1 Introduction

There has been a growing requirement for learning with the rapid technological and social change. The traditional face-to-face learning cannot meet such a requirement and online learning and blended learning are being widely used to meet different learning demands. It has been recognized that information and communication technologies can improve learning by enabling more learners to have access to learning resources and providing learners and teachers with unprecedented flexibility and convenience (Shen et al. 2002).

Yet, to date technology is mostly used in the distribution and management of learning materials. Most of current online instructional systems just simply deliver course materials over the Internet. They offer a “one size fits all” approach to the delivery of learning materials, the support for personalization of learning is sadly lacking from most online instruction systems (Cristea 2005). In such systems, information is presented to learners through a set of instructional sequences with predetermined outcomes. Learners are directly told about a solution for the problem under study or are taught how to get a solution by a predetermined mode and route. They are denied any opportunity to independently explore possibilities, make their own solutions, and actively construct new knowledge in the process of learning.

Clearly this online instruction mode cannot satisfy present day’s needs for customizing learning to meet particular demands as it is cost prohibitive to develop customized systems for each particular need. This paper claims that what is needed are generic systems that can be used to easily customize online learning environments by putting predefined components together in different ways to satisfy different learning needs. We propose an approach shown in Fig. 1 to identify such components. Here we first look at what is happening in the real world, the current practices in learning and teaching online. Then rather than simply translating each practice to a computer implementation, it develops a general conceptual model that can be used to describe any number of practices. At the same time it identifies the common components that can be applied to various learning requirements. This is the first step in introducing the kind of structure needed for implementing flexible computer learning systems. The conceptual model terms can then be used to model various particular instances. Since learners often work together with their fellows to collaboratively solve practical problems, collaborative learning conceptual model is used to model the collaboration among learners. A mapping is provided to convert these models to an implementation, usually a learning space. The corresponding services are then developed using technological means such as software agent technology.

Fig. 1
figure 1

Our approach to developing personalized online services

Full size image

In the paper we will use this way to investigate the required services to facilitate knowledge construction of individual learners in online learning, and present an agent-based approach to implement the individualized supportive services. The paper is organized as follows. In the next section, the relevant learning theories are reviewed to build a theoretical background for this study, where it is emphasized that the personalized supportive services will mainly aim to facilitate constructivist learning and they will also reflect other valuable learning theories. In Section 3, a general conceptual model to describe the common practices in learning and teaching online is built. In Section 4, the services to assist individual learners to constructively build knowledge are identified based on the requirement in the online learning process. Section 5 combines the existing technologies together to realize the customized supportive services for individual learners. The involved multi-agent architecture is explored in Section 6. The implementation of the individualized services is addressed in Section 7, where a system prototype is presented. At the final are the conclusions and some future research directions.

2 Theoretical background

In order to facilitate knowledge construction of individual learners through providing them with customized supportive services, it is essential to investigate how the individual constructs knowledge and how to support the construction from the perspectives of cognition. Learning theories have investigated into this issue for a long time and different learning theories have drawn different results. Among them objectivism and constructivism are the major and pervasive learning theories.

According to objectivism, knowledge objectively exists outside of the mind of the individual in the world and can be transmitted directly from the head of a teacher to the heads of learners (Jonassen 1991). Thus, the learner is to passively receive the information, whereas the role of the teacher is to send the information. Constructivism, however, argues that knowledge cannot be transmitted but must be individually constructed and socially co-constructed by learners through interaction with the surrounding environments (Jonassen 1999; Mayer 1999). The learner is an active knowledge-constructor, whereas the teacher is a cognitive guide who provides the constructor with individual guidance and scaffolding to support the construction. This subtle difference has profound implications for how to support and facilitate learning. The five among the differences between the two learning theories are highlighted in Table 1 (French et al. 1999).

Table 1 Comparisons of constructivist learning and objectivist learning
Full size table

Constructivist learning is the central conception of constructivism. The fundamental epistemological assumptions underlying constructivist learning can be summarized as follows (Gagnon and Collay 2004):

  • Knowledge is physically constructed by a learner who is involved in active learning

  • Knowledge is symbolically constructed by a learner who is making his or her own representations of the information presented to him/her

  • Knowledge is socially constructed by a learner who conveys his/her meaning making to others

  • Knowledge is theoretically constructed by a learner who tries to explain things he or she doesn’t completely understand.

Learning by adopting a constructivist method can generate more significant results than by adopting other methods (Wilson et al. 1995) because constructivist learning focuses on knowledge construction of learners. The prime problem for constructivist learning is that it requires learners to possess a certain skill on independent learning. Earlier research has illustrated that not all learners are equally capable of adequately constructing knowledge on their own (Britain 2004). For instance, some may lack necessary prior knowledge or abilities to independently choose a learning resource and adopt a proper method to conduct a learning process. Consequently, supportive services are necessary to assist learners to study independently. These must include services that go beyond simply presenting course materials, but supply a wider range of technological facilities, tools and services to support the learning process.

