Abstract: The aim of the study is to develop an intelligent web-based system that allows integration, easy access and sharing of domain specific Knowledge-Based Systems (KBSs). Development of KBSs requires continuous adaptation and extension of already deployed knowledge bases. Often, these modifications consider the improvement of the Knowledge Base (KB). Furthermore, examination of the KBSs in its actual user environment could broad the KB. The proposed knowledge sharing environment allows different clients to access, maintain and upgrade domain specific knowledge bases. The proposed framework is generally applicable for KADS-based KBSs. As a matter of fact, we have applied it to agriculture diagnostic KBSs.
INTRODUCTION
The dynamic nature of knowledge has long been realised: knowledge evolves over time as experiences accumulate; it is revised and augmented in the light of deeper comprehension; entirely new bodies of knowledge are created while at the same time others pass into obsolescence. However, this dynamism has rarely been acknowledged in the engineering of knowledge-based systems: systems of static knowledge elements are assumed to exist in unchanging environments. The increasing desire to deliver knowledge across distributed environments such as the World-Wide Web highlights the extent of this gulf: the Web is ever-changing and systems must be able to accommodate change if they are to succeed and thrive. To be able to describe the dynamics of knowledge, there seems to be a need for some level of formality. The role of formality is twofold: to give a concise account of what is going on and to use this account for practical purposes in maintaining and analyzing the life cycles of knowledge components from their creation, through successive phases of modification, to their eventual retirement (Schorlemmer et al., 2002).
The presented research focuses on enhancing the process of knowledge-based systems construction by providing an integrated development, deployment and sharing environment that aims to provide up to date KBSs. This research proposes a knowledge sharing and collaboration environment with a prototypical application in the agriculture field. It gives a way to collaborate and share knowledge between web-users. The environment covers many aspects related to the content, the editing facilities and exchange of knowledge. It provides system users (client) with different views that enable each of them to access permitted knowledge according to user profiles. Two types of users are involved in this system, the first users type is domain experts that belongs to different agriculture disciplines who constitute the main source of knowledge. The other system client is the end users (extension workers) who use the KBSs to support him in making his daily decision in giving the appropriate advice to growers. Based on their daily life expertise, users may require to update the domain knowledge. As a result, a need for specific domain experts to review and confirm user modifications arises. The proposed knowledge sharing environment allows both end users and domain experts to apply her/his modifications in structured manner without needs of knowledge engineer. It is obviously difficult to bring together information from different knowledge sources about a subject of common interest. Therefore, the proposed environment aims to close the gap between feedback knowledge obtained from the usage of the KBSs in real world environment and the domain expert opinion. This is an important step in KBSs development because it allows monitoring the system in its actual user environment. Then, specific domain experts extend the KB in their area of expertise based on experiment of KBSs in real life environment.
Applying Multi-Agent Systems (MAS) (Wang et al., 1999) that provide a degree of flexibility (generic) to different application areas. Software agents are broadly defined as software entities capable of getting information from their environment, reacting to it, communicating with other agents and achieving a goal of performing some complex task on behalf of their users (Ustymenko, 2008). Effectiveness of the proposed knowledge sharing platform rely on its ability to automatically accumulate and refine end users modifications and transfer it to specialized domain experts to be reviewed. Knowledge exchange and sharing is necessary in this case to build an intelligent service that carried out by specific agent. The society of agents in our system consists of multi-agent diagnostic expert system, set of communication agents and global coordination agent. Collaborative communication between different agents is based on JADE platform. (http://jade.tilab.com).
Knowledge sharing framework such as the Rapid Knowledge Formation project (RKF) (Tecuci et al., 2004) has been developed. RKF aims at building power-tools for knowledge acquisition. The central objective of this study was to enable distributed teams of subject matter experts to enter and modify knowledge directly and easily, without the need of prior knowledge engineering experience. Disciple-RKF, the implementation of the most recent version of the disciple approach, is an agent shell with a knowledge base structured into an object ontology that describes the entities from an application domain and a set of task reduction and solution composition rules expressed with these objects. The development of the knowledge base of a disciple-RKF agent is based on importing ontological knowledge from existing knowledge repositories and on teaching the agent how to perform various tasks, in a way that resembles how subject matter experts teaches a human apprentice.
