Wai Keung Pun, Craig McDonald
& John Weckert
Knowledge Management Group
School of Information Studies
Charles Sturt University
Locked Bag 675, Wagga Wagga, NSW 2678 Australia
{dpun | cmcdonald | jweckert}@csu.edu.au
http://infstud.riv.csu.edu.au
Traditionally expert systems have been built from knowledge elicited from
domain experts. However, knowledge in applied science domains is grounded
in published sources like research reports, text books, articles and the
like. This corpus of knowledge is typically inconsistent, dated, dispersed,
and so on. The project described in this paper aims to construct a putative
Knowledge Management System. The core of the system is a knowledge server
that represents each publication and expert as a separate knowledge base,
and a meta-knowledge base to allow different kinds of access to the server.
Different client systems can be connected to the knowledge server to meet
different user needs, such as forecasting, advice, explanation, education,
and training. The server can also be a resource for researchers and research
managers, by allowing hypothesis testing and reviews of the literature.
Knowledge re-engineering is not necessary, as the system simply embodies
what is in the domain. The knowledge is being represented in conceptual
graphs and the test domain is irrigation. The work is being supported by
the Cooperative Research Center for Viticulture.
Before the knowledge created by applied science research can form a normal part of industry practice, it must be published, presented at conferences and seminars, built into training and education courses, and slowly 'percolate' through the community. This process can take considerable time, and much detail is lost or misinterpreted along the way. The Cooperative Research Centre for Viticulture (CRCV) in Australia is investigating methods of building applied research results into a knowledge-based system as a matter of course so that new knowledge can be put to use in the grape growing industry. Such a system would provide a vehicle for quick and complete promulgation of research results. We envisage a future where knowledge created in the laboratory and in the field can be reported to a knowledge-based system and become immediately effective in viticultural practice.
The project described here aims to find ways of representing applied research papers and reports directly in a knowledge management system (KMS), and of establishing the "meta-knowledge" necessary to properly mobilise the knowledge embedded in the literature. Such a system will enable multiple kinds of access to the knowledge, by decision support systems or computer-aided education systems for example, which will use the knowledge in different ways, for advice, forecasting, education and training, explanation and so on. It will also be a resource for researchers in hypothesis testing and research management. A prototype KMS is being built in the irrigation of grapevines as a means of evaluating the KMS approach.
Human knowledge takes two forms: private and public (Kemp, 1976). Private knowledge is that held in and used by the minds of humans. In its public form, knowledge is published as periodical articles, research papers, conference proceedings, technical reports, textbooks and so on. The applied sciences create public knowledge through research and publication, but current methods of organising and mobilising this knowledge are flawed. Considered as a whole, the applied science literature is:
Clearly, there is a large knowledge management problem to be addressed here. Current approaches to the problem come from either information management technology (document indexing and bibliographic databases which store and deliver papers) or expert systems technology (advice giving systems built from consensus knowledge of domain experts). The former requires a person to interpret the information delivered while the latter is often pervaded by imprecision and/or uncertainty (Grabot 96).
The research project described here aims to employ knowledge based technology to deal more effectively with the knowledge management problem. The KMS under development will collect and consolidate knowledge in a form that is explicit and accessible, while still preserving the context of each research publication. By avoiding some of the problems in current knowledge management, the KMS will be a powerful tool for technology transfer, allowing complete, unbiased and justifiable responses to industry problems and for research management. In the future, research results will be input to the KMS as though they were data. Of a parallel domain, forest science, McRoberts et al. (1991) say:
Computerized database management systems have been accepted
as essential aides to the human mind for decades now. No one
would dream of trying to manage a large forest inventory on paper
or in the minds of humans any more. Computerized knowledge base
management systems are making it equally wasteful to manage forest
science knowledge in paper journals and books, or in the minds of
human scientists. The volume is too large and, thanks to the
advances in AI, the computer can now cheaply store and retrieve
knowledge as easily as it can store and retrieve data. (p20)
The KMS will incorporate and integrate new knowledge that is being created in applied research projects around the world.