Constructivists believe that knowledge is constructed by learners through actively interacting with and exploring the surrounding environments (Jonassen 1991). Accordingly they have to actively take their own learning activities to make meaningful understandings of the study theme. To support and facilitate their meaning-making, teachers, tutors or even the online instruction system should not offer them with the solution for a problem under study or impose them to attain a solution using a designated mode. Rather, they should be provided with a constructivist learning environment (CLE).

The CLE is a rich learning environment, where learners may work together and support one another as they use a variety of tools and information resources in their guided pursuit of learning goal and problem-solving activities (Wilson 1996). Its core is to facilitate learners engagement in active manipulative, constructive, intentional, complex, authentic, cooperative and reflective learning activities (Jonassen and Rohrer-Murphy 1999). It provides learners with opportunities to construct new knowledge based on prior one from authentic experience. Learners are allowed to confront problems full of meanings. While solving a problem, learners are encouraged to explore possibilities, invent alternative solutions, collaborate with others, try out ideas and hypotheses, revise their thinking, and finally reach to a best solution. In short, the CLE provides (Jonassen 1999):

  • a learning space for problem representation and manipulation

  • related cases, experience and appreciation for conveying multiple perspectives

  • cognitive tools for various metacognitive activities

  • support for learner-centered learning activities

  • computer-mediated communication tools for conversation and collaboration

  • criteria and methods for evaluating learning outcomes.

The understandings of objectivism and constructivism, and in particular the CLE, will be used as a theoretical background for the study presented in this paper. Building upon this background, we will design supportive services for individual learners based on his or her unique learning characteristics. We encourage the individual to build knowledge by using constructivist ways. In the meantime, we do not reject any other valuable learning theories, including the objectivist ones. Although in the above constructivism and objectivism were conveyed as two extremes in order to contrast their differences, they are not mutually exclusive in actual educational practices. We provide supportive services for the individuals based on a synthetic utilization of the two learning theories, which not only reflects constructivist perspectives for learning but also reflects the objectivist ones. The degree to which constructivist principles will be reflected in the services for an individual will be customized based on his or her unique learning characteristics.

In the following sections, we will follow the way showing in Fig. 1 to analyze the common requirements in online learning and develop appropriate solutions to meet these common needs, with the flexibility to tailor generic solutions to meet individual specific requirement as required.

3 Description of the practices in learning and teaching online

To develop a general conceptual model to describe any number of practices in learning and teaching online, two approaches will be presented in this section based on the investigation into the current practices. One is a top level description whereas the other is a more detailed conceptual framework to describe a unit of learning. The former identifies general components whereas the latter defines a framework to integrate the components into an online learning process according to the unique learning characteristics of individual learners.

3.1 Top level description

In the top level, we first look at the online learning process itself and the kinds of components that make up the learning process. The supportive services are then identified based on the requirements for the process steps. The main components of such a conceptual model for the online learning process are shown in Fig. 2. They include:

  • learning environment where learning takes place, which refers to the physical as well as to the social environment in which learning takes place and might include physical entities, tools, and people

  • learning goal describing what to be achieved through the learning activities

  • learning plan defining what learning activities will be carried out and the sequence of the activities to be taken to achieve the learning goal

  • learning activity describing the action to be performed for the learning plan, e.g. writing a report, evaluating a problem

  • subject metadata providing explicit references to learning resources needed in learning

  • learning method defining the way how to conduct learning, including which learning material will be used, what actions will be taken, and how to evaluate the learning in each step of the learning activity

  • supportive services describing the services provided for various learning methods respectively.

Fig. 2
figure 2

The top level conceptual model for describing learning activities and processes

Full size image

This top level description illustrates a framework for placing the learning activities within a practical context. Thus for example, supportive services depend on the environment in which learning takes place. The environment can be a university or a business entity. The learning proceeds in accordance with a learning plan. How the plan is set depends on the learning environment and the learning goal. The revision of the learning plan depends on the practical progress of the learning in the environment. Similarly the method used to support learning activities depends on available supportive services and learning process. It is used to identify the kinds of engagements to be supported.

