Elimination by aspect and rule based decision making

Submitted by Sukaina Bharwani | published 25th Mar 2011 | last updated 13th Jan 2020
Please note: content is older than 5 years

Tools for Knowledge Elicitation (KnETs)
Sukaina Bharwani

The need to understand empirical data in a formal way is imperative when faced with multiple responses of humans to their environments. Techniques for classifying and formalising elicited knowledge can be used to enhance our understanding of our data and to reveal new avenues for enquiry. This method supports traditional participatory fieldwork methods and can also provide input for agent based models. Thus, it provides a formalised link between qualitative and quantitative representations of knowledge and their interaction.

Knowledge Elicitation Tools (KnETs) incorporate methods used in anthropological fieldwork combined with classical knowledge engineering techniques from computer science. The purpose of this is to alleviate the weaknesses inherent in both methods, to provide an overall robust and practical process, from knowledge elicitation to knowledge representation. The fusion of these techniques has resulted in a four-stage process which incorporates consistent verification and validation on data as it is collected. The application of this innovative methodology to this domain is successful precisely due to the mutual benefits that each technique provides by addressing current bottlenecks in both processes of ethnographic data collection and knowledge engineering processes.

Figure 1 Stages within the knowledge elicitation process.Figure 1 Stages within the knowledge elicitation process.

The methodology requires intensive interaction with stakeholders. After interviews with key domain experts, other experts/stakeholders are consulted to assess the representativeness of the knowledge gained from the chosen experts and their ability as key informants. Further knowledge is elicited from the main experts using an interactive game other data can be consulted as a consistency check of the data that is provided.

Possible relationships within the data that is produced from the game are revealed through the use of a machine learning algorithm. Analysis of the output from this algorithm allows the construction of decision trees. These prototypical rules are further validated in an iterative process, using an interactive learning program during further interviews with stakeholders, to refine and prune the decision trees. Validation can also be carried out during further interviews.

A combination of anthropological and ethnographic methodology to collect qualitative data and knowledge engineering techniques to formalise this data has been usefully employed in this suite of tools to better understand domain knowledge. This method can provide clarity to qualitative data collected during fieldwork, to reveal new avenues for enquiry and to broach the realm of tacit knowledge. All of this is achieved while remaining engaged with stakeholders and enabling their involvement in the entire process, from elicitation to validation.

Combination with WEAP

Knowledge elicitation tools (KnETs) can be used to develop matching methods. An explicit example is being developed to combine a KnETs-derived decision tree with a physical water allocation model built using the scenario-based Water Evaluation and Planning System (WEAP) software.

Future developments

KnETs will eventually be developed to link its analysis of micro level decision making with more macro level decisions using the elimination by aspects methodology. This has the potential for addressing some of the weaknesses of vulnerability mapping applications.

Implementation in decision-support tools

The elimination by aspects approach would need be driven by an expert system framework. To support a wide user base, and coupling/integration with other software it could be implemented in the Java programming language. A suitable environment for instance could be the Java Expert System Shell (Jess).