Interviewer's toolbox

Interviewer's toolbox

Computers ind. Engng Vol. 17, Nos 1-4, pp. 73-78, 1989 Printed in Great Britain. All rights reserved 0360-8352/89 $3.00+0.00 Copyright © 1989 Pergamo...

390KB Sizes 0 Downloads 58 Views

Computers ind. Engng Vol. 17, Nos 1-4, pp. 73-78, 1989 Printed in Great Britain. All rights reserved

0360-8352/89 $3.00+0.00 Copyright © 1989 Pergamon Press plc

ll%'11~WI~' S TOOLBOX

Janice M. Karasz Technical Department Cameron University Lawton, Oklahoma 73505

ABSTRACT

Knowledge elicitation is the life blood of the expert system. Qualifications for a knowledge engineer (KE) to elicit knowledge are, "You must be good with people and possess good interview skills." While this is true, there are some techniques the KE can develop to better meet these qualifications. Understanding the expert and the problem of expression gives insight into the person while standard interview techniques and cognitive interviewing techniques build interviewing skills. IWI~O[KK~ION

The knowledge base of the expert system depends upon the transference of knowledge and heuristics like a user depends on good software. Poor software may eventually get the job done but leaves the user somewhat frustrated and unsatfsfied. One may not know what good software is but bad software is easily identified. The performance of the expert system in terms of its usefullness, capabilities, and reliability rests with accurate and valid knowledge that is reliable. This performance is proportional to the success of the knowledge elicitation process, also known as the knowledge acquisition and transfer process. The elicitation process depends upon the interaction between the expert and the KE. Thus, the KE must be aware of these interactions and be capable of employing techniques to effectively elicit the knowledge needed. The interaction between the expert and the KE involves four areas: (i) (2) (3) (4)

Personality and communication skills of the expert Knowledge expression abilities of the expert Interaction awareness process The extraction techniques used by the KE.

THE EXPERT

The expert's personality is as varied as those found in everyday life. The personality that the KE must he concerned with is the one related to the knowledge elicitation process. Hopefully the expert will be articulate, cooperative, and enthusiastic. This expert is a KE's dream. However, if the expert is not of this type the KE needs to be able to identify the type and identify an approach to effectively communicate, thus aiding the elicitation process. Following are descriptions of several problem types of experts.

b~able to articulate. The expert may not be able to articulate the knowledge to another person. In this case the KE must obtain most of the knowledge through observation. The expert must teach the KE like he would teach any aspiring expert. However, since the KE has no previous experience in the domain area, the KE becomes an apprentice. Time may not permit the KE becoming a full fledged expert but he will become more proficient communicating with the expert. The inability of the expert to c(mlmunicatemaybe due to the situation that the expert may not have English as his first language or the expert may have a communication handicap.

Uncooperative. An expert may be uncooperative and refuse to share the knowledge be possesses. The KE must work to overcome this attitude problem by securing top management support and by establishing rapport with the expert. The KE may encounter from the expert much the same feelings that a systems analyst encounters from a user. Extra time invested in educating the expert as to the benefits of the expert system, in public relations, and in a genuine caring usually dissolves many of these feelings.

Evasive. The evasive expert is similar to the uncooperative expert but his actions are in a 73

74

Proceedings of the 1lth Annual Conference on Computers & Industrial Engineering

much more subtle manner. It may take a little time to identify this person for he may appear to be your dream expert. When things appear to be going great but you have little to show for your time, indicates that the expert is evasive. The KE may pose two or three specific questions and evaluate the responses to validate his suspicions. The KE needs to re-evaluate the relationship and apply the same techniques applied to the uncooperative expert. m--hling. The rambling expert has good intentions but wants to give so much information that it is hard to evaluate what is and what is not relevant. The KE needs to learn to direct the expert with such statements as: That's not important right now, we'll get to that later. I really want to know just about "x" right now. Here is a list. Please evaluate only what's on the list. The information that the expert is giving m a y b e important but at this time in the KE's education process it has little or no meaning. Therefore, the KE should leave the door open to discuss it further at a later date. The KE must be a guide to open doors and to decide how involved one becomes once inside the door. Expert 's expert. The "expert's" expert also has good intentions. However, be has worked in the field so long that he forgets that others do not possess his vocabulary of terms. Thus he uses words that have no meaning to the KE. The K E m u s t learn some jargon and the expert must lower his vocabulary level. ~[PRESSION

The expert must express the facts and heuristics in such a manner that the KE is able to translate the knowledge into some appropriate representation which the expert system inference engine understands. Commonly the knowledge representation is in the form of If-Then rules. However, the expert may have one or more problems expressing his knowledge and heuristics. Deeply embedded. The expert's knowledge and heuristics may be so deeply 6rabedded that the expert does not really understand what he knows or how he knows it. The decision process may be attributed to experience that something occurs without being able to explain how or why it does; it may be due to intuition or a gut level feeling; or the knowledge may not be defined enough to express it in the desired If-Then rules.