A prototype KMS is being constructed with two components. The first is a set of knowledge bases each representing the knowledge in a particular research paper or report. In the KMS, each publication is treated like a small single and independent knowledge base. The second component is a meta-knowledge base that represents aspects of each research publication. These aspects influence the selection of which knowledge base is applicable in a particular instance. The KMS will be used by a range of interface systems that will employ it in different ways. For example, a decision support system will use the KMS as a model of a domain to allow scenario processing. An expert system will give advice using the KMS as a knowledge base and justify the advice on the basis of the publications from which the KMS has been built. A Computer Aided Instruction (CAI) interface would allow the KMS to form the basis of courses in the domain. Researchers and research bodies can use the KMS as a source for literature reviews and hypothesis testing. Each of these interface systems will have specific systems components suitable to their purposes but will rely on the KMS as the source for their domain knowledge. As each new research report becomes available it is represented as a new document-related knowledge base and so participates immediately in the various uses to which the system is being put. Figure 1 shows the KMS architecture.
Figure 1: Knowledge Management Systems Architecture |
The research involved in the construction of the KMS centers on the development of methods for knowledge extraction from literature, knowledge representation in conceptual graphs (Sowa, 1984), knowledge query, and access to KMS by the interface systems mentioned above.
The Cooperative Research Centre for Viticulture (CRCV) in Australia carries out basic and applied science research on grape vines and their management. As part of its technology transfer program, the CRCV has developed an expert system, AusVit. The system provides advice to vineyard managers and grape growers about pest and disease risk in their vineyards and what appropriate action might be taken. The system also advises on irrigation, chemical use, and the like. The advice is based on vineyard profile, data from weather stations and user input from vineyard monitoring, all of which is interpreted by a series of disease simulators and a rule-based expert system. A chemical database provides details of the active components in agricultural chemical products, their application and registration information. The components of the system are shown in Figure 2.
Figure 2: The Inputs and Components of AusVit |
The rule base has been built using the traditional expert systems approach (Travis, 1992). The CRCV is interested in transforming AusVit from a traditional expert system to a KMS. An aim of the CRCV is to ensure that the results of its commissioned applied viticulture research are transferred to industry, and it sees the KMS as a vehicle to facilitate that transfer. A pilot study of building a knowledge base from the literature was conducted in the Botrytis Cinerea module of AusVit (McDonald & Ellison, 1994) and over the next two years the expert rule bases and simulations in one module of AusVit will be replaced by a KMS.
Irrigation plays an important part in viticulture. It is a powerful technique for improving vine performance, because it allows an environmental factor (water) to be placed under managerial control. Grapes are grown in Australia in regions with annual rainfall as low as 250mm to as high as 1100mm. In those regions with low rainfall, irrigation is necessary, because without it vineyards would be uneconomic due to water stress. Research into irrigation for grapevines has therefore increased significantly during recent years. To show how a KMS for vineyard irrigation might work, the following section gives an example of literature being used as a source for a set of Conceptual Graph (CG) knowledge bases and Section 5.2 shows how these CGs might be used in different ways to meet different user needs.
As an example of developing conceptual graph knowledge bases from published literature, three papers, Goodwin (1995), McCarthy et al. (1993), and Williams and Grimes (1987), have been selected and CG representation of their content created. These CGs are intended as examples only.
Evaporation (ES): Water stored in the soil is lost by evaporation from
the soil surface. The extent to which evaporation from the soil
surface contributes to evapotranspiration depends on the frequency of
wetting of soil, the area of soil surface wetted, and the proportion
of the wetted soil surface that is shaded. .......
Air temperature, humidity, and wind speed at ground level also affect
evaporation from the soil surface. .......
Transpiration (EF): Loss of water from vine foliage. Water vapour in
the air spaces within leaves diffuses to the outside air through
numerous valve-like pores (stomata) on the surface of the leaves. ....
Evapotranspiration: As both processes involve the use of radiant
energy they are collectively called evapotranspiration (ET = ES + EF). .......
If all pores in the soil are filled water and no air then soil is
saturated and is at the Drained Upper Limit (DUL). .......
Eventually a level of soil water is reached when plants can no longer
extract enough water and they begin to wilt. When plants wilt by day
and fail to recover at night, the soil is at the Lower Limit (LL).
.......
Plant Available Water (PAW) is the amount of water held in the soil
between DUL and LL and is the water that can be used by the plant.
It can be expressed as millimetres of water per metre of soil
(mm water / m soil). The amount of available water that a soil
profile can store depends on its texture ranging from 33 to 208 mm
per metre. .......
Actual vineyard water use (soil evaporation + plant use) is reported
to be as low as about 250 mm to more than 800 mm. .......