3.2 Learning process description

The top level description illustrates the kinds of components that make up a learning process. We still need a way to, based on the particular learning characteristics of individual learners, put all the relevant components together to build a learning process where his/her knowledge construction can be facilitated through a wide range of supportive services. We use the conceptual framework shown in Fig. 3 to describe learning activities and processes and the related supportive services to facilitate the individual’s knowledge construction. The framework is developed based on the earlier work in the description of learning activities, particularly the Educational Modelling Language developed by Koper (2001). It describes learning activities and processes and the corresponding services to support learning through a series of units of learning (UOLs). A UOL is an abstract term used to refer to any delimited piece of education and training, such as a course, a lesson, a module, and so on (IMS 2003). As shown in Fig. 3, a UOL is composed of seven compound fields, i.e. metadata, roles, content, methods, assessments, cases and plans. Each contains more elementary fields, constructing a complex hierarchical architecture (Pan and Hawryszkiewycz 2004). The major fields and their roles for the realization of the services to support knowledge construction are outlined below.

  1. (1)

    Metadata field is to provide a general description of the UOL, including its ID, title, prerequisites, and learning objectives.

  2. (2)

    Roles field is to specify the intended users of the UOL, such as learner or tutor. They can be further categorized through the Property field, where an identifier of the category is defined and a brief description to the feature of the users in the category is provided. This design enables the supportive services to be customized according to the unique features of individual learners.

  3. (3)

    Content field is to describe all the learning resources and all the learning activities related to the UOL. This design is to enable the system to provide supportive services not only concerning learning resources but also various learning activities, and their combinations.

    A learning activity is an action some learners will do in the learning process. Within the description of each learning activity, type specifies the category of the activity, what gives a textual description of what will be done in the activity, complete indicates the status change after the activity is completed (the status change will affect the sequence of learning activities), and activity output specifies the artifact files for evaluating the outcomes of the learning activity.

    A learning resource is a learning material some learners will use in the learning for the UOL. It can be in any form but mostly is a Web resource. Within the description of each learning resource, traits specify the specific features of the resource, content object indicates the medium type of the resource and its exact location, communication object illustrates the requirements for the communication facilities, and tool object specifies the prerequisite tools and facilities for using the resource.

  4. (4)

    Methods field is to define all the learning methods to achieve the objective of the UOL for different categories of learners, through putting the predefined components together in different ways. A learning method is referred to a way to conduct learning, including the learning materials to be used, the learning activities to be taken and their sequences, the evaluation methods, etc. The learning methods are categorized based on the learning characteristics of their targeted learners and divided into different groups accordingly. Each group is put into an activity structure field. This implies that the learning methods defined in an activity structure field suit a particular category of learners.

    Within an activity structure, there can be multiple activity sequence fields. An activity sequence defines a particular “learning flow” in which a sequence of learning activities is performed during the learning process. Such a design is to let the system provide learners with multiple optional activity sequences that suit their learning characteristics. Each activity sequence can be associated with more than one learning resource, ensuring an activity sequence can be conducted by using different learning resources. Each activity sequence can be associated with more than one assessment approach, allowing the outcome of learning to be evaluated from different aspects. Also, each activity sequence can be associated with multiple related case study materials, letting the system scaffold knowledge construction through providing multiple optional case study materials.

    Within an activity sequence, a learning step can be a reference to a learning activity defined in the UOL or a reference to another UOL. In the latter case, a learning step is the execution of other UOL; this way lets a hierarchical architecture of a module or a subject be defined if required. The activity sequence may also define a conditional learning path of the given learning steps, allowing sequencing, conditions and repetitions of some learning activities. The type field is to specify the methodology category of the activity sequence, e.g. knowledge-acquisition, problem-based, case-based, project-based, informative testing, learning by designing, discovery learning, etc. Service options declare the computer supported collaborative tools that are required in the learning activities, e.g. discussion forums, etc. Traits indicate the particular features of the sequence.

  5. (5)

    Assessments field is to provide a description of the assessment approaches for the learning activities defined in the UOL. Within the description of each assessment approach, type specifies its category, e.g. question-answering, interactive program, etc.; traits specify the specific features of the assessment approach; and source indicates where to find it.

  6. (6)

    Cases field is to provide a description of the case study materials related to the UOL for scaffolding the learning for the UOL. Within the description of each case study material, type specifies its category, e.g. document material, interactive program, etc.; traits specify the specific features of the case study material; and source specifies where to find it.

  7. (7)

    Plans field is to link a learning method with its targeted learner category. It contains a series of plan fields, each of which is a pair of a property ref and an activity structure ref. The former refers to an ID of the property defined in roles whereas the latter refers to an ID of the activity structure defined in methods. The field is crucial to dynamically organize a specific learning mode and provide associated supportive services for individual learners.

