Knowledge Elicitation (KE)

Knowledge Elicitation (KE)
February 15th, 2007
Knowledge elicitation is the process of eliciting tacit knowledge, i.e. bringing out the knowledge present in the conscious and sub-conscious mind or facilitating the expert in recalling and redefining their rules of thumb, work practices, processes, etc. with the help of a knowledge engineer.
What is Tacit Knowledge and Why Elicit Knowledge?
Tacit knowledge is the knowledge that exists in the mind of people; it cannot be easily verbalized, formalized or expressed into words. It is a mixture of deliberations, subjective insight, intuitions, rules of thumb and judgment that is more likely to be personal and acquired by experience and expertise.
It often takes a long time in years to gain expertise in any particular domain so the tacit knowledge possessed by domain experts must be elicited, captured and made available for current use and preserved for future use.
Knowledge Elicitation Techniques
The KE techniques can be broadly classified into two main categories:
Direct or Indirect elicitation methods Interaction with the domain experts
Direct Elicitation Methods
Direct elicitation methods such as storytelling, case-study, interviewing and question-answer rounds provide required information directly from domain experts. Most of the time knowledge engineer knows what knowledge will be elicited during the KE sessions.
Indirect Elicitation Methods
Required information is not directly obtained from domain experts but the knowledge engineers need to analyze the results of KE sessions to elicit domain knowledge. Some of the indirect elicitation methods include observation, workspace analysis and work pattern analysis. Some times indirect elicitation methods provide additional information due to following reasons:
The indirect KE sessions led to thorough examination of the problem that the knowledge engineer would not have anticipated and would not have asked or explored during direct interaction with the domain experts. Some topics are not as verbal for direct questions as compared to full and detailed questions. Implicit knowledge ignored during the direct KE sessions.
Interaction with the domain experts
KE techniques can be grouped by the type of interaction with the domain experts such as interview methods, Case Study and Simulation.
Some Popular KE Techniques
INTERVIEW METHODS: These methods include structured, semi-structured and unstructured interviews, concept mapping, knowledge mapping, problem discussion, questionnaire surveys, ER diagrams and data flow analysis. CASE STUDY ANALYSIS: Problems in the domain area are identified and discussed via the case study analysis method. CRITIQUING: Results of previous KE sessions or KE sessions with various domain experts are validated by this method. ROLE PLAYING AND SIMULATION: The domain expert is given problems for solving by either creating live or simulated scenarios. These methods help identifying implicit knowledge. OBSERVATION, WORK SPACE ANALYSIS and WORK PATTERN ANALYSIS: Knowledge engineer observes the expert(s) performing tasks. The observation could be a simple on-site observation or work space analysis where 100’s of photographs of the expert’s workspace is taken and analyzed to determine how the experts performs his tasks. In Work pattern analysis method the patterns of activities in the work space are analyzed by the knowledge engineer. K-AUDIT: Please refer to previous posts on k-audits dated06/27/2006 and08/02/2006. KNOWLEDGE MAPPING: Can be used on the fly while interviewing experts to organize flow of thoughts and ideas with words, images, colors, numbers and relationships. To read more about knowledge mapping please read “Knowledge Mapping in Legal Research and Litigation” byPooja Songar.
Knowledge elicitation was initially termed as knowledge extraction during late 1980’s when the knowledge engineers and programmers were facing challenge in developing expert systems. Knowledge engineers and programmers were given credit just for codifying the elicited knowledge and not the time consuming process of interaction with experts. Knowledge extraction requires several interactive sessions with domain experts whereas coding is easy and relatively quick.
Robert R. Hoffman (AI Magazine, 1987) suggested following techniques for proper knowledge extraction:
BOOTSTRAPPING: Knowledge engineer must have basic domain knowledge prior interacting with the expert so that the expert does not have to define basic terms used in his daily routines METHOD OF FAMILIAR TASKS: Perform on-site observations and analyze what makes people take certain decisions for certain situations or problems. (UN)STRUCTURED INTERVIEW SESSIONS: Perform various structured, semi-structured and unstructured interview sessions with experts. ROLE PLAYING WITH CONSTRAINTS: Give problems to experts with constraints and limited information. METHOD OF TOUGH CASES: Experts gain expertise by working and observing several different levels of cases - easy, moderate and tough during their career. Some people gain expertise very early in their careers as compared to others because they have encountered tough cases very frequently. Priority must be given to tough cases during KE process as high quality and quantity information will be extracted from tough cases.
Important Points to Note
Interview sessions may be one-to-one, one-to-many and many-to-many. KE can also be automated when the knowledge possessed by human beings is buried in documents or some other media. KE becomes challenging when there is one single expert in the domain or no expert in the domain. It is also very profitable when a single or very few experts exists in a domain as eliciting their knowledge and making it available in a system can make their expertise available at several locations at the same time. Validating the results or information obtained from the experts may be an issue. KE is a time consuming process and requires significant expert’s time during KE sessions such as interviews, case studies, etc. Trying to get schedule appointments with domain experts can be an issue. Indirect KE techniques may provide additional information but has an overhead for analyzing the KE sessions data to get the required information. Concept mapping is used in most of the expert systems in use today as it easily presents the elicited knowledge, easy for understanding and modifying the knowledge in future. KE can help in developing Knowledge Based Systems (KBS) in several different fields whether it is medicine, weather forecasting, law firm, manufacturing, civil etc. KE is a time consuming and expensive process and should only be considered in domains where knowledge does not get outdated very frequently.
Other Resources to Check
Knowledge Elicitation Lecture by Robert R. Hoffman given at Florida State University on 3/17/04. This lecture can be viewed in Real Player or Windows Media Player and can be downloaded fromhttp://www.lsi.fsu.edu/fase/course/knowledgeelicitation.html