[TRANSPIRATION] -
(OBJ)->[WATER: #]->(STORE)->[PLANT: #]
(LOSS)->[LEAF_SURFACE]->(ATTR)->[AREA: #]
[EVAPORATION] -
(OBJ)->[WATER_CONTENT_OF_THE_SOIL: 1/4 DUL]<-(OBJ)<-[STORE]->
(AGNT)->[SOIL_TEXTURE: Sandy Loam = *x]
(LOSS)->[SOIL_SURFACE]->(ATTR)->[AREA: #]
[EVAPOTRANSPIRATION] -
(LINK)->[EVAPORATION]->(OBJ)->[RADIANT_ENERGY]->(CHRC)->[USE: #]
(LINK)->[TRANSPIRATION]->(OBJ)->[RADIANT_ENERGY]->(CHRC)->[USE: #]
[WATER_CONTENT_OF_THE_SOIL] -
(CONTENT)->[DRAINED_UPPER_LIMIT]
(CONTENT)->[LOWER_LIMIT]
[SOIL_TEXTURE: SAND] -
(ATTR)->[SOIL_PROFILE: Moderately Coarse]
(ATTR)->[TEXTURE_RANGE]->(MEASURE)->[MEASURE: 104mm]
[DRAINED_UPPER_LIMIT] -
(OBJ)->[WATER_CONTENT_OF_THE_SOIL]->(MEASURE)->[MEASURE: 54mm]
[LOWER_LIMIT] -
(OBJ)->[WATER_CONTENT_OF_THE_SOIL]->(MEASURE)->[MEASURE: 31mm]
[PLANT_AVAILABLE_WATER] -
(OBJ)->[SOIL_TEXTURE: *x]<-(MEASURE)->[MEASURE: 1metre]
[ACTUAL_VINEYARD_WATER_USE] -
(RSLT)<-[NUMBER: #]<-[SOIL_EVAPORATION]<-(ARG)<-[ADD]->
(ARG)->[PLANT_USE]->(RSLT)->[NUMBER: #]
vineyard irrigation is best defined as the efficient application of
water to maximise profit and minimise environmental degradation. .....
The aim of an irrigation is to replace the water used by the vineyard
since the previous irrigation. The timing and the amount of
irrigation will depend on the rate of water use and the quantity of
available water held in the root zone. Knowledge of vineyard water
use is therefore a critical component of irrigation scheduling. .....
Water stress is a physiological reaction of a vine to a limitation in
supply of water. Some of the physiological responses of grapevines
include: closing of leaf stomata, reduced photosynthesis, reduced
cell division and loss of cell expansion. .......
The vines must use up the total storage of available moisture from
rainfall before an irrigation is necessary. When to start irrigating
is therefore a function of how much water is stored in the soil and
the daily rate of water use by the vineyard. .......
When to start irrigating (days from bud burst) =
Soil water storage (litres) / Daily vineyard water use (litres/vine/day) .......
[IRRIGATION_AIM] -
[EVENT:
[REPLACE]->(OBJ)->[WATER_USE]->(LOC)->[VINEYARD: #]] -
(SUCCESSOR)->[EVENT:
(PAST)->[PROPOSITION]: [VINEYARD: #]<-(LOC)<-[IRRIGATE]->(OBJ)->[WATER_USE]]
[WHEN_TO_START_IRRIGATING] -
(RSLT)->[DIVIDE] -
(RSLT)->[STORE] -
(AGNT)->[SOIL: #]
(OBJ)->[WATER: #]
(MEASURE)->[MEASURE: # Litres]
(RSLT)->[WATER_USE] -
(LOC)->[VINEYARD: #]
(OBJ)->[VINE: #]
(FREQUENCY)->[FREQUENCY: Daily]
(MEASURE)->[MEASURE: # Litres]
[IRRIGATION] -
(FREQUENCY)->[TIMING: #]
(QTY)->[AMOUNT: #] -
[EVENT:-
[DEPEND]->(OBJ)->[WATER_USE] -
(MEASURE)->[RATE: #]
(STATE)->[AVAILABLE]
(SOURCE)->[PLANT: #]->(LOC)->[ROOT]]
(NECESSARY)->[PROPOSITION: [WATER_USE] -
(LOC)->[VINEYARD: #]
(OBJ)->[KNOWLEDGE]
(SUPPORT)->[SCHEDULING]->(INST)->[IRRIGATION]]
An important aspect of this study was to establish irrigation regimes
that reflected best estimates of vineyard potential evapotranspiration
and then apply water equivalent to that ET. .......