Fig. 3
figure 3

The major components in a UOL and their mutual link

Full size image

As seen from the role of the seven fields, the framework can be used to describe any number of practices in teaching and learning online. At the same time, it can also be used to model particular instances by combining the relevant components, including the sequence of learning activities, learning resources, assessments, and case studies, in a particular way for individual learners. Thus, it enables system to provide personalized services for the individual through customizing the supportive services. With such framework, we can design supportive services for different learners through careful learning design for each category of learners (Koper and Tattersall 2005).

4 Supportive services for individual learners to build knowledge

The previous section has identified the general components in online learning and defined a framework to integrate the components into an online learning process according to the unique learning characteristics of individual learners. What supportive services should be provided for the individuals? This section will structure an online learning process and then, based on the practical requirements for online learning, identify the supportive services to facilitate the individual’s knowledge construction.

4.1 An instance of an online learning process

From the perspectives of knowledge construction, an online learning process can be structured and an instance can be described in Fig. 4. It is initiated and driven by a learning goal. Here the activities are represented by black dots. After a goal is constituted based on a project or a problem under study, the learner experiences a guided process to reach it. The process starts with a goal, which is followed by building a plan to achieve the goal. This includes defining the learning activities and designing the methods used in these activities. The methods are chosen based on his or her particular cognitive characteristics and learning history. Then, the learner carries out the plan to construct knowledge in a domain. He/she performs the learning activities in sequence as defined in the plan. Each learning step is a particular learning activity, such as accessing learning resources, discussing with other learners, doing assignments, doing self-assessment, requesting assistance from teacher or fellow learners. When the learning is progressing, the learner manages the plan to align learning towards his/her goal. He/she records the learning activities that have been done, evaluates their outcomes, and then changes the current plan if it is necessary based on the evaluation. The revised plan will immediately affect the learning process; the relevant activities or sequences will be aligned. The learning based on the updated plan will be evaluated again and this may result in the further revision of the plan. The learning proceeds in this way until the evaluation illustrates that the learning goal has been achieved.

Fig. 4
figure 4

An instance of a constructivist learning process

Full size image

Using this way to describe an online learning process takes the process as an individual learner’s self-motivation and self-regulation process of building ideas and concepts through reflection, abstraction and interaction with other learners. The center of learning is placed on concept development and understanding through appropriate activities. A prime emphasis here is the self-adjustment organized by the learner while the learning is proceeding.

4.2 Personalized services to facilitate knowledge construction

The description of an instance of an online learning process shown in Fig. 4 provides some significant clues on what supportive services should be provided to assist individual learners to construct knowledge in the learning process. The services must include: (1) providing access to appropriate learning materials and learning strategies that meet the learning requirements and match to his/her unique learning characteristics; (2) fostering meaningful interactions with content, teachers, and other learners; (3) promoting collaborative learning among learners in groups; and (4) aiding to timely and accurately evaluate learning outcomes.

The services are directly aimed to solve the problems probably emerged in the practical online learning process and are customized for the individuals to meet his/her particular needs in online learning. Furthermore, the services are also provided for different learners in different ways. The purpose is to facilitate effective and efficient knowledge construction for all individual learners in the online learning process.

Figure 5 shows a few examples of the personalized services we have already developed to assist individual learners to construct knowledge by using software agent technology. When a learner builds a plan for a learning goal, an agent will advise him or her of several suitable plans to reach his/her goal. When he or she executes a plan to pursue his/her goal, the agents will provide him/her with a wide range of supportive services to assist him/her to conduct relevant learning activities. These include suggesting learning resources for the learning theme he/she is studying, suggesting online discussion forums for the theme, suggesting suitable assessment methods for doing self-assessment, suggesting people for getting assistance, etc. Meanwhile the learning activities taken by the learner are monitored and evaluated by the agents, and suggestions for learning adjustment will be presented to him or her whenever necessary. The overall objective of these personalized supportive services is to assist him/her to advance the learning steps towards the achievement of his/her learning goal.

Fig. 5
figure 5

Examples of the services to promote construction of knowledge

Full size image

5 Combining technologies together to customize the supportive services

To customize the supportive services identified in the previous section for individual learners, an agent-based approach is proposed. In the approach, several existing technologies are combined together to realize the personalized supportive services for individual learners. The relevant learning theories are used as a pedagogical foundation. The conceptions of learning objects and learning designs are respectively applied to dynamically organize learning materials and learning modes. The top level conceptual model and the more detailed conceptual framework, investigated in Section 3, are used to define and specify learning activities and processes and associated services for individual learners. Moreover, software agents are incorporated into the online learning environments to realize the customized services for the individuals. In this section, the overall architectural framework of the approach will be outlined, and the way to catering for learners in individual manners will be explored.