Furthermore, a conceptual framework for facilitating the modelling and development of heuristic reasoning services such as diagnosis, therapy planning and patient monitoring in clinical domains was developed by Taboada et al. (2002). The purpose of that framework was to provide support for (1) the representation of medical knowledge in a structured manner, (2) the design of clinical tasks, (3) simplifying the development of executable clinical reasoning services from their design and (4) making available its use on the web, with two main objectives: helping to increase the use of decision-aid systems in the clinical practice; facilitating the design of these systems by reusing information and knowledge sources distributed in the Web.
Bioinformatics Knowledge Management System (BKMS) has been developed by Abdullah et al. (2008) using multi-agent systems in order to facilitate knowledge sharing among bioinformatics communities as well as to demonstrate it into the system wise. The researchers provided how the agent technology could be utilized in the BKMS as a system model for serving the communities that is developed by using groupware such as Lotus Notes software. The achievement in conducting this framework of the BKMS is an added value for the any organization that need to implement the BKMS as a system, which could be helped the CoPs to work together in achieving their aims and mission statements.
An agent mediated community of a practice system using MAS-CommonKADS methodology was developed by Peyravi et al. (2007). Each single agent, even those in the same community, needs its own autonomy in order to model an organization and its individuals correctly using this approach, the natural model for an agent based on knowledge sharing system has been resulted. All models of MAS-CommonKADS methodology required for developing the multi-agent system was described. An implementation resulted design model was presented. Rebeca was used for modeling and verifying this system.
PROPOSED KNOWLEDGE SHARING FRAMEWORK
The Knowledge Sharing Framework (KSF) focuses on providing an efficient framework for accessing, maintaining and refinement generic knowledge bases.
Knowledge Sharing
Knowledge sharing involves the transfer or dissemination of knowledge
from one person or group to another. World wide web can be effectively
used for knowledge transfer between individual of the same domain. The
aim of the proposed Knowledge Sharing Framework (KSF) is to corporate
different knowledge sources in order to be able to build a complete Knowledge
Base (KB) that has at least as much information as each KB components.
This implies that knowledge is collective and able to evolve since it
is liable to be extended through the interactions of the various system
entities which store various parts of the available knowledge. This knowledge
sharing framework is generally applicable to different domain. As an example,
the system is applied in the agriculture domain. Expert systems are recognized
as an appropriate technology for knowledge sharing because they address
the problem of transferring knowledge and expertise from highly qualified
specialists to less knowledgeable personnel. Moreover, the development
of an agriculture expert system requires the combined efforts of specialists
from many fields of agriculture domain and must be developed with the
cooperation of the farmers and extension officers who will use them (Broner
et al., 1992). Knowledge transfer is always taking place from research
to extension, from extension to farmers and even from farmer to farmer
(Rafea and Shaalan, 1996). Different knowledge sources are considered
here. The first one (the core) is a centralized generic diagnostic knowledge
based-system. This generic model has the ability to derive diagnoses for
differently structured individual systems from the agriculture domain.
The generic diagnostic model (El-Korany et al., 2004) was developed
based on CommonKADS model of expertise (Schreiber et al., 2000)
that integrates the domain knowledge and the problem solving knowledge.
Domain knowledge contains all knowledge about diagnosis in the agriculture
domain, regardless of the crop. The problem solving process describes
the reasoning process. The second source of knowledge is the knowledge
acquired from end user (extension worker) based on their daily life expertise.
End users can extend the domain knowledge based on feedback from the usage
of the ESs in real world environment. Researchers have agreed that end
users are only allowed to specify certain kinds of modifications, i.e.,
domain-specific knowledge regarding concepts and domain relation but not
modifying problem-solving knowledge (Kim and Gil, 2000). The remaining
knowledge source is domain specific experts in different agriculture diagnostic
fields who review and confirm end user modifications. The role of the
domain experts is to decide to include end users modifications in the
centralized knowledge base.
