KAOS modeling editor is a tool that enables semi-automatic goal modeling via browsers. This tool draws/derives a goal model from a requirements description of a software system. This tool allows software engineers to discover and organize goals that the envisioned software has to accomplish. Engineers can also edit the diagram manually using a graph interface or DOT source file. Moreover, the goal description can be stated in natural languages. The tool is currently available in Github1 and can be installed in a PC with Node.js and Python. The demo video2 is also available on Youtube.
CEUR ceur-ws.org
In recent times, the requirements that software must meet have grown increasingly complex, mirroring the expanding scale of software development. Consequently, software developers often grapple with the challenge of discerning precisely what their clients expect from their products, striving to avoid either excessive features or deficiencies. To tackle this issue, a goal-oriented requirement analysis model, referred to as the goal model, is employed. In our work, we propose an open-source editing tool that automatically generates an initial KAOS goal model based on the provided requirements description, detailing the states or objectives of a software system.
Numerous studies have explored goal modeling, including eforts to extract and visualize
goals through data mining techniques [
Figure 1 showcases our proposed editor. This tool functions in environments with Node.js and
Python and is compatible with web browsers like Google Chrome and Firefox. On the left side,
there’s a text field for entering requirement descriptions. After the user input and activation by
pressing the button below, the editor autonomously generates a goal model on the right side
after a set time. You can manipulate the goal model’s structure directly using the right-side
interface. Our KAOS modeling editor is based on GVE[
This section describes the automatic construction of a goal model using the proposed tool.
Parsing sentences into words Initially, the editor parses the sentences into words. Each word has information about its part of speech, and its dependencies on other words.
Extraction of important sentences Following this, crucial sentences are identified based on specific keywords. These critical sentences pertain to the primary requirements outlined in the requirement description. To illustrate, in the context of describing an automotive control software system, sentences containing words like “car,” “driving,” and “speed,” as well as sentences following them, are considered to contain essential requirements.
To extract these vital sentences, our proposed tool leverages a concept known as an “epoch phrase” as a guiding principle for identification. An epoch phrase is a specific phrase that not only includes words occurring a predetermined number of times within the requirement description but also incorporates particular terms like “envisioned” and “to be developed.” For instance, phrases such as “The envisioned auto-driving software” or “The speed control system to be developed” fall within the epoch phrases category when terms like “auto-driving,” “software,” “speed,” and “system” appear a predefined number of times in the descriptions. In the case of a specification for an automotive control system, these terms undoubtedly refer to the automotive control system under development and typically precede related requirements. Consequently, sentences within the same paragraph following these epoch phrases are singled out as pivotal sentences.
In order to ensure rapid transportation, / advcl
obl trains should run fast, / by reducing unnecessary delays, /
obl at a high frequency.
In order to ensure rapid transportation
trains should run fast by reducing unnecessary delays at a high frequency
Extraction and linking goals Figure 2 illustrates the creation of a KAOS goal model based on word-to-word dependencies. In general, one word can be considered “dependent” on another. As demonstrated in Figure 2, the word “ensure” in the first line displays an “advcl” dependency originating from the word “run” in the second line. The arrows in the figure visually represent this connection. An “advcl” dependency arises when an “adverbial clause modifier” is present in the text. An adverbial clause modifier is a clause that modifies another word, like adverbs such as “very,” “usually,” “pleasantly,” and so on. Our tool creates a goal model by breaking down sentences into phrases based on identified word dependencies. It forms parent-child connections between these phrases according to the dependencies that link them. For instance, the “advcl” dependency links the phrase in the first line of Figure 2 with the phrase “trains should run fast” in the second line, resulting in an automatic parent-child relationship between them. However, connecting goals from diferent sentences through dependencies is challenging since dependencies are limited to words within a single sentence. In such cases, our tool compares nouns in one goal with those in another and establishes a parent-child relationship between goals with the most similar nouns. This is based on noun co-occurrence, indicating that the goals may share the same topic, describing a common concept.
Figure 3 shows a part of a preliminary KAOS goal model that the proposed editor automatically generates. The other part of the model was omitted due to space limitations. To assess the accuracy of the tool, we conducted an experiment comparing the generated goal model with the modified goal model. We used two metrics to evaluate the applicability of the editor. First, we used NDCG, which measures the accuracy of the ranking prediction, and the number of editor operations. To begin with, it was necessary for us to measure how closely the automatically generated model resembled the ideal model’s structure and to assess the diferences. For this purpose, we prepared rankings for both the automatically generated goal model and the manually adjusted goal model, arranging the nodes in each by their depth. We then compared to ensure more effective access to
state of the art books periodicals and proceedings
The new UWON library system should address such problems
It should provide interactive online facilities for book acquisition user registration loan management bibliographical search and book reservation
Access to such facilities should be restricted to convert its library new system while reducing operational costs through software based solution integrating all department libraries
The new system should interoperate these rankings using the NDCG metric to evaluate the accuracy of the model’s automatic generation. Next, we manually used the editor to measure which operations and how many times it took to modify the preliminary goal model to make a valid goal model. This is a direct indicator regarding the usefulness of the tool. In essence, it’s like having a rule such as “extract phrases containing infinitives as goals.” The algorithm doesn’t distinguish contextually relevant infinitives from irrelevant ones; it extracts them all. Solely relying on algorithmic judgments for context is a significant hurdle. To address this fundamentally, it’s advisable to explore the integration of large-scale language models like GPT-4 in the future, capable of understanding and interpreting context.
In this study, we introduced our tool, the KAOS Modeling Editor, designed to autonomously create initial KAOS goal models from requirement descriptions. It identifies crucial segments using epoch phrases and then extracts and connects goals based on dependencies and noun co-occurrence. The highest accuracy achieved for initial models is 0.940 by NDCG, with a maximum generation time of 30 minutes. Our future goals include integrating LLMs like GPT-4 into the goal model implementation process to improve accuracy and processing speed.