5.1 Overall architecture

Based on the understandings of objectivism and constructivism, and in particular the CLE, we have developed an agent supported learning system (ASLS) to facilitate the individual’s knowledge construction. Its overall architectural framework can be depicted in Fig. 6. Individual learners log into the system for learning via networked digital devices. They can study individually in an electronic learning space, and if they wish, they can learn collaboratively with other learners in groups through the learning spaces. The agents work in the background, independently of learners, observe and monitor the events that happen in the learning spaces. Any change taking place in a learning space made by an individual learner is represented as an event. The agents autonomously detect these events and take actions to respond them.

Fig. 6
figure 6

Overall architecture of the ASLS

Full size image

To customize supportive services for individual learners, the agents build and timely update profiles for the individual learners through collecting detected events and inducing from these events. They monitor and evaluate the learning of the individuals and present suggestions and advise them when necessary. All the suggestions or advice are tailored to the individual; they are dynamically organized based on the actual learning scene and the learner’s profile. The agents present suggestions or advise learners through the learning spaces.

All the rules related to provide what supportive services for different learners in which learning scenario are stored in the UOL database. The UOL database contains a collection of careful designed learning units, each of which, by using the conceptual framework shown in Fig. 3, defines how to organize the learning process to achieve the objective of the unit according to the unique learning characteristics of different learners. The agents provide services for individual learners mainly according to these rules.

In the ASLS, the agents provide supportive services for learners in a form of suggestion or advice. Learners are not imposed to take any learning steps. They can decide whether and how much to use the service provided by the agents. Their autonomy in learning is sufficiently supported and encouraged. Conceptually, the agents work like a “constructivist teacher”. They encourage learners to actively build knowledge by using their own ways. They support learners independently develop and explore their own learning methods for a study theme and actively construct meaningful understandings of the theme. Meanwhile, they will provide the individuals with spontaneous services to direct and scaffold the process whenever necessary. As identified in the previous section, the services are directly aimed to help them solve various problems which probably emerged in practical learning. This design contrasts with most current online instruction systems, where they just deliver course materials to learners and leave themselves to decide how to use these materials to build knowledge. It also contrasts with most intelligent tutoring systems, where learners are forced to accept the services chosen by system based on the used learner models, expert models and tutorial models. In fact, they are imposed to conduct learning by following a pre-set path defined by system. As Jonassen (2000) argued, those paths cannot possibly specify all of the ways in which learners may go about trying to solve a problem due to their different backgrounds, interests, styles, motivations, and capabilities. As a result, learners are often forced down a learning path that does not suit them or even limits the development of their cognitive abilities.

5.2 Catering for learners in individual manners

All the supportive services in the ASLS are customized so as to accommodate learners in an individual basis. The service contents are dynamically organized according to the learning characteristics of individual learners. The agents provide personalized supportive services for individual learners through two dimensions, the service content and the presentation mode. In general, different learners are provided with different services and by different manners.

In order for the agents to provide the individuals with supportive services that exactly match to his or her learning characteristics, there must be ways to let the agents know his/her learning characteristics. Thus, it is necessary to investigate into the learning models for different categories of learners. Yet, in general, learner models are currently considered as an intractable problem. The classification of learners usually makes use of the objective knowledge about the learner’s previous knowledge, skill level performance, and learning styles (Shi et al. 2004).

In the ASLS, learner profiles are used to describe the learning characteristics of individual learners. The profile of a learner is characterized by the following three dimensions: (1) knowledge constitution, (2) cognition ability, and (3) learning styles. All the services for an individual learner are dynamically organized according to the three dimensions stored in his or her profile and the practical learning scene. The knowledge constitution seeks to describe what an individual has already learned so far, i.e. the learning history. The cognitive ability seeks to describe the ability or skill that an individual constructs knowledge. It is categorized by six levels based on Bloom’s taxonomy: knowledge, comprehension, application, analysis, synthesis and evaluation. The learning styles seek to describe the mode by which an individual studies. They are characterized by a set of his/her preferred styles in online learning derived based on Kolb’s classification of learning styles.

In ASLS, the agents try to provide supportive services for the individuals in a mode that meets his or her learning requirements. In the selection of the mode of presentation, different requirements of learners with different cognitive skills are taken into consideration. It can be easily observed that learners require different presentation modes of services and different intervention degrees in online learning. As an example, for a prompt message presented by an online system in a particular scene, some learners may feel it is annoying because it is unnecessary for them, but on the other side, other learners may think it is valuable and helpful, and some of them may even think it is not enough for example they may wish to receive more such prompt messages in the similar scenes. Without doubt, different strategies should be adopted to offer the services for different learners so that all of them can have emotional connections with these services; facilitating their enjoyment of the learning situations. The overall principle used to deal with this issue in our implementation includes: (1) to reduce the cognition overload of learners as big as possible; and (2) to assist learners as non-annoyingly as possible. In our current implementation, different strategies are used to present supportive services for different learners according to their cognitive ability level. For example, learners with lower cognitive ability level are probably presented supportive services in a step by step mode. The contexts are clear and concrete instructions for what to do next and how to do. Moreover, all the assistance is actively presented to them. For the learners of higher cognitive ability level, only limit assistance is offered, mainly through suggestions, reminding or prompting.

