The comprehensive syntax of iStar modeling language allows requirements analysts to clearly capture stakeholder's needs, as well as dependencies among stakeholders. However, such iStar models cannot be automatically laid out using typical layout algorithms, such as hierarchical layout and circular layout. Thus, constructing and adjusting iStar models are laborious tasks, especially when dealing with large-scale models. In this paper, we propose a tentative approach to automatically lay out iStar models using a force-based layout algorithm. In particular, our approach has been designed by taking into account the syntax of iStar models in order to ensure both neatness and understandability of resulting models.
iStar modeling framework has been used as an e cient approach to capture and analyze stakeholder's needs in the early phase of requirements engineering. In particular, the comprehensive syntax of iStar modeling language allows requirements analysts to clearly capture stakeholders' requirements, their rationales, as well as dependencies among stakeholders.
However, to the best of our knowledge, there is no algorithm to automatically
lay out iStar models. Typical layout algorithms, e.g., hierarchical and circular
layout algorithms, cannot serve this purpose, because they do not t iStar
syntax. As a result, analysts have to spend a signi cant amount of time dealing
with layout issues when creating or modifying iStar models. Such a problem is
exacerbated by the growth of model scale, resulting in a scalability problem [
Challenges of automatic layout of iStar models inherit from their comprehensive syntax. In particular, there are two views of presenting iStar models, each of which has its particular requirements for layout. Firstly, the SD (Strategic Dependency) view exclusively presents actors and their dependency relations, the layout of which should avoid crossed links while maximizing model readability. Secondly, the SR (Strategic Rationale) view includes internal details of actors, e.g., goals and tasks, which are supposed to be presented as a tree-like structure.
Using force-based layout algorithms is a classic graph drawing method, which
simulates a physical system with edges acting as springs and nodes as repelling
objects [
In this paper, we present our ongoing research of automatic layout of iStar models. In particular, we propose a forced-based algorithm to automatically lay out elements in the SD view, which has been customized based on the syntax of iStar models in order to produce meaningful layouts. The automatic layout of SR view is not covered in this paper. We will discuss our initial ideas towards this issue and left its in-depth investigation for future research. It is worth noting that terminology used in this paper is aligned with iStar 2.0 modeling language.
When it comes to layout of the SD view, a primary goal is to calculate a reasonable position for each actor. To this end, we need to rst transform an iStar SD model into an undirected relational graph, which is then used as input of our force-based layout algorithm. Speci cally, each actor in the SD model is turned into a node, and relationships among actors (e.g., dependency and part-of ) are abstracted as edges in the relational graph. It is worth noting that the direction of an edge does not a ect the outcome of our layout algorithm, and thus is not considered in this paper.
Dependency relationships among actors are essential in the SD view, re ecting to which extent two actors interact with each other. In order to produce meaningful layout, our approach takes into account such dependency relationships. Speci cally, the more dependency relationships exist between two actors, the closer they are. As a result, when abstracting edges among actors, we also calculate their corresponding weights and store them in an edge weight matrix EW , The element in the matrix EWij is the number of dependency relationships between actors. For example, EWij = 3 means that there are three dependency relationships between actor i and actor j. Fig. 2 shows an example of the transformation from an iStar SD model to a relational graph.
Let G(VG; EG) denotes a relational graph, VG represents the node set of G,
and EG represents the edge set of G. We use FR (Fruchterman and Reingold)
layout algorithm [
area number of nodes (1)
As shown in Eq. 1, k is calculated as the radius of the empty area around a node, where the constant C is found experimentally. In such a way, all nodes are able to be uniformly distributed in a canvas. Subsequently, given d the distance between the two nodes, attractive forces fa and repulsive forces fr are de ned as below: fa(d) = d2 (2) fr(d) = k2 (3)
k d
Our layout approach takes into account the weight of each edge. Speci cally, an edge with bigger weight indicates that its endpoints will be arranged closer. Thus, we improve the attractive force formula as fa(d) = EW d2 (4) k
For edges that are derived from dependency relations, the corresponding dependum will be modeled in the middle of edges. As such, two nodes should not be placed too close to each other. To this end, a threshold is de ned to control the minimal distance between connected nodes.
