While classic textbooks are increasingly available online, future e-textbooks have the potential to t within a rich ecosystem of electronic learning tools commonly used in schools and universities. We describe work towards that vision, based on the Platform for User Model Personalized Textbooks (PUMPT) platform. This integrates learning data from multiple e-textbook and other learning tools, notably a learner management system with its data from class assignments, quizzes and exams. All such data can be used to provide an Open Learner Model (OLM) to support learners in self-monitoring, re ection and planning that is based on a holistic view. Our core contribution is the new PUMPT platform for e-textbooks that are integrated with other learning tools. 1
The simplest form of e-textbook is like a paper textbook, with the same content,
but with valuable a ordances, such as easy, fast and powerful searching,
highlighting, bookmarks, translation and easily accessed links to external sources
[
We want to go beyond this, with learning data from the e-textbook integrated
with data from other learning tools that students use in typical university
subjects. Core to our approach is a long term user model and its associated Open
Learner Model (OLM) [
Most work on personalised teaching systems has a learner model as part of
the system [
Commons License Attribution 4.0 International (CC BY 4.0).
The gure shows how the e-textbooks and other learning technology, such
as the LMS, can make use of a Personal User Model for Learning (PUML) [
At the same time this approach ensures the PUML is stored securely on a cloud storage system. This reduces the risks of data loss associated with pure local storage (for example, due to data corruption from device faults, change of technology and user error). Dropbox provides very high reliability and security as well as insulating a user from regular maintenance and upgrades. Importantly, it also means that the PUML can be available to the learner on any personal device with access to the PUML in the Dropbox.
We now turn to the detailed architecture of our PUMPT platform. Figure 2 shows the authored e-textbook content at the left this has the typical elements of e-textbooks, expository content, interactive activities (e.g., multiple choice questions, ll in the blank questions etc).
As the web browser runs locally on the learner's machine, all personalisation and interactivity occurs locally to produce the interface the learner sees. It sends data from the interaction, including information about page views, to the PUML. The user model in the PUML is available to the learner via the OLM. The model in the PUML is also available to personalise the e-text experience.
We now drill down to the technical details that make this possible. The e-textbook is implemented as a set of web pages generated from an extended version of the \reStructured Text"2 markup language. A preprocessor scans the input les in the markup language and converts the extended directives into raw HTML, passing the normal reStructured Text markup through to the input les that the Sphinx3 document processor uses to generate the nal HTML pages. The extended directives that the textbook author can use include: { page meta data: used to report when a user views the page; { tb: learner model evidence from ll-in-the-blank questions; { mcq: learner model evidence from multiple choice questions; { likert: learner model evidence from answers to Likert-scale questions; { OLM: display the open learner model; { log: other logging created by the etext author.
{ showhide: show or hide content based on the user model These are converted to HTML/Javascript which is processed in the browser during user interaction.
Personalisation of the e-text currently takes a very simple approach. Authors
write the e-textbook content as \reStructured Text" markup text. They can use
2 https://docutils.sourceforge.io/docs/user/rst/quickref.html
3 https://www.sphinx-doc.org/en/master/
our extended directive, showhide, to mark up text that is conditionally
presented, depending on the value of a component in the user model. This supports
selective presentation of content, including navigation links. This is similar to
our early work [
The personalisation process that generates what an individual learner sees is controlled by the Personis user modelling system. As the learner loads and interacts with each page of the e-text, the markup directives and the learner model drive the personalisation. We are well aware that this personalisation process imposes a load on the e-text author. They need to add the relevant directives in the markup for the e-textbook. The textbook author also needs to de ne the learner model by specifying the learner model ontology which has: { contexts, which form a hierarchy for example, an e-text on the programming language, C, may have contexts for high level aspects such as control structures and data structures and within these there may be more negrained contexts, such as loops and selection within control structures; { components which are the leaves of the ontology hierarchy for example, for-loops and dangling-else within the context C/control structures/loops; { each component can have rules that are examined when each \tell" adds evidence to the component. Rules may add evidence to other components or send data to external applications; { each component has one or more resolver functions that interpret the available evidence to return the value of the component.
Depending on the nature of the textbook, the design of this ontology may be reasonably straightforward for the author as they design the goals of the e-textbook. Essentially, the design aims to make the simplest personalisation simple to implement, and more sophisticated forms possible with more e ort.