6 Software agents for the services

As stated above, software agents are employed in the ASLS to realize the individualized services aiming to facilitate effective and efficient learning experiences for individual learners. This section describes the agents used in the system and illustrates how they collaboratively work towards the achievement of the goal.

6.1 Process agents

Software agents are an attractive paradigm for developing intelligent software system (Jennings and Wooldridge 1998). An agent in this study is an active, persistent software component, situated in the online learning environment, that is capable of taking flexible, autonomous, and reasonable actions to the change in that environment in order to assist learners to construct knowledge (Jennings 2000). It can exhibit several specific properties, such as autonomy, social ability, responsiveness, and pro-activeness. These distinguished properties enable users to utilize agent technology to build intelligent entities in AI projects. Already a lot of such agent based systems have been reported on the literature, ranging from comparatively small systems such as personalized email filtering to large, complex, or even mission critical systems, such as air-traffic control, workflow management, information retrieval, data mining (Jennings et al. 1998). Generally, agents are used as an overarching framework for bringing together the component AI subdisciplines that are necessary to design and build intelligent entities.

In particular, software agents have been applied in online learning (Shang et al. 2001). One is the pedagogical agents (Baylor 2004) that take a role in learner’s learning process such as tutors, mentors, or co-learners. Another is the agents that provide specific functions (McArdle 2005), such as interface agents or navigational agents. This study proposes a new class of agents, process agents, which promote online learning not through understanding the academic contents of the subject that learners are learning, but rather through providing a broad range of process-oriented services to assist learners to get through the learning process. They assist learners to get access to learning materials, provide learning plans for learners to achieve their learning goal, suggest online discussion forums for learners to participate in the discussion about the theme under study, aid learners to timely evaluate learning outcomes and adjust learning whenever necessary. Process agents are domain independent and thus they do not need to understand the knowledge of a particular subject as pedagogical agents usually do.

6.2 Agent classes in the ASLS and their tasks

The agents we have developed to implement the services to facilitate knowledge construction of learners are a group of individual agents with specific expertise, which forms a multi-agent society. Each agent has a particular intention and undertakes a particular task. They work together towards a common goal: customizing the supportive services for individual learners to facilitate his or her knowledge construction in online learning. Currently three categories of agents are being employed in the ASLS to realize the services:

  • Learning assistant agent (LAA) is an agent for an individual learner. Each learner is assigned an agent when he or she logs in for learning. It interacts with the learner and serves as a personal assistant in his or her learning process. It continuously observes the behavior and actions of the learner to maintain a profile for him or her. It manages the profile as the learning proceeds, and provides the profile information to other agents when being requested.

  • Planning agent (PA) is an agent responsible for customizing learning methods for the individuals through putting the relevant components together in the ways that match to his/her particular learning characteristics. The principal task of the PA is to assist the individuals to determine a learning method for a particular goal, which includes determining the learning materials to be used, the learning activities to be taken and their sequences, the evaluation methods, etc. All of these are determined based on the practical learning scenarios and the unique learning characteristics of the individual learners. The PA also assists other agents to help the individuals align learning towards his/her goal when being requested by other agents.

  • Managing agent (MA) is an agent responsible for managing the learning process of a unit. Each UOL has a specific agent, a MA, for assisting individual learners to perform the learning activities defined in the UOL. The MA manages the learning of a UOL by following the learning plan for the UOL being used by the learner. It delegates work to the learning activities according to the plan. The MA keeps track on the progress of the learning activity in the UOL and provides assistance for learners in revising learning plan accordingly. This includes monitoring the submission of the artifact file for the UOL, evaluating the artifact file or asking a domain expert to evaluate it and then receiving the evaluation result from the expert.

The individual agents in the multi-agent architecture support a hybrid architecture that combines a reactive reasoning and a BDI based proactive reasoning (Wooldridge 1999). When an agent detects an event, it uses the event to match the event–condition–action rules that it has been appointed to. According to the matched rule, the agent may execute an action by a reactive reasoning or activate a goal to be achieved, i.e. initializing a BDI based proactive reasoning (Lin et al. 2003). In both cases the result will change the environment and the change will produce new events. The new events will further result in other agent actions. The agents respond events in this way towards their goal which they have been designed for.