It was interesting to note that the relationship between applied water
and growing degree days (GDDs) was linear. This would indicate that
ET and vine growth were temperature dependent. .......
A constant level of soil moisture did not occur for the 0.4 ET
treatment, however, until 1000 GDDs after budbreak at Kearney
Agricultural Centre. The level of soil moisture that resulted from
the 0.4 ET treatment throughout the growing season were sufficient
to induce a water shortage for vines in this treatment. .......
The plant based measurements of vine water status indicated that
grapevines receiving less than 1.0 ET in this study were under
stress. These measurements have been used by many as a measure of
the degree of stress experienced by the vine during a period of water
deficits (Smart 1974, Hardie and Considine 1976, Kliewer, Freeman and
Hossom 1983, Liu et al. 1978). .......
[VINE_WATER_STATUS] -
(MEASURE)->[MEASUREMENT]->(METHOD)->[METHOD: PLANT BASED]
(INDICATE)->[STATE:
[VINE: GRAPEVINES = *x]->(RECEIVE)->[ET]->(LESS)->[NUMBER: 1.0]]
->(CAUSE)->[STATE: [VINE: *x]->(ATTR)->[UNDER_STRESS]]
[MANY]->(QTY)->[PERSON: #]->(INST)->
[EVENT:
[MEASUREMENT] -
(METHOD)->[METHOD: Plant Based]
(OBJ)->[VINE: #]
(CHRC)->[THE_DEGREE_OF_STRESS]
(POINT-IN-TIME)->[DURING_A_PERIOD_OF_WATER_DEFICITS]]
(NEG)->[PROPOSITION:
[OCCUR]-
(TREATMENT)->[ET]->(QTY)->[NUMBER: 0.4]
(OBJ)-> [SOIL_MOISTURE]->(MEASURE)->[CONSTANT_LEVEL]]
The KMS is designed to be used in different ways for meeting different users needs. Here we show how the KMS can be used as an expert system to simulate the problem-solving behaviour of an expert, can be employed as a decision support system to allow scenario processing, and can be also used as a computer aided instruction system for education and training.
In order to remain competitive, grape growers and vineyard managers often depend on agricultural advisers and experts to provide advice for their decision making. This advice is costly and expert assistance is not always available when the grape growers and vineyard managers need it. In this situation, the KMS can be used as an expert system. For example, a dialogue between KMS and user might be as follows:
User : "Given the current state of the vineyard, should I irrigate?"
KMS : "Yes. Irrigate to field capacity."
User : "How did you come to that advice?"
KMS : "Your ET is less than .01 that implies water stress (Williams
and Grimes, 1987). Your soil type is sand so you should
irrigate to 31mm, the Drained Upper Limit
(McCarthy et al. 1993). ..... "
The KMS would be using the CGs both to come to a decision and to justify that decision specifically on the basis of the literature. As an expert system, the KMS combines CGs from many sources including the experiential knowledge and intuitive reasoning of many experts. The KMS would use a meta-knowledge base to select suitable knowledge bases for the response.
The grape growers or the vineyard managers could test scenarios through the decision support system (DSS) interface to the KMS, for example:
User : "If the weather is hot over the next week, will I need to
irrigate?"
KMS : "Current water_content_of _soil is 40%. Hot weather implies
high radiant energy and high evapotranspiration
(McCarthy et al. 1993).
Expected stress_level in one week is ?? (Goodwin,1995)."
This kind of 'what-if' processing reasons with the CG's by setting up the conditions that would apply in the scenario nominated by the user. Another possibility is the use of the knowledge bases to determine what scenarios would be necessary for a specified outcome to occur (Richards & McDonald, 1995). For example:
User : "Under what conditions will I need to irrigate next week?"
KMS : "Current water_content_of _soil is 40%. If there is Hot
weather then there is high radiant energy and high
evapotranspiration (McCarthy et al. 1993).
Expected stress_level in one week is ??
If there is no rain you will need to irrigate (Goodwin, 1995)."
The KMS can also be employed as a computer aided instruction (CAI) system to support education and training. For example, students enrolled in the irrigation module of a viticulture course might be presented with text followed by questions that they answer online:
KMS : "Vineyare irrigation is best defined as the efficient
application of water to maximise profit and minimise
environmental degradation. .......