Faulty zepozting. The expert may not report an influential factor. For example, a car owner calls a mechanic telling him that his car won't start. The mechanic arrives and attempts to start the car by checking various parts of the engine. After 15 minutes of checking, the owner admits that the car was very low on gas and then it just quit running. The fact that the car was very low on gas is a very influential factor in this problem and solution. The solution could have been reached almost immediately if that information had been known. On the other hand, noninfluential factors may be reported. After 15 minutes the car owner could have said that the problem may he due to the washing and waxing that the car received the day before. If the mechanic is uneducated like a KE, this information would have to be considered until such a time it was proven irrelevant to the problem-solution. Special cases. Many experts want to make everything a special case. During the development of an armed robbery suspect identification expert system (AREST), one of the experts was asked if a particular combination of traits that matched a suspect with a witness description was sufficient to classify the suspect as a probable suspect (Badiru). The expert instead of stating a simple "yes" or "no" wanted to say "it depends" and then would go off on a tangent. That special case may be important to include or may occur so rarely that it is not worth including. The trick is to keep the expert on track. Special cases may be explored when the expert system nears its final developmental stages. (Yche]:° In many areas, knowledge may commonly be expressed in formulas, in graphics, or a combination of formulas and graphics. Since the expert is accustomed to communicating with others in his field in this manner, it will be difficult for the expert to change to If-Then rules.

Karasz: Interviewer's toolbox

75

In 1955 Luft and Ingham (Luft) introduced a model describing the interaction that takes place between two individuals or groups. The model is called the Johari Window and is graphically represented as a matrix relating to the two entities involved. Each entity has known and unknown areas. Thus the matrix has two rows and two columns. Each intersection is referred to as a quadrant and represents different characteristics of the relationship. Quadrant i, labeled open, is the information that a person knows about self or subject and is willing to share. Quadrant 2, blind, is the information that is unknown to a person but is known by the other person through observation. Quadrant 3, hidden, is the information that is known by a person about self but is not shared. Finally quadrant 4, unknown, is the information that is unknown to both persons. The interaction process is dynamic and thus is constantly changing. At the beginning of any interaction quadrant 1 is minimized, quadrants 2 and 3 will be larger in size, and quadrant 4 will vary in size depending upon how much the expert understands of his knowledge and heuristics (Figure i).

Known to self

Unknown to others

Unknown to self

HIDDEN

UNKNOWN

Figure 1 The KE's objective (Figure 2) is to increase the exchange of information between two people which is represented by quadrant i. The expert must be willing to share the knowledge and heuristics with the KE. This area will increase more as the expert trusts the KE and a good rapport has been established, thus decreasing quadrant 3 significantly. Quadrants 2 and 4 will decrease some but perhaps the changes will not be as significant as quadrants 1 and 3. As the KE acquires more knowledge about the subject and is able to present information back to the expert, quadrant 2 may decrease. If the expert accepts the observation as being accurate, that information may then be transferred to the domain represented by quadrant i. The exchange of knowledge and questions may naturally generate new knowledge. Therefore, less information is unknown, thus its representation in quadrant 4 becomes smaller. There are however unanswered questions and rejected or unshared observations that causes information to remain in quadrants 2 and 4.

Known to self Known to other Unknown to others

J

OPEN

J

HIDDEN

I

Unknown 1

BLIND UNKNOW~

Figure 2

SYST~C

D&'I~m~VIEM]~G ~ Q % q ~ S

The KE initiates the interview with an expert. The interview technique involves the actual interview process and guidelines for effective information collection. The interview process begins with making an appointment with the expert and identifying the purpose of the interview. The purpose of the interview should also provide the expert with a personal reason to appear at the appointed time. Other amlmenities such as where, when, who, and what will happen should also be communicated to the expert. These steps may seem trivial to a busy KE but the success of the knowledge tranference process depends upon the rapport established with the expert. Once the interview is scheduled, the KE develops a "game plan". Objectives for the interview are identified. Also the information the expert needs from the KE and the information the KE needs from the expert is identified. The information must then be exchanged in the most effective way. Many interviewers use questions as the means of exchanging information. However questions should he used in moderation. Also avoid trying to do everything in one interview. The KE may need to analyze the information and summarize and/or prototype the information as the KE understood it. Another interview will allow the expert the chance to validate the KE's understanding, see progress in the project, feel an integral part of the process, and perhaps present new ideas to the expert through the KE's eyes.

76

Proceedings ofthe llth Annual Con~renceon Computers & Industrial En~neefing

The KE directs the interview using several types of interaction techniques. are shown in Figure 3 (Rubin and Farley 108).

These

II~tTa~CTION 'rsu~IQOES Reponses

Example

Effect

Continue Closed Ended Question

Yes, I see... What is the distance from A to B? How would you describe... This is what I hear you say... This system will help you... What if the situation were... You dressed warmly because it's cold.

Keeps the expert talking Demands a specific answer

Open Ended Question Replication Informational Advise/Opinion Analysis

Allows the expert freedom to express Validation of the expert's words Provides information to expert KE attempts to expand to understanding Conclusions that the expert may not express

Figure 3

These responses are combined by the KE for the purpose of directing the interview and effectively obtaining the necessary knowledge.