6.3 Interactions of the agents

To achieve the designated goal, the agents in the multi-agent architecture need interactions with each other to take collaborative actions. They interact through exchange of messages. A message carries a piece of information that the sender wants the receiver agent(s) to know. Agent messages in this architecture follow a specific format that obeys the KQML standards (Finin et al. 1994).

Because it requires coordinated efforts by a number of agents to assist and guide a learner to get through a learning process, the interaction among these agents is very complicated. Here we illustrate the interaction and collaboration by a simplified example where only the selected agents are involved. The LAA captures the learning requirements and learning characteristics from its owner learner through monitoring and observing his/her learning. It maintains a profile for the learner and provides this information to other agents when being requested. When the agent perceives its owner learner has set up a learning goal, it requests the PA for learning methods to assist its owner to reach the goal. The PA, based on the learner’s profile, customizes learning methods for the learner by putting the relevant components together in a particular way, and, through the LAA, provides these methods to the learner for making decisions. Every learning methods determines a strategy for how to conduct learning towards the goal, including the learning materials to be used, the learning activities to be taken and their sequences, the assessments, the case studies, etc. When the learner starts to learn a unit, a MA is created for assisting the learner to perform the learning for the unit. The MA, based on the sequence of the learning activities scheduled in the plan being adopted, delegates a learning activity, creates a learning space for the learning activity, and builds a MA for it as well. The agent for the activity further creates MAs for actions to assist the learner to conduct related actions in the activity and reports the progress to the MA. The MA will delegate another learning activity after it receives the report from the agent for the activity on the completion of the activity it is associated with. If a problem occurs in the execution of the plan, the MA reports the problem to the PA, and the latter will revise the plan accordingly.

In the above process, some agents start working and some stop running along with the progress of a learner’s learning. A specific mechanism, message delivery broker (MDB), is responsible for the administration of all the agents in the system. Each agent in the system has a unique ID, and has a message box that stores incoming and outgoing messages. The MDB is announced to all the agents when they are established. Every agent knows it and registers in it when being launched. The MDB stores the information of all agents, includes their ID, creator, as well as the services they have registered to provide. Hence, it can transport messages between the registered agents.

7 Implementation of the services

A prototype of the ASLS has been developed and several personalized services to facilitate knowledge construction of individual learners have been implemented. This section will briefly introduce the system prototype, and then illustrate how the agents realize these services through an example.

7.1 System prototype

In the prototype, the agents work in the background, independently of learners, observe and monitor their learning, and provide supportive services for learners when necessary. All the supportive services for the individuals are dynamically organized based on the actual learning scenes and his/her unique learning characteristics. The agents actively offer supportive services for individual learners when they believe it is necessary. Also, learners can ask the agents for providing assistance at any time when they want.

Here we use Fig. 7 to illustrate what the agents can offer the individual in his or her online learning. The figure is the screen shot of an interface the prototype displays when a learner requests the agent services. The “Agent Services” window in the interface provides a basic vision for what supportive services the agents can offer an individual in a learning process. As seen from the window, the agents can offer:

  • learning materials for the current learning theme

  • learning plans for the current learning theme

  • online discussion forums for the current learning theme

  • evaluation modes on the learning of the current theme

  • case study materials for scaffolding the learning of the current theme and

  • contact people to collaborate or to get assistance for the current theme.

Fig. 7
figure 7

The supportive services provided by the agents

Full size image

In addition, the agents allow the individual to view his or her profile the agents have maintained for him/her so far, and to customize the service settings to meet his/her particular requirements.

All these services are customized for the individual according to his/her unique learning characteristics based on the actual learning scenes. The agents attempt to facilitate his or her knowledge construction through these customized supportive services.

7.2 Implementation of the services

A number of supportive services to facilitate knowledge construction of individual learners have been successfully implemented in the system prototype. These include assisting them to get access to learning materials, and other kinds of assistance such as those concerning learning plans, learning assessment, case study, and online discussion forums. All the supportive services for an individual learner are dynamically organized based on the actual learning scene and his or her profile according to the rules stored in the UOL database. The way to assisting individual learners to dynamically adjust learning has been reported in (Pan and Hawryszkiewycz 2006). In the remainder of this section, we will, through an example, present an overall vision on how the agents customize the supportive services for individual learners.

