Guidelines for using UML association classes and their effect on domain understanding in requirements engineering

Abstract

The analysis and description of the application domain are important parts of the requirements engineering process. Domain descriptions are frequently represented as models in the de-facto standard unified modeling language (UML). Recent research has specified the semantics of various UML language elements for domain modeling, based on ontological considerations. In this paper, we empirically examine ontological modeling guidelines for the UML association construct, which plays a central role in UML class diagrams. Using an experimental study, we find that some, but not all, of the proposed guidelines lead to better application domain models. We use a process-tracing study to investigate in more detail the effects of ontological guidelines. The combined results indicate that ontological guidelines can improve the usefulness of UML class diagrams for describing the application domain, and thus have the potential to improve downstream system development activities and ultimately affect the successful information systems implementation.

Appendices

Appendix 1: sample diagrams (fast-food operation) used in the study

See Figs. 6, 7.

Fig. 6

Class diagram developed by following the ontological guidelines

Fig. 7

Class diagram developed by violating guideline 2

Appendix 2: experimental materials

2.1 Variables: modeling and domain knowledge

• 1. To what extent do you know data modeling concepts (such as classes, operations, and attributes)?

• 2. To what extent do you have experience in using data modeling concepts (such as classes, operations, and attributes)?

• 3. To what extent do you know UML association class?

• 4. To what extent do you have experience in using UML association class?

• 5. To what extent do you know the operation of a food restaurant?

• 6. To what extent do you have experience in the operation of a food restaurant?

• 7. To what extent do you know the operation of a hotel reservation?

• 8. To what extent do you have experience in the operation of a hotel reservation?

2.2 Variable: perceived usefulness of the diagrams

• 1. To what extent do you think that the diagrams helped to answers the questions?

• 2. To what extent do you think that the diagrams made it easier to complete answering the questions?

• 3. To what extent do you think that the diagrams enhanced your effectiveness on answering the questions?

2.3 Variable: domain understanding

• 1. A customer tried to order food. She has selected the food she wanted to purchase but no food was delivered to her. What could have caused this problem?

• 2. A Driver went about his route to drop off the ordered food. However, when he reached a delivery point, he could not deliver the ordered food. What could have caused this problem?

• 3. On a particular day, the partner of the restaurant ordered ingredients for preparing food. The ingredients did not reach on the expected delivery date. What could be the possible reasons?

• 4. A guest was not a privileged hotel guest but was allowed to get a car pick up service. How could this have happened?

• 5. A guest had 7 days of reservation in the hotel. At the end of the stay, the guest did not pay for her stay. How could this have happened?

• 6. A privileged guest received the pick up service even after his membership expired. How could this have happened?

2.4 Task on developing UML class diagram

In the following space draw a UML class diagram for the description below using at least one association class.

A hospital treats patients. For each treatment, the hospital needs to record the doctor, the treatment code, and the date.

2.5 UML class concepts

 

Concept	Definition	Example
Class	A class is set of objects that share the same properties and/or behaviors	Person and hospital are concepts and therefore are modeled as classes
Attribute	Attributes are properties held by the members of a class. Attributes can have constant (such as date of birth) or variable values (such as address)	The person class can have name and address as attributes
Operations	Operations are functions or services that are provided by all the instances of a class to invoke behavior in an object	The two operations of the hospital class are register patients and treat patients
Subclasses	A subclass has more attributes or/and more operations than the general class	A patient is a subclass of a person
Association	Association is the relationship among instances of classes	Hospital and patient are related as hospital treats patients
Association class	An association class is an association that has attributes or/and operations of its own	Registration is an association class that has attributes registration number and registration date

2.6 Training on answering problem-solving questions

Please look at Fig. 8 carefully. The figure is drawn using the concepts mentioned in the earlier page. The figure describes the following situation.

Fig. 8

A patient admission situation

A patient class has the attributes name and age and an operation get treated. Admitted patient is a subclass of patient as it has additional attributes—admission date and bed number and an additional operation—get admitted. The physician class is associated with the patient class.

