Default Knowledge Preprocessing

As we know, there is default knowledge in every experiment; e.g., the default knowledge related to devices is the major portion of default knowledge. Since the devices have properties of stereotype and inheritance, use frame is suitable to represent knowledge. Besides, the pre-defined default knowledge can reduce the construction cost because it reduces the reworking when designing a new experiment.

In this thesis we propose a method to do knowledge acquisition for the preprocessing of devices and output the pre-defined default knowledge that is the knowledge frame hierarchy of devices with their attribute slots and action slots.

The proposed knowledge acquisition method is based on repertory gird which is a widely used method for knowledge acquisition. The original purpose of using repertory grid is to extract the knowledge to distinct different elements. In our approach, we use repertory grid in order to get the attribute slots and action slots of

24

device frame in constructing the hierarchy of devices. The proposed algorithm KAofDK is shown in Algorithm 1.

In Step 1 we list all the frequently used devices in the domain of specific curriculum level. Then the teacher can add new devices or remove some devices of them. In Step 2 the teacher can classify those devices into some subclasses by their heuristic. In Step 3, we handle each subclass respectively. In Step 4 the teacher lists the common attributes and actions for each subclass, and that are the reasons why the teacher can classify those devices in the same class in Step 2. Those attributes and actions listed in this step would be the device frame slots and form the first level of device hierarchy. From Step 5 the repertory grid is used to acquire attributes and actions of each device in a subclass. First, add all devices in the subclass as elements of the repertory grid. Then repeatedly choose three of the elements, ask teacher to describe the difference between one and the other two, the differences would be the constructs of repertory grid. After finishing all the iteration of asking, teacher then fills the repertory grid with three kinds of value:

1: means the device tends to have trait.

2: means the device has no tendency about trait or opposite.

3: means the device tends to have opposite.

In order to cut down attributes that have similar meaning but have different notation, we compute the row similarity. If the similarity is greater than threshold, we provide teacher the information to decide whether to combine the two attributes. With the aim to group devices as a second level of device hierarchy, the column similarity is used to provide teacher the information of grouping devices with higher similarity over given threshold.

Algorithm 1. The knowledge acquisition to form the device frame hierarchy Algorithm: KAofDK

Input

Devices that are frequently used in the domain.

Output

The slots for each device frame and the device frame hierarchy of default knowledge.

Step 1. List all of the frequently used devices.

Step 2. Classify listed devices into some subclasses by heuristic.

Step 3. For each subclass, repeat from step 4 to step 6.

Step 4. Ask teacher the common attributes and actions of this subclass. Form the first level of device frame hierarchy.

Step 5. Use repertory grid to get device frame slots.

Step 5.1. Add the devices to be the elements of repertory grid.

Step 5.2. Repeatedly choose three of the elements, ask teacher the attributes to distinct one element from the other two.

Step 5.3. Fill the repertory grid with three kinds of value.

1: means the device tends to have trait.

2: means the device has no tendency about trait or opposite.

3: means the device tends to have opposite.

Step 6. Compute the similarity to merge attributes and form the hierarchy.

Step 6.1. Compute the row similarity for attributes merge.

Step 6.2. Compute the column similarity to group devices and form the second level of device frame hierarchy.

Step 7. Finish all subclasses, Output the device frame hierarchy of default knowledge and the slots of each device frame.

26

Example 11: Knowledge acquisition process using five containers Step 1.

We have devices of condenser, U-model tube, test tube, wide mouth bottle, beaker, graduated cylinder, phosphoric acid, nitric acid, Petri dish, test tube rack, separatory funnel rack, burette clamp, centrifuge tube holder and wash bottle.

Step 2.

The devices in step 1 were classified into five subclasses.

1.Container: wide mouth bottle, beaker, graduated cylinder, test tube, wash bottle 2.Tube: condenser, U-model tube

3.Rack: test tube rack, separatory funnel rack 4.Clamp: burette clamp, centrifuge tube holder 5.Chemicals: phosphoric acid, nitric acid Step 3.

From now on we focus on the subclass of container.

Step 4.

Asking teacher the common attributes and actions of those containers.

The teacher may answer:

Container has common attributes: capacity Container has common actions: pour in, pour out

Figure 8. the first level of container in device frame hierarchy

Step 5.1

Add devices as elements of repertory grid.

Table 11. Fill out the element of repertory grid wide

Repeatedly choose three of the elements, ask teacher the differences between the three, the differences would be the construct of repertory grid.

Q1.What is the difference between wide mouth bottle, wash bottle and beaker?

A1.(The teacher answers) heat, temperature

Q2.What is the difference between beaker, wash bottle and graduated cylinder?

A2. graduation, heat

Q3.What is the difference between graduated cylinder, beaker and test tube?

A3. heat, graduation

Q4.What is the difference between test tube, wide mouth bottle and wash bottle?

A4. cover, acid endurance

Table 12. The original construct of repertory grid wide

temperature no temperature

graduation no graduation

cover no cover

acid endurance no acid

endurance

28

Step 5.3

Fill the repertory grid with three kinds of value.

Table 13. Filling out the values of repertory grid wide

Compute the column similarity, since beaker and test tube have higher similarity, we suggest teacher to group them, as “heatable container.”

Table 14. Count the column similarity to group devices wide

The output of container subclass frame hierarchy with attributes slot and action slots is shown in Figure 9. The first level of device frame “container” is obtained from step 4. The second level of device frame comes from step 6. In level 2, each device inheritances slots from parent frame, and has additional slots that has value “1” in

repertory grid.

Figure 9. The result of frame device hierarchy

Since the frame knowledge representation is understandable, through a mechanism teachers can easily add new device frame or modify slots to maintain the knowledge.

Considering other components of default knowledge, operation can be appended to previously constructed device frame hierarchy. Since the structure and scene did not have attributes of inheritance, they can be preprocessed or learned by surveying experiment cases.

30