Initial Social Personalised Adaptive E-Learning
4.2 Architecture
Based on the study of prior and related work (sections 2.2 and 2.3), and the elicited system requirements (section 3.7), the architecture of the initial social personalised adaptive e-learning system has been designed. As shown in Figure 8, it adopts a classical layered structure (inspired by the Dexter model [71], and the more recent SLAOS model [39]), extended with a clear and well-defined social flavour: a Storage Layer, a persistence infrastructure for physical entities; and a Runtime Layer, parsing adaptation strategies to present adaptations and social interactions, and tracking learner behaviours.
In the system architecture, models are represented in the Storage Layer, and can be accessed via the four basic functions of persistent storage, CRUD (create, read, update and delete) [74]. The Adaptation Model (AM) and Presentation Model (PM) are similar to typical adaptation models and presentation models in adaptive hypermedia. The other models present in the architecture are loosely based on the LAOS [41] and SLAOS [39] frameworks. Whilst at a conceptual level, the similarity is stronger to previous frameworks, for the actual implementation, it was
often useful to break models into smaller ones, which interact with each other. Hence, models represented by the architecture often contain sub-models. They and their interactions are described in the following.
Figure 8 System architecture
• The Concept Model (CM) defines the knowledge cell with the minimum
granularity that contains the basic information about this concept, such as its title, tags and description.
76
• The Domain Model (DM) is a knowledge network defining a domain map,
which consists of a set of CMs and concept relationships (structure). This inherits the classical DM defined in the LAOS [41] framework.
• The Resource Model (RM) represents concrete learning content, which can
be a text, an image, an audio, a video, etc.
• The Topic Model (TM) wraps around and basically contains a CM,as well
as one or several RM(s), so it is called a ‘wrapper model’.
• The Module Model (MM) is an overlay over the DM – and defines goal
and constraints maps, which are a subset of structured concepts within a domain map, with goals and constraints given by module (course) constructions. Similarly to the domain maps, goal and constraints maps can, and in practice, often do contain hierarchical structures. This is similar to the Goal and Constrains Model (GM) in the LAOS framework [41]. Here a MM is a self-contained module, which contains structured TMs.
• The Knowledge Model (KM) is an overlay over the DM and CM – a
subset of structured concepts, mapping the learners to the concepts, such as ‘learnt’ or ‘ready-to-learn’ a concept, which can be updated according to the learner’s activities, similar to typical user models in adaptive hypermedia.
• Learner Model (LM) represents a learner’s cognitive preferences and
knowledge space. Whilst the former are recorded in its metadata, the latter is recorded in a KM. It is also called a ‘wrapper model’, as an LM wraps around and basically contains a KM
• The Interaction and Connection Model (IM) is an ‘abstract model’, which
can be ‘instantiated’ as a ‘message model’, a ‘tag model’, a ‘like model’, a ‘share model’, a ‘comment model’ or a ‘note model’ (a ‘comment’ can be seen by others, whilst a ‘note’ can only be seen by its author) and so on. It represents one of the pre-defined social interactions, such as messaging, tagging, liking, sharing, commenting and noting,performed by a learner (with characteristics stored in an LM) to another learner (with characteristics stored in another LM), or to a topic (with metadata stored in a TM, which belongs to an MM). While interacting with each other, social connections are built and maintained by the Socialisation Model (SM).
• The Socialisation Model (SM) is a social network that maintains social
relations between learners. These relations could be either built when learners interact with one another, such as sending or replying to a message, or built when they interact with the same content, such as registering on the same module, or commenting on the same topic. As an
SM is derived from Interaction and Connection Models (IMs) and Learner Models (LM), it is called a ‘derived model’. By doing so, the connected learners could be grouped into a learning community, so as to better support adaptive (expert) peer recommendations and social interactions.
Additional to the above models contained in the Storage Layer, this system architecture also includes an Adaptation Rule Parser (ARP), an Action Tracker (ATR) and a Learner Behaviour Parser (LBP), which are in the Runtime Layer, detailed as the following.
78
• The Adaptation Rule Parser (ARP) combines and analyses the adaptation
rules provided by the AM, in order to support adaptive and/or adaptable client-side presentations, e.g., a webpage that presents a module, a topic, or a profile, etc. In particular, it determines whether to present certain learning content, learning paths and expert peers, and how to present them in terms of personalised navigations, layouts and schemes.
• The Action Tracker (ATR) is embedded in each of the client-side
presentations, e.g., a webpage that presents a module, a topic, or a profile, etc., generalised by ARP. It is explicitly and non-intrusively tracks learner actions, and sends them (raw logging data) to LBP for further process.
• The Learner Behaviour Parser (LBP) collects and analyses the raw
logging data sent by the ATR, to interpret learner actions into meaningful learner behaviours that contains information about learners’ intentions, feeling, behaviour patterns, etc., and then triggers the updating function in the related IM, SM, and LM in the Storage Layer, for persistent storage.