Use of Multi-Agent to Support Community of Practice
According to Ferber (1999), a multi-agent system is a system composed
of a population of autonomous agents which cooperate with each other to
reach common objectives, while simultaneously each agent pursues individual
objectives. Multi-agent system can be viewed as a system in which agents
can communicate, exchange their individual knowledge and cooperate in
order to solve complex problems and to achieve collective or individual
goals. Multi agent system could be developed and used to facilitate and
support the community of practice (CoP) (Maier et al., 2005). In
this case, agent also could be categories in terms of its roles in collaborative
context like in knowledge communication, knowledge monitoring and others.
Three main agent categories are involved in the proposed system like:
| • | Service agent that to assist in local production activities, like ES agent |
| • | Monitoring agent that aims to monitor, collect and publish modified knowledge, like global coordination agent |
| • | Interaction agent to assist with cooperative work and to mediate communication among people, like user community agent and domain expert community agent |
A prototype for knowledge sharing environment was designed and implemented using JADE platform. JADF (Java Agent Development Framework) is a software framework fully implemented in Java language. It simplifies the implementation of multi-agent systems through a middle-ware that complies with the Foundation for Intelligent Physical Agents (FIPA) specificationsf. Collaborative communication between agents occurs in the form of request and reply messages. Messages between different system components are exchanged based on JADE platform. Agents provide two important capabilities in the combined system:
| • | Communication: Ability of sharing information among agents |
| • | Cooperation: The joint effort of different agents that decompose the system aiming at achieving one goal (Kone et al., 2000). This goal originally would go beyond their individual capabilities. Cooperation-agents are responsible for coordination the overall operation |
SYSTEM ARCHITECTURE
From the architectural point of view, the system provides a framework to coordinate the behavior of several specific agent types. The proposed architecture has the advantage to be fully generic, only the type of knowledge should be modified. The society of agents in this system consists of multi-agent diagnostic expert system, set of communication agents and global coordination agent. The diagnosis knowledge is distributed among several agents, each agent is an autonomous expert for a certain agriculture crop. The end user community agent and the domain expert community agent employ intelligent data collector working on the different client sides. Those agents communicate by passing messages to the global coordination agent.
Multi-Agent Diagnostic Expert System
Expert systems running on the Internet can support a large group of
users who communicate with the system over the network. The applied multi-agent
expert system constitutes a set of agents capable of diagnosing disorders
of different crops in the agriculture domain. Each expert system agents
provides the reasoning capabilities as well as the domain knowledge which
is required to carry out its tasks for a certain agriculture crop. A diagnosis
agent is responsible for producing diagnosis of disorders based on a set
of input observations.
All agent communication takes place through by passing messages to the Expert System Coordinator Agent (ESCA) (Fig. 1). The main functionality of this agent is to react to external request (normally from the global cooperation agent) and select the appropriate expert agent and assigning the request to it. A communication protocol that aims to ensure concurrent interactions between ES agents working on the resolution of a problem, together and interact with the environment is provided through ESCA.
| Fig. 1: | The overall structure of multi agents system |
Domain Expert Community Agent
Diagnostic KBS combines different specializations that are well defined
in the agricultural domain. The agricultural diagnosis domain has been
classified into six groups, namely: fungal, insect agent, nematode, snail,
mites and nutrition deficiency. This classification is done according
to the diagnosis taxonomy in the agriculture domain (Shaalan et al.,
2004). It has the following advantages that contribute to the suggested
framework. First, it tries to simulate what may happen in real world in
a very fine-grained manner. Second, partitioning of domain knowledge makes
it easy to maintain and refine the domain knowledge. Each of diagnosis
class is assigned to a specific domain expert (s) who is authorized to
review and confirm any user modifications. The main functions of the Domain
Expert Community Agent (DECA) are:
| • | Establish a community of experts specialized in different agriculture diagnostic fields |
| • | Receive the set of guilty rules from the global coordinator agent and distribute among domain experts according to their specialization |
| • | Retrieve the confirmed modifications from domain experts and transport it to global coordinator agent |
It is also responsible for authenticating the domain expert of the KSE and authorizing them to the appropriate resources. It manages these activities through a database of domain experts.