As summarized in Algorithm 1, our approach rst transforms a textually represented iStar model into a relational graph, while forming an edge weight matrix. After that an improved FR algorithm is iteratively applied to calculate the position of each node, based on which a graphical iStar model can be constructed. 2.2
As for the layout of SR view, which is left for future work, we discuss here our initial thoughts towards this issue. In general, this layout task can be divided Algorithm 1 A customized force-based layout algorithm Input: a textual speci cation of a iStar model Output: a graphical representation of the iStar model 1: Transform the iStar model to a relational graph G; 2: Calculate an edge weight matrix EW ; 3: Apply the improved FR algorithm; 4: f Calculate k 5: For i = 1 to iterations Do 6: Calculate repulsive forces; 7: Calculate attractive forces; 8: Adjust the total displacement; 9: if the distance between two nodes is less than 10: adjust moving direction; 11: g 12: Construct the graphical iStar model into two sub-tasks: laying out all the intentional elements inside actor boundaries and positioning each actor boundary as a whole in the entire picture. Since most intentional elements are supposed to be organized in a tree-like structure, an intuitive solution is using a hierarchical layout algorithm. However, for quality requirements and corresponding contribution links, which are laid out in a different manner, we need to design particular algorithms to deal with their layout. Once elements inside actor boundaries can be well arranged, we can then treat each of them as a single element and layout them together with actor nodes. To accommodate this, we will need to further adjust the force-based algorithm proposed in this paper. 2.3
Once we extend our proposal to cover the layout issue of SR view, we plan to implement the entire layout approach as an independent library, which is able to be integrated into various iStar modeling tools 1. As a starting point, we plan to rst implement the layout approach on top of our previous goal modeling tool MUSER 2, serving as a prototype tool.
With the help of the prototype tool, we plan to evaluate the e ectiveness of our layout approach through a series of studies. Speci cally, we are interested in investigating the following research questions: { Can our approach create understandable layout of i models? We plan to carry out several user studies, in which we will ask iStar experts to assess to which extent the resulting layout of our proposal can be understood. { Is our approach scalable to large-scale models? We will evaluate the scalability issue of our approach by applying it to a set of large-scale models. 1http://istar.rwth-aachen.de/tiki-index.php?page=i%2A+Tools 2http://disi.unitn.it/~li/MUSER/Intro.html { Can our approach lead to faster modeling practices? We plan to design a controlled experiment to compare the performance of two groups, one of which is assisted by our approach. In particular, we aim to collect feedback from participants via focus group interview, based on which we can improve our approach to better support iStar modeling. 3
We have been aware of a number of proposals that are relevant to ours. OpenOME
and jUCMNav provide some basic graph layout functions to automatically
arrange the goal elements, which mainly focus on laying out intentional elements
inside actors [
Di erent from most iStar modeling tools, Grau et al. propose J-PRiM which
does not show the i* elements in a graphical way but in a tree-form hierarchy [
Graph visualization helps users to gain insight into data by turning the data
elements and their internal relationships into graphs [
In this paper, we present a tentative approach to automatically lay out iStar models using a customized force-based layout algorithm. As ongoing research, our proposal mainly focuses on the layout of iStar SD view thus far. In addition, we have discussed possible ways of extending our approach in order to lay out elements in the SR view.
Our approach can be applied to textual representations of iStar models, the
bene ts of which are twofold. Firstly, modelers are able to exclusively focus
on the content of models to be built without being bothered by layout issues.
As such, our approach can save much e ort on the creation of iStar models,
especially when modeling large-scale scenarios. Secondly, there are increasing
demands of improving usability of iStar modeling tools in order to better deal
with scalability issues [
As for future work, we plan to rst extend our proposal is to deal with automatic layout of the SR view. After that we will implement our layout approach in a particular iStar modeling tool, based on which we will evaluate the usefulness of our approach through a series of empirical studies.
Acknowledgments. Tong Li acknowledges the support of BJUT Startup Funding No.007000514116022. Xiaolin Du acknowledges the support of Beijing Postdoctoral Research Foundation No.Q6007012201601.