A physician is dissatisfied with her work. Why might this be?

2.7 Sample answers

Using Fig. 8, you come up with answers by making inferences based on the information in the diagram combined with your own background information. For example, to come up with answer 1 (in Table 9), you have to look at the classes admitted patient and patient in Fig. 9 and infer that some patients might not be admitted.

Table 9 Possible answers with explanation

Fig. 9

A patient admission situation—sources of answers

Appendix 3: the ANCOVA statistical technique

The ANCOVA technique evaluates the effect of each treatment or control variable by first calculating the mean of the dependent variable (e.g., problem-solving scores) for each experimental group (“treatments”) or control variable. Next, the sum over all observations of the squared differences of the dependent variable score from the mean of the dependent variable score of the group of each observation is computed, called the sums of squares within groups.

$$ {\text{SS}}_{\text{within}} = \sum\limits_{i = 1}^{g} {\sum\limits_{j = 1}^{{n_{i} }} {\left( {Y_{ij} - \bar{Y}_{l} } \right)^{2} } } $$

Here, Y denotes the dependent variable, a bar denotes the mean, g is the number of groups, n _i is the number of observations in group i, and c is the overall number of observations. Also, the sum over all groups of the product of the number of observations in that group and the squared differences of the dependent variable mean for that group from the overall mean of the dependent variable is computed, called the sums of squares between groups.

$${\text{SS}}_{{{\text{between}}}} = \sum\limits_{{i = 1}}^{g} {n_{i} (\bar{Y}_{l} - \bar{Y})^{2} } $$

Next, each of these “sums of squares” is divided by their degrees of freedom, defined as the number of data points for that calculation minus the number of parameters calculated. This equals the number of groups minus one (the overall mean is calculated) for the sums of squares between groups, and the number of total observations minus the number of groups (each group has a group mean that is calculated) for the sums of squares within groups. This yields the “mean sums of squares” per degree of freedom.

$$ {\text{MS}}_{\text{within}} = \frac{1}{g - 1}{\text{SS}}_{\text{within}} \quad {\text{MS}}_{\text{between}} = \frac{1}{n - g}{\text{SS}}_{\text{between}} $$

The logic of the ANCOVA rests on the observation that, if the treatment variable had no effect on the group means (i.e., all group means are equal, and thus equal to the overall mean), the mean sums of squares between the groups would be equal to the mean sums of squares within groups; in other words, their ratio should be one. This ratio is called the F statistic, reported in our tables in the text, and it is distributed according to an F distribution.

$$ F = \frac{{{\text{MS}}_{\text{between}} }}{{{\text{MS}}_{\text{within}} }} $$

One can test whether the F statistic that is calculated is significantly different from the expected value of one. This is done by calculating, from the F cumulative distribution function, that probability with which the observed F statistic would be found, if the true F statistics was one. This is the P value reported in our tables in the text. If this probability is sufficiently low (generally this cutoff is assumed to be 0.05), one concludes that the observed F statistic does not come from a distribution for which the true F statistic is one; in other words, the true F statistic is different from one, thus the ratio of mean sums of squares is different from one, and therefore the dependent variable mean differs between the treatment groups or categories. To assess to what extent the different treatment groups or categories explain the observed variation of dependent variable scores, one can compare the sums of squares calculated between the groups to the sum over all observations of the squared difference of the dependent variable from the overall mean of the dependent variable. This is the r ² value, which can be adjusted to account for the effects of sample size and number of groups.

$$ {\text{SS}}_{\text{total}} = \sum\limits_{i = 1}^{n} {\left( {Y_{i} - \bar{Y}} \right)^{2} } $$

$$ r^{2} = \frac{{{\text{SS}}_{\text{between}} }}{{{\text{SS}}_{\text{total}} }}\quad r_{\text{adj}}^{2} = 1 - \frac{n - 1}{n - g}(1 - r^{2} ) $$

Generally, a higher r ² value is better, though a value of approximately 0.25 is suggested to indicate a medium-strength effect [31, 32].