Alla Petrova

Jeff Wartes

Trish Winters

Abstract *

Program Description *

Calculating Results *

Functionality *

Design *

Exercises *

Alternative Approaches *

Limitations *

Problems *

Future Work *

References: *

The Conflict Resolution Assistance system (CRA) is a diagnostic part of a larger system that hasn’t been developed yet. The purpose of CRA is to identify one or more conflict resolution skills that the user may need to develop or improve in order to resolve the conflict successfully.

The system will then give feedback to the user and direct her to one or more Learning Sections that would interactively help the user to learn the missing skills. However the Learning Sections haven’t been developed yet.

The system is designed for people over the age of 18. If the user or the person the user has a conflict with are younger than 18, the system would give a warning, but lets the user make the final decision if he/she wants to use the system.

The program begins by gathering biographical information about both the user and the person that the user is having relationship trouble with. The user is then taken through a series of modules, each of which focuses mainly on a specific aspect of conflict resolution.

1. The first module examines the user’s general beliefs about conflict resolution.
2. The second and the third modules – Abuse and Addictions, are intended to be screening modules for the big system. This means that if the user collects a high score in one or both modules, the system will direct the user to the corresponding Learning Section before the user can go further. There is a very good psychological reason for doing this. However, in this version the user can go through all the questions, even if the scores for these modules are high.
3. The next three modules – Projection & Conflict Resolution, Hurt, Accusations – deal with specific skills of the user in this particular conflict.
4. The Body Symptoms module examines possible connections between the current conflict and physiological reactions the user might have to it without realizing it.
5. The last module – Recommendations, analyzes the scores the user collected through all the modules, gives the recommendations, and directs to the corresponding Learning Sections (that are not there yet).

Since psychological knowledge is so inexact, it is impossible to draw any conclusion on the basis of one fact or question. That is why we employed a scoring method for the evaluation of various skills involved in conflict resolution. We kept track of the following eleven scores:

High score indicates that the user needs to learn or improve this particular skill, and will be referred to the correspondent Learning Section. However, even without the Learning Sections being developed, pointing out to the particular skills that are needed for resolving the conflict I believe has a value to the user.

We calculated maximum score for each variable, then divided it by three. The lowest third considered being low, the medium third – moderate, and the highest third – high. If the score is high, the system gives strong recommendation(s) with a high priority. If the score is moderate, the system would give moderate recommendation(s) with a lower priority. If the score is low the system won’t give any recommendation. You can find the scoring scales and the range of priorities in the Design Spec Document.

The questions asked of the user are multiple choice questions. The answers are checked for validity. If the answer is not valid, the user will be prompted to answer again. The incorrect answer is ignored entirely.

We make gender distinctions in our questions. We also refer to the name that the user entered as the other person involved in the relationship. This information is gathered in the introductory module of the system.

The user is asked in the beginning of each module if they wish to continue with the program or exit. If the user chooses to continue, they will be asked the next series of questions. If they choose to quit, they will exit the system.

After going through the entire program the user receives recommendations in the order of priority.

The system consists of 9 separate modules. The user works through them in a linear manner.

The sequence of questioning is often important because the answer to one question may determine if other questions should be asked. It’s also important from the point of view of making sense to the user. In most of the modules there are several sections where it doesn’t necessarily matter in which order the sections are asked. The last module, Recommendation, is not linear at all. The system matches the final scores to the recommendations and fires the right ones. However, since the priority of the recommendations was a requirement, the recommendations that fired would pop up in the order of priority.

You can find detailed description of the system design in the Design Spec Document.

It would probably be possible to code a more general form of the rules that we used. The rules are very similar because we agreed on a common coding technique for both the question rule and the scoring rule based on the answer of each question. If we could create a small set of general rules, we could probably define a set of facts that contain the differences between the rules. This might result in smaller, more easily modifiable code.

Our method of scoring each question was based on either adding or subtracting an integer to a particular problem score. The integer was determined by the expert and depended on the answers given to each question by the user. The problems with this scheme is that the weighing of all integers needs to be consistent within a given scale across all modules and all questions. This means that if one question is weighted incorrectly, it could throw off the conclusion and recommendation. We could use certainty factors instead of integer scores to accomplish the same result. Questions would still have to be weighted within a relative scale, but certainty factors might be more intuitively obvious. They might also make unbalanced weighting easier to locate, and it might make adding more modules easier, since it wouldn’t throw off the relative scale of the integer value. This would also mean that when we give our recommendation we could indicate exactly how certain we are of the given conclusion.

On the other hand, we understand that it’s extremely difficult to estimate certainty factors for each single piece of information received from the user as well as for the rules. From this point of view the scoring approach seems easier to implement.

In addition, both of the mentioned alternative approaches would mean more complex code and a greater time investment.

The questions are asked in a text format. It would be nice if there was a GUI with buttons that the user would click. This way we could avoid making the user type in an acceptable response.

The system doesn’t give the user the opportunity to reject a recommendation as we planned in the Proposal. This is because in a psychological system you can’t give out one remedy to the user. You want to make her aware of the entire range of the options and possibilities. That is why the user gets all the recommendations in the priority order.

The scope of relationship conflicts covered by the system is wide but not exhaustive. Some situations could be covered partially. However even in this case system would give a good partial awareness of the situation.

The system requires a certain level of introspection from the user. It means that the audience for the system is somewhat limited.

Sometimes it’s hard to express certain ideas, which makes the text rather heavy and difficult to comprehend.

The system needs more testing and psychological validation.

The system still resembles a linear flow system more than a pure forward chaining expert system. We feel this problem would go away as we increased the depth of the knowledge base. However, we were not able to achieve much depth due to time constraints.

As we mentioned in Alternative Approaches, we used an integer score to represent the degree of a problem. This made it impossible to determine exactly what questions produced those scores when producing a diagnosis, and the relative weighting of the scores had to be carefully monitored.

• Development of the Learning Sections of the System. However we feel that it would be too large for one expert shell. So we probably will have to use several shells with the Black Board capability.
• A more friendly user interface, once that supports a mouse for example, might improve the feel of the program.
• Polishing and improving the text.
• Additional testing and psychological validation.
• Possible employment of the certainty factors in the future versions.

Psychologist: Dr. Peter Thomas
Techniques in marriage and family counseling. Watts, Richard (Ed). Memphis,
Tenn: Family Service of Memphis. 1(1999).
An introduction to the profession of counseling. Nugent, Frank. Columbus,
OH, USA: Merrill Publishing Co. (1990).