End Users Community Agent
The knowledge sharing framework allows users to register and deploy
their local KBS, modify it and receive updates from their colleagues in
the community based fashion. Therefore, the two main functionality of
End User Community Agent (EUCA) are:
| • | Track users modifications |
| • | Establish a community of practice |
Each of these functions is illustrated in the following subsections.
Track Users Modifications
Users need to access and interact with KBS from different perspectives
depending on the tasks they are completing. End users (extensions, highly
experienced growers) interact with the system to complete their sessions
by applying their own diagnostic cases and obtaining the system result.
When the system does not (does no longer) function as required, users
may require to update any of the domain specific knowledge. Since in real
life, the symptoms of disorders are likely to change, also may others
disorders need to be added/removed from the original KB. When the End
User Community Agent (EUCA) receives a user request to join the
community, it authenticates the user to the KSE, provide him with a modifiable
local KBSs, user refinement module and create log files for this user
to keep track of her/his modifications. When the end user applies any
knowledge modification, the refinement module provides him with a complete
guidance about the effect of her/his modifications on the domain knowledge
to avoid any knowledge inconsistency (Bahnasy and El-Korany, 2008). When
the user confirms her/his modification, the log file is activated. Regularly,
the end user community agent send approved user modifications to the global
cooperation agent. End user community agent manages all user activities
through data base. It is also responsible for applying the collaboration
mechanism between the global cooperative agent and end users.
Establish a Community of Practice
The shared knowledge building and experiences from a community of
practice could provide the technical advisor with the confidence to support
other academics with similar knowledge of practice (Molphy et al.,
2007). Knowledge update done by each individual user solely may be of
added value to other users in the community. Therefore, there is a need
to provide system users with the contribution of the updates applied by
different community members. End users community agent aims to constitute
a community of users with joint expertise to form a Community of Practice
(CoP). Since each individual user applies her/his modification on her/his
local KB, this modification is opaque to her/his colleague. Therefore,
the main functionality of End User Community Agent (EUCA) is to allow
other users in the community to share each knowledge update (of their
interest) and to decide whether to include it in their individual KB or
to neglect it. This is accomplished by first classifying users into a
set of group based on their interest. Then, knowledge modification is
supplied to group of users of the same interest. User interest is obtained
based on the customary modification type that is frequently done by each
user. As describe earlier, the agriculture disorder taxonomy is decomposed
into the following clusters namely: fungal, insect, nematode, snail, mites
and nutrition deficiency. End user profiling is done based on this taxonomy
i.e., depends on weight (heaviness) of users modifications applied on
a specific diagnostic class(es).
Global Cooperative Agent
Agent cooperation is a key issue in a multi-agent system (Lejouad-Chaari
and Mouria-Beji, 1998). The Global Cooperative Agent is (GCA) the core
of the presented environment. It provides the deployment and cooperation
services to its clients over the network. The process of knowledge monitoring,
exchange and integration is the main responsibility of the global cooperative
agent . Moreover, it is also responsible for publishing the modified versions
to corresponding users. This can be done through:
| • | Monitoring knowledge exchange |
| • | Update Generic KBSs |
Monitoring Knowledge Exchange
In the proposed architecture, the central KBSs have two types of users:
the end users and the domains experts, each working on his corresponding
side. While the end users apply their changes on their local KBs, domain
experts would refine these changes before applying it to the centralized
KBSs. When the end user community agent sends the accumulated users modification
to the GCA, it proceeds and executed the incremental refinement algorithms.
A set of guilty rules is then transmitted to domain expert community agent
which in turn distributes it among domain expert based on their specialization.