The aim of an irrigation is to replace the water used by the
vineyard since the previous irrigation. The timing and the
amount of irrigation will depend on the rate of water use and
the quantity of available water held in the root zone.
Knowledge of vineyard water use is therefore a critical
component of irrigation scheduling. .......
Water stress is a physiological reaction of a vine to a
limitation in supply of water.
Question : List the factors that remove water from the soil."
User : "vine transpiration, evaporation, drainage"
Their answers could be verified by reference to the CGs and student learning enhanced by other systems facilities such as information retrieval or simulations based on the CGs.
The case study described above raises many interesting issues concerned as much with the nature of applied science itself as with the technicalities of constructing a CG based KMS. However, the system has the potential to become an effective vehicle for technology transfer and knowledge management. It will have:
It is known that conceptual graphs are suitable for representing and processing knowledge due to their strong expressive ability and well-defined operations. Other reasons for using conceptual graphs, which are adopted as a knowledge representation language in this project, are that they allow advanced explanation, knowledge transportation and knowledge re-use and have potential to subsume a range of other forms of representation. AusVit is a part of a growing trend to manage scientific knowledge using computer systems. Information technology has an extraordinary rate of change and its ability to deal with highly complex and voluminous data is increasing rapidly. It is already the primary vehicle for recording information and it will become the primary vehicle for mobilising knowledge. Systems builders of the future will have to come to grips with the issues of knowledge management rather than knowledge engineering.
Davis, M. & Compton, P. (1991). Hypothesis Testing: A New Information
Retrieval Task for Libraries. In Libraries and Expert Systems, eds.
Craig McDonald & John Weckert. Taylor Graham: United Kingdom, pp. 142-148.
Goodwin, I. (1995). Irrigation of Vineyards: A Winegrape Grower's Guide
to Irrigation Scheduling and Regulated Deficit Irrigation. Institute
of Sustainable Irrigated Agriculture: Tatura, Vic., Australia.
Grabot, B. & Caillaud, E. (1996). Imprecise Knowledge in Expert Systems:
A Simple Shell. Expert Systems With Applications, 10, No. 1, pp.
99-112.
Kemp, D. A., (1976). The Nature of Knowledge: An Introduction for Librarians.
Clive Bingley: London & Linnet: Connecticut, USA.
McCarthy, M. G., Jones, L. D. & Due, G. (1993). Irrigation - Principles
and Practices. In Viticulture, 2, eds. B. G. Coombe & P. R.
Dry. Australian Industrial Pub: Adelaide, Australia, pp. 104-127.
McDonald, C. & Ellison, P. (1994). Knowledge Acquisition from Domain
Literature. In AI and Natural Resource Management: A Workshop in
Conjunction with AI'94, University of New England, 22 November 1994, Armidale,
Australia, pp. 83-91.
McRoberts, R. E., Schmoldt, D. L. and Rauscher, H. M. (1991). Enhancing
the Scientific Process with Artificial Intelligence: Forest Science Applications.
AI Applications, 5, No. 2.
Mühlemann, S. 1995. Das verschüttete Kapital. Spezial/Wissensmanagement,
BILANZ, 1:76-80.
Rauscher, H. M. & Dale, W. J. (1993). Managing the Global Climate Change
Knowledge Base. AI Applications, 7, No. 4.
Richards, D. & McDonald, C. (1995). Bridging the Expert System - Decision
Support System Gap: Adapting and Re-using an Existing Expert System. In
Proceedings of Sixth Australian Conference on Information Systems,
School of Information Systems, Curtain University of Technology, Perth,
Western Australia, p219-229.
Sowa, J. F. (1984). Conceptual Structures: Information Processing in
Mind and Machine. Addison-Wesley: Reading, Massachusetts.
Travis, J. W. et al. (1992). A Working Description of the Penn State Apple
Orchard Consultant, An Expert System. Plant Disease, 76, No. 6.
Williams, L. E. & Grimes, D. W. (1987). Modelling Vine Growth - Development
of a Data Set for a Water Balance Subroutine. In Proceedings of the
Sixth Australian Wine Industry Technical Conference, Adelaide, South
Australia, ed. T. Lee. Australian Industrial Publishers: Adelaide, Australia,
pp. 169- 174.