A key part of the system is the publish/subscribe4 paradigm for data handling. As the user interacts with the e-text pages, messages are \published" to \topics" on a local pub/sub message broker. The PUML manager subscribes to these topics, and then receives each message the PUMPT publishes. These update the user model (using the Personis \tell" operation). This is a very simple and elegant approach to managing data from user interaction as it avoids explicit handling of asynchronous server calls. It is also very exible and extensible as there can be more than one subscriber to topics. The publish/subscribe message broker used in the system is Eclipse-paho mosquitto5. This program is compact (<0.25M) and e cient, easily capable of handling the modest message load of a single user. It allows raw TCP connections as well as a web-sockets (a thin protocol layer over TCP used by web applications). Mosquitto is available for Linux, MacOS and Windows.
Notably, there is no central server for managing user models. Each learner runs their own copy of the MQTT message broker and the PUML system and accesses the PUML data in Dropbox (or other cloud storage). Both these components (broker and PUML) run in a docker container6 for ease of installation and isolation. This removes the need for authentication in order to access the PUML since the docker container is running purely locally on the learner's desktop or laptop and is not accessible from the outside network. The model data is stored in Dropbox which requires separate user authentication and is private to the user.
Web pages in the e-text contain the Eclipse-paho MQTT javascript library7 that allows web-socket client connections to the MQTT broker. When a user interacts with the page, for example to do a Multiple Choice Question, messages containing the user response are published to the local broker (using the \localhost" address). These messages are then processed by the PUML manager and the model updated with evidence from the messages.
Inside the PUML, as new evidence is received, Personis resolvers and rules[
A PUML is an example of a Personal Data Store8. From wikipedia: \Personal data services or personal data stores (PDS) are services to let an individual store, manage and deploy their key personal data in a highly secure and structured way. They give the user a central point of control for their personal information (e.g. interests, contact information, a liations, preferences, friends). "
In our case the Personal Data Store contains learning data gathered from user interactions with the e-text. However, the data contained in the PUML could come from many sources, not just the e-text. It could, for example, be collected from multiple e-texts and gather evidence for common learning objects. It could be data from a learner management system (eg Canvas) and include interaction events and test scores. 3
We have presented an overview of the architecture of PUMPT, our e-textbook platform. We developed PUMPT because as authors and teachers, we wanted to create e-textbooks and could not nd a platform with the features we wanted. We now brie y describe them. This paper relates mainly to one of our goals - to enable learners to control their own learning data and its use. Also, we wanted a platform that made authoring manageable in the sense that the technical burden of creating the learning materials was acceptable. At the same time,
we wanted the platform to provide exibility in authoring and in creating new tools for authoring expository content and self-assessments. For these aspects, we settled on use of a markup language (even though a WYSIWYG interface would be nicer and this is on our roadmap). A key goal was to ensure that we could create the e-textbook with an Open Learner Model to support the learner in key metacognitive processes of self-monitoring, re ection and planning their learning. The creation of OLM interfaces is a current project.
We have also been deeply aware of the authoring challenges of even a conventional textbook. This paper has described the personalisation mechanisms we have created these are the most basic approach of marking material for optional presentation to the learner. This will create a signi cant additional content authoring burden on teachers. At the same time it does provide enough exibility for some very valuable forms of personalisation that we consider important. We note that the use of the term, personalisation, is used in many ways in educational contexts. For example, if students are allowed to proceed at their own pace, even using paper learning materials, this is a valuable form of personalisation. In our PUMPT roadmap, this form should be supported. For example, we would like to enable a learner to use their OLM to monitor their progress. But we also consider it critical to enable them to do this in relation to a meaningful standard. A very simple one could be created by the teacher who is using the textbook they could create a timeline for progress so that this could be presented in the OLM so the student could see their progress and compare it with the teacher's expectations. Supporting this form of personalisation would require a simple way for the teacher who is using the e-textbook to create such a timeline. This is part of a larger vision for supporting personalisation that places a layer of information over the textbook. For example, a typical textbook has several chapters and the teacher using it may want to distinguish chapters (or parts of them) that are: { core to the subject and should be the top learning priority for students; { only relevant for students who want to do advanced material (and should only be tackled after the core has been mastered); { not part of the subject and can be skipped.
There are other forms of personalisation, which have recently been distinguished
as customisation, individualisation and adaptation [
A distinctive driver for the design of this architecture has been to provide personalisation that is driven by a learner controlled Personal User Model for Learning (PUML). This builds on our long term work on the Personis user modelling framework that aims to ensure that teaching system frameworks and learning materials are designed, from their foundations to enable the user to scrutinise and control their own learner model and its use.