Figure 8 sketches out the implementation procedure that the agents customize learning materials for an individual learner according to his or her learning characteristics based on the learning goal he/she is pursuing. In this procedure, a unit is first identified from the UOL database through matching the learner’s goal to the objective of the unit. The UOL record for the unit is captured from the UOL database. Then, all the learning materials for the unit are extracted from the UOL record. These are the ones that can be used for learning the unit, but only some suit the learner; some do not suit him or her in terms of learning characteristics. Accordingly, at the next, the suitable ones will be chosen from them based on the learner’s profile. At the first, the ones that suit the learner in terms of the cognitive ability level are chosen. Then these are further filtered based on the learner’s learning styles. Only the learning materials whose traits most match the learner’s learning styles are left. Finally, these learning materials are presented to the learner.

Fig. 8
figure 8

Customizing learning materials for a learner

Full size image

To judge if a particular learning material is suitable for a learner, the agent compares the Traits field of the learning material against the learner’s profile. As described previously, the learning styles of a learner are used as one of the dimensions describing the learning characteristics of the learner in our work. They are stored in his or her profile and are characterized by a set of his/her preferred styles in learning. Every learning material stored in a UOL record has a traits field that describes its traits. If a learning material can accommodate a particular learning style, then the style is put into its traits field. Consequently the comparison of the traits field of a learning material against a learner’s profile can tell if the learning material is suitable for the learner. The fit degree is calculated by summing all the learner’s favored styles that can be accommodated by a learning material. A learning material is considered as an appropriate one for the learner if its fit degree is larger than a designated threshold value. A learning material is considered as the optimal one for the learner if its fit degree is larger than any other learning material’s.

By the above procedure, the agents have customized the learning materials for the learner that most match his or her learning characteristics. In order to promote active learning of individual learners, the agents do not force the learner to accept any of the learning materials they have customized for him or her. Instead, they present these materials for the learner in a form of suggestion or advice to let him/her make decisions. Figure 9 shows a typical scene, where a learner is being presented with three learning materials, which have been customized for him or her to learn unit “Database logical design” based on his/her profile. The learner is free to accept any one of them or reject them by using his/her own preferred one. Here the agents provide the learning materials with multiple options and allow the learner to use his/her own one. Such design is to ensure the learner can use a learning material that most satisfies his/her particular requirements. It encourages him or her to actively build knowledge in his/her personal preferred ways. This is consistent with constructivist theories for learning.

Fig. 9
figure 9

Presenting learning materials to a learner

Full size image

The implementation of other customized supportive services is similar to the one presented here. The cognition level of a learner is taken into account at first when a supportive service for the learner is dynamically organized. Then, his or her particular learning styles are further considered to determine the service content that most suits him or her. Also the service is provided for the learner in a form of suggestion or advice to encourage the learner use his/her own way to build knowledge.

8 Conclusions and further work

To make online learning more productive, a wide range of supportive services for online learners are necessary. A principal challenge is learners may require different supportive services due to their particular backgrounds, interests, styles, motivations, and capabilities. This study attempts, by following the way showing in Fig. 1, to identify a generic system that can be used to easily customize online learning environments to cater for different needs of learning by putting predefined components together in different ways. An agent-based approach to realize the customized supportive services for individual learners to facilitate his or her knowledge construction is proposed. A prototype has been developed and a preliminary evaluation has been carried out. The research and practice, although still at a preliminary stage, indicates this way is effective to provide personalized supportive services to facilitate knowledge construction of individual learners.

The personalized supportive services include the provision of learning materials and methods for conducting learning as well as the dynamic adjustment of learning processes. These services are aimed directly to help the individuals to solve various problems he or she probably encounters in online learning and thus benefit him or her to continue the pursuits for his/her learning goal. Meanwhile the services are non-intrusive; learners are free to accept them or ignore them by using their own ones. The autonomy of learners in learning can be sufficiently supported. They can ask agents for assistance at any time when they want and the agents also actively offer suggestions or advice for them. Furthermore, all the service contents contain multiple options to facilitate and encourage learners to use constructivist approaches to build knowledge. These services can facilitate learner’s active construction of meaningful understandings of the study themes and their creative participation in online learning.

All these services are tailored to individual learners and the service contents are dynamically organized and generated depending on his or her cognition abilities and learning styles stored in his/her profile. The related rules are stored in the UOL database developed through careful learning designs. The conceptual framework shown in Fig. 3 is used to define and specify learning activities and processes and associated supportive services for individual learners. Facilitating personalization of learning can be further enhanced through two means. One is to refine the learning characteristics of learners stored in learner profile. The other is to adjust, extend and optimize the conceptual framework so that more learning routes that suit different learners could be described.

Next, we will extend this study. The supportive services will be refined in both scope and depth to effectively assist online learners to construct knowledge. The architecture of the agents, the communication between the agents, and the strategies for decision-making will be optimized to organize the supportive services in a more timely and robust manner. The systematic evaluation will be another work.