, the adoption of our model enables a system to support all the personalisation actions described in Brusilovsky (1996) and many of those in Brusilovsky (2001). Implemented systems such as Adaptive HyperMan (Mathe and Chen 1996), ELM-ART (Brusilovsky et al. 1996), Hypadapter (Hohl et al. 1996) and AHA (De Bra et al. 2002a) can all be represented using our model.
In its aims, our model subsumes the work on the adaptive hypermedia reference model AHAM (De Bra et al. 1999, De Bra et al. 2002a). However, the explicit aim of the AHAM is to represent techniques that have been implemented and hence serve as a primary reference for comparative studies. In contrast to AHAM, the work reported here aims to induce a set of personalisation actions from a formal definition of a core of hyperlink functionality. In this way, we provide a methodological approach to the area that goes beyond the effort embodied in the AHAM.
The work presented herein concurs with that in Koch and Wirsing (2002) in which the Munich reference model (MRM) for adaptive hypermedia applications is introduced. However, both the AHAM and MRM models are based on the notion that hypermedia applications are closed in nature and can be described using the framework of the Dexter Model (Moline et al. 1990).
The framework for personalisation of hyperdocuments introduced by this paper can be contrasted with the MRM. While both seek to model personalisable hyperlink interaction, the MRM is a reference model that characterises the architecture required for hyperlink-based personalisation, and the model that underpins this research (Ohene-Djan 2000) characterises a set of formally defined personalisation actions. Furthermore, while the model described in the MRM views hyperlink-based systems as comprising user interface and database components, our work views these components as outside the scope of hyperlink-based personalisation. Our view that the design aspects of documents contained within e-books are central to their usability is shared by Landoni et al. (2000b), Wilson (2002) and Marshall et al. (2001). However, we aim to use personalisation as a value-adding strategy rather than, for example, visual clues (Landoni et al. 2000c).
Several customisable hypermedia systems have been developed as commercial products, a representative example of which is Microcosm and its Web version, Webcosm (Davis et al. 1992, Davis et al. 1993). These products are concerned with the integration of hypermedia, database and information retrieval technology to tackle the problems involved in providing access to multimedia repositories and providing protocols for the interoperability of hypermedia systems.
These systems characterise the functionality that hypermedia systems can provide to hyperlink-aware and -unaware applications (i.e. applications that have no notion of linking). In contrast, the work reported here views such a characterisation, at a suitable level of abstraction, to be an essential starting point from which to induce a set of comprehensive personalisation and adaptive actions.
It is clear that current e-book technology has some way to go before it can provide the richness of interaction that owners of paper-based books benefit from. Personalisable hyperdocuments accessible via e-books, however, can be designed to possess characteristics that in some ways supersede those of paper-based books.
The challenge the research underlying this paper aims to meet is how to model, at a suitable level of abstraction, the space of possibilities for personalisation actions that could be made available to readers of hyperdocuments.
Through personalisation of the interaction process that takes place between the reader and the e-book, we have shown it is possible to perform actions that enable users to come closer to satisfying their individual information retrieval needs.
Section 5 exemplified how our framework is a step towards bringing together the complementary benefits, highlighted in section 3, in paper materials owned by a student and hyperdocuments that the student can only use but not personalise. The framework constitutes a contribution to the goal of making e-books better able to address users' information needs on an individualised basis, as advocated in e.g. Benyon et al. (1997) and Brusilovsky (1995).
Benyon, D., Stone,
D., and Woodroffe, M. (1997) "Experience with Developing
Multimedia Courseware for the World Wide Web: The Need for
Better Tools and Clear Pedagogy". International Journal
of Human-Computer Studies, 47(1):197-218
Bray, T., Paoli, J., and Sperberg-McQueen, C. M. (1998) Extensible Markup Language (XML) 1.0. Technical report, W3C
De Bra, P., Aerts, A., Smits, D., and Stash, N. (2002a) "Aha! meets aham". In Adaptive hypermedia and Web based systems: Second International Conference, AH 2002, edited by Bra, P. D., Brusilovsky, P., and Conejo, R, LNCS (Springer), pp. 213-222
Halasz, F. and Schwartz, M. (1990) "The Dexter hypertext reference model". In Proceedings of the Hypertext Standardization Workshop, NIST Special Publications, Vol. 500-178 (Gaithersburg, MD: National Institute of Standards and Technology), pp. 95-133
Hockemeyer, C., Held, T., and Albert, D. (1998) "RATH -- A Relational Adaptive Tutoring Hypertext WWW-Environment Based on Knowledge Space Theory". In Proc. CALISCE 98, edited by Alvegård, C., pp. 417-423
Koch, N. and Wirsing, M. (2002) "The Munich reference model for adaptive hypermedia applications". In Adaptive Hypermedia and Web Based Systems: Second International Conference, AH 2002, edited by Bra, P. D., Brusilovsky, P., and Conejo, R, LNCS, Vol. 2347 (Springer) pp. 213-222
Landoni, M., Wilson, R., and Gibb, F. (2000b) "From the visual book to the web book: the importance of good design".
In Proceedings of the Fourth European Conference on Research and Advanced Technology for Digital Libraries (ECDL 2000), pp. 18-20
Marshall, C. C., Price, M. N., Golovchinsky, G., and Schilit, B. N. (2001) "Designing e-books for legal research". In JCDL'01: Proceedings of the 1st ACM/IEEE-CS Joint Conference on Digital Libraries, Digital Libraries for Education: Technology, Services, & User Studies, pp. 41-48
Moline, J., Benigni, D., and Baronas, J. (editors) (1990) Proceedings of the Hypertext Standardization Workshop, NIST Special Publications, Vol. 500-178 (Gaithersburg, MD: National Institute of Standards and Technology)
Ohene-Djan, J. (2000) "A Formal Approach to Personalisable, Adaptive Hyperlink-Based Interaction". PhD thesis, Department of Mathematical and Computing Sciences, Goldsmiths College, University of London