(II~ITIVE T E I - R I ~ Cognition is defined in Webster's dictionary (Webster) as, "...i. the process of knowing or perceiving; perception. 2. anything that is known or perceived." The cognitive interview technique is ~mployed by the KE in order to perceive what the expert knows. This technique consists of four general methods of stimulating memory recall. The KE guides the expert to: i) Describe the decision making process 2) Be complete 3) Describe the decision making process in reverse 4) Change perspectives. Before the expert begins to describe the decision making process the expert should brainstorm. The expert writes words or topics just as they come to mind, not in any sequence of time or importance. This will give the KE a check list for future reference and also a vocabulary list to work from. The next step is to organize the brainstorming by category and sequence. This can be done graphically as a cognitive map. To begin the cognitive map, select a topic that is a catalyst for the other topics. This catalyst may be the goal, the solution reached after applying heuristics, or it may be a name for the knowledge base. Topics are placed in categories around the main catalyst. If more than one topic is listed in a category, the category is named and that topic category becomes a sub-catalyst to explore. Another way to develop a cognitive map is to ask the following questions: i. What comes to mind when someone asks, "Where is the professor?" write each topic down. 2. What made you think of office hours? What made you think of research time? Etc. 3. What connection do you see between location and office hours? Etc.

Question 1 is brainstorming. If this has been done you may choose to skip this question and move on to questions 2 and 3. Doing question 1 again at a later date may produce a more complete list than the previous one. The two lists can be compared with differences noted. Question 2 allows the KE to understand why the expert thought of each topic. Question 3 allows the KE to begin developing heuristics and an understanding of the interaction that exists between topics. Figure 4 is a cognitive map for locating a professor.

Karasz: Interviewer's toolbox

77

Professor's Location

J

I Class ii:30 FYNF 9:30 10:30 MWF ii:00 TTH

Research 1:00-3:00 M 1:30 3:00-5:00 MTW

Office Hours 8:30 MWF 8:00 TTH 1:00-3:00 TTH 3:00-5:00 TH

i

Not Available < 8:00 Daily > 2:00 F > 5:00 Daily

Lunch 12:30 MWF 12:00 TTH

Figure 4

The KE must have some understanding of the domain of the expert. He needs to feel comfortable with the jargon and the relationships that exist. Once this general education is accomplished, the expert can begin method 1 of the cognitive interviewing process. Method 1 requests that the expert construct the decision making process. This means identifying a goal, sequencing related categories, and stating the heuristics to reach the goal. This decision making process can be drawn in a tree diagram or may be presented in a matrix form to a commercial expert system that employs induction techniques to generate the rules (Thompson). The expert needs to be as complete as possible in describing the decision making process. If the prototype is to be built for test purposes, this will mean only that: all the information must be included that is necessary to reach an acceptable conclusion or goal. As the system progresses through its developmental stages the knowledge base will become more and more complete. As a check of the decision making process, have the expert start with the first topic in the sequence of topics. The expert should be able to link one topic £o another in the same sequence as previously described thus reaching the same goal. If the sequence varies then further exploration of the decision making process is needed. Finally, the expert needs to change perspectives. There is an old saying, "You can't see the forest for the trees." If the solution to a problem is conceived in only one manner, the solution may work but may not be the best or most efficient. Have the expert look for other solutions or relationships. The KE can pose questions or inject his knowledge to help the expert step to another position to view the problem-solution.

(X]MC~SION. Knowledge elicitation is too important to be left to chance. It has the power to make or break an expert system. Therefore the KE needs more direction than, "You need to be good with people and have good interview skills." While these things are vital, one wonders how do I develop these skills. Understanding the expert's style and expression limitations provides insight into the person, while interview techniques provide the tools to accomplish the knowledge elicitation process.

78

Proceedings of the 1lth Annual Conference on Computers & Industrial Engineering R l g - ~

Badiru, Adedeji B., Mathis, Janice and Holloway, Robert T. (November 1987), "AREST to Aid Crime Investigations," AI Interactions, Vol, 3, pp.5-6. Geiselman, R. Edward and Nielsen, Michael, "Cognitive Memory Retrieval Techniques," The Police Chief, March, 1986, p. 69-71. Geiselman, R. Edward and Fisher, Ronald P., "Interviewing Victims and Witnesses of Crime, "National Institute of Justice: Research in Brief, December, 1985. Heimlich, Joan E. and Pittelman, Susan D., Semantic Mapping: Classroom Applications, An IRA Service Bulletin, International Reading Association, 1986. Luft, Joseph, Group Processes: An Introduction to Group Dynamics, National Press Books, Palo Alto, Ca., 1970, p. 11-20. Meyer, David E., "On the Representation and Retrieval of Stored Semantic Information, "Cognitive Psychology, Vol i, No. 3, Aug, 1970, p. 242-300. Rubin, Stanford E. and Farley, Roy C., Intake Interview Skills for Rehabilitation Counselors, Arkansas Rehabilitation Research and Training Center, 1980. Thompson, Beverly and Thompson, William, "Finding Rules in Data," BYTE, Vol Nov, 1988, p. 149-158. Williamson, Mickey, "Breaking the Expert Systems Bottleneck," Computers in Science, Nov/Dec, 1987, p. 62-65.