Moreover, when expert community agent responds and sends confirmation
of domain experts back to the GCA, it provides expert system coordination
agent with these modifications to be applied to original KBs and publish
the new version. Knowledge exchange is transferred between agents by applying
message-passing protocol. In multi-agent environments, messages should
be schedulable, as well as event driven. They can be sent in synchronous
or asynchronous modes. Furthermore, the transportation mechanism should
support unique addressing. Applying direct communication protocol, each
agent want to communicate must explicitly name the receipt or sender of
the communication (Kone et al., 2000). In this scheme the following
are required.
| • | Describe the prototypical scenarios between agents, the conversations are identified through the following primitives: |
|  | • | Send (agent, message) A message is sent to agent |
|  | • | Receive (agent, message) receive a message from agent |
| • | Represent the events (interchanged messages) between agents which reflect the relationships between the agents via their services |
The use of global ontology helps us to define the interchanged knowledge structures. These functionality are fully supported through using JADE platform.
Update Generic KBs
The knowledge engineering process involves (at least) three parties:
the intended end-user of the system, the domain expert who provides the
expertise that forms the basis of the system and the knowledge engineer,
who acquires the knowledge of the domain expert and uses it to construct
the system. The hope of knowledge engineering has long been to remove
the knowledge engineer from the adaptive loop and provide tools that enable
the domain expert to adjust the system to changing knowledge requirements.
This functionality is applied through the GCA that is responsible for
regularly gathering all user modifications kept by the end user community
agent. These modifications are stored in a global modifications database
to be used by the incremental refinement algorithm. When the domain experts
approve user modifications, it is transmitted to the original KB to be
integrated and published to end users again.
OVERALL PROCESS
The intended form of representation for the KB repository is XML structures. Due to the structural nature of XML and its high interchangeability, representing domain knowledge in XML would provide syntactical standardization well suited for domain knowledge exchange when there is an agreement on their intended semantics. The suggested knowledge sharing platform decomposed the following functionality:
| • | Local refinement |
| • | User profiling |
| • | Incremental refinement |
Each of these functionalities is descried in details in the following subsections.
Local Refinement
At the user site, the local refinement module handles user modifications
that affect the semantic of the knowledge base. For example, when a user
wants to add new rule (or any KB component) and as she/he is not familiar
with the classification criteria upon which the rule clustering taxonomy
was built. As a result, she may add this rule at incorrect rule cluster.
The objective of the local refinement module is to advice the user about
her/his modification request and its effect on knowledge semantic to avoid
any inconsistency. When any user submits her/his modification to her/his
local KB, the local log file is activated. The details of the local refinement
module is explained in by Bahnasy and El-Korany (2008).
User Profiling
After the user admits her/his modification to her/his local KB, this
modification is circulated to the community of users based on their profiling.
Thus, only users of common interest are notified about the newly added
knowledge. This helps users to save their time as they only get particular
modifications instead of receiving all knowledge update. Moreover, when
more than one user confesses the same modification on her/his local KB,
this provides an indication about the importance of this modification.
Accordingly, this ratio is transferred to the global cooperative agent
to be considered by incremental refinement module. User profiling is decomposed
from two main phases.
| • | User profile phase |
| • | Ranking procedure |
User Profile Phase
User profile phase is used to identify persons inside a community
with similar interests. Users in the same group are held together by a
common interest in a body of knowledge and are driven by a need to share
problems, experiences and best practices (Kim and Gill, 2000). Thus, in
the proposed environment, the process of user profiling is based on user
frequency of applying modifications to the same knowledge cluster. This
helps locate and cluster similar users to facilitate relevant knowledge
sharing. Moreover, it helps to find users who apply the same changes on
their local KB. User profile algorithm is shown in Fig.
2.
| Fig. 2: | User profiling algorithm |
| Fig. 3: | Ranking algorithm |
Ranking Procedure
A Ranking procedure is used to indicate the ratio of approved modifications
by the community of users. It represents the number of end users of the
same profile who apply the same modification to their local KBs. This
ratio represents the confidence of the applied modification and is used
by the incremental refinement techniques. In order to calculate this ratio,
a token that represents the knowledge modification is circulated to the
all users in the same profile. When a user selects to include this modification
in her/his local KB, the token is updated. When the number of users exceed
certain threshold (for example 3), this modification is transferred to
the Global Coordinative Agent (GCA) to be used by incremental refinement.
Figure 3 presents the ranking procedure used.
Incremental Refinement
The aim of the incremental refinement phase is to recommend to the
domain expert the possibly of including the collected user modification
to the original KBs. This is done based on two important factors: knowledge
similarity and ranking ratio. The former indicate the similarity measure
(Rong and Edward, 2007) of the updated knowledge w.r.t knowledge consistency.
While the later represents the ratio of approved modifications by more
than one user in the same community. The incremental refinement consists
of two important phases:
| • | Knowledge localization |
| • | knowledge inconsistency removal |
Knowledge Localization
Knowledge localization is the process of locating the updated knowledge
in the appropriate knowledge cluster. When a knowledge base component
is updated, the system suggested the suitable knowledge cluster that should
include this newly knowledge component based on the structure and semantic
similarities of knowledge. For example, newly diagnostic rules are obtained
from user in form of antecedence-consequence pairs. Antecedence represents
the symptoms (finding) that lead to certain disorder, while consequence
presents the disorder. The user can freely selects any rule cluster to
add in this new rule. The new rule must have at least one dominant finding
match with the specified rule cluster from the user. Otherwise, it has
no similarity and cannot be added to that rule cluster. The knowledge
localization phase is responsible measuring the similarity of any modified
rule (for example) to all rule clusters. The task of classifying a rule
to the nearest (most similar) rule clusters, is most likely to explain
on a firm theoretical foundation using Bayesian statistics. The key problem
here is to get the likelihood of the given description (findings) given
the classes. Applying Bayesian Classification theory (Hanson et al.,
1991) to get the most confidence knowledge cluster that matches each newly
added knowledge components. The formula used attempts to understand more
about the knowledge base component by considering its semantic similarity
w.r.t other knowledge clusters.
| • | P(c) prior belief (probability of rule cluster) |
| • | P(F|C) likelihood (probability of the finding match inside each cluster) |
| • | P(F) = Σ P (F|C) P(C) Finding evidence (marginal probability of the finding overall the classes |
| • | P(C|F) posterior belief (probability of class C to include finding F) |
The mathematical formulas say how likely any particular finding would be confidence to a certain cluster. Next, confidence of each finding to all cluster is calculated. The confidence values of each finding indicate whether it is dominant and subsidiary features w.r.t diagnosis class. Dominant finding are those prototypical (Rosch, 1975) for the diagnosis class membership, more importantly, while subsidiary finding are those overlaps between classes.
Knowledge Inconsistency Removal
At the domain expert site, each domain expert (according to his specialization)
receives a set of guilty rules. Those guilty rules are categorized with
two different beliefs. The first belief represents the ranking ratio (votes)
which is the actual number of users that select to apply this modification
in their local KB. The second is the confidence measure obtained from
the knowledge localization. Now, based on these values the domain expert
can decide whether to accept these modifications or not. Then the domain
expert community agent will send the accepted modification to the global
coordination agent to be transmitted and included in the centralized KBS.
CONCLUSION
The aim of this study is to facilitate the sharing, access, updating of generic diagnosis agriculture KBSs residing on server and allow both end users and domain experts to apply her/his modifications in structured manner without needs of knowledge engineer. The society of agents in this system consists of multi-agent diagnostic expert system, set of communication agents and global coordination agent. Unlike other knowledge sharing platform, the proposed knowledge sharing environment is capable of gathering and integrating feedback knowledge from end users. This feedback knowledge reflects the real life experience is included with the original KBs in order to enhance the process of constructing knowledge bases without a need for knowledge engineer. Moreover, the existence of specialized domain experts who confirm any modification before updating the centralized KB robust the construction process of domain specific KBSs.