Gesellschaft fr Informatik e.V.

Lecture Notes in Informatics


Meta-Modelling and Ontologies, Proceedings of the 2nd Workshopon Meta-Modelling, WoMM 2006 P-96, 85-108 (2006).


2006


Editors

Saartje Brockmans, Jürgen Jung, York Sure (eds.)


Contents

Evolutionary method engineering - A case study in meta modeling

Becker J. , Seidel S. , Pfeiffer D. and Janiesch C.

Abstract


Meta modeling is a widely established means for developing conceptual modeling methods (CMM). Here, we show how a CMM for the structured conceptual design of management information systems has been developed in an evolutionary process based on meta modeling. The aim of management information systems is to satisfy the information need managers have to successfully accomplish their tasks. The quality of management decisions is highly dependent on the information they are based on. A structured conceptual design of management information systems is a crucial task that has to precede their implementation and monitoring. Several conceptual modeling methods have been developed in order to support the specification of data warehouse structures and management information systems. However, none of them was found to be appropriate to bridge the communication gap in the process of requirements analysis. Thus, in an ongoing research effort, a CMM has been designed to adequately support the conceptual design of management information systems. Several case studies were conducted and in an iterative process the findings were incorporated to improve the initial CMM. The result of this process is a CMM quite different to the original one. The aim of this paper is to elaborate on the evolutionary development of a CMM using meta modeling and to show how it has successfully been applied in multiple case studies. 85 The aim of management information systems is to satisfy the information need managers have to successfully accomplish their tasks. The quality of management decisions is highly dependent on the information they are based on. Thus, a structured conceptual design of management information systems including a comprehensive information requirements analysis is a crucial task that has to precede their implementation and monitoring [JKD01]. Information requirements analysis has to support the conceptual specification of information needs [WS04]. To facilitate information requirements analysis there is demand for a conceptual modeling method (CMM) that is understandable to both IT developers and system users. By this, management support and user participation in the development process can be increased dramatically [BDR03; WW01]. Furthermore, the CMM should hold a degree of formality that allows for the derivation of data warehouse and OLAP structures and thus fosters the implementation of the resulting information system [Ho02]. The aim of this paper is to present an ongoing research effort in which a CMM for adequately supporting the conceptual design of a management information system has been designed. We particularly focus on how the method was developed by means of meta modeling. The work we describe here spans across a period of altogether nine years. During this time five case studies and multiple iterations through the research process were performed. Due to word limitations, in this paper we present only two cases. However, the selected cases document all significant changes in the CMM. We show how the CMM has been applied to conceptually design management information systems. In an iterative process the findings of the case studies were incorporated to improve the initial CMM. We distinguish between evolutionary method improvements and method customizations. Evolutionary method improvements arise in projects with particular companies but may be used within other companies or projects as well. Method customizations address requirements of a certain company or project and are not expected to be used in further projects with different companies. In this paper, we focus on evolutionary aspects of method engineering and thus on evolutionary method improvements. The structure of the paper is as follows: First, we elaborate on the foundation of our research endeavor by explaining the research methodology that guided us through the evolutionary process of method engineering. Second, we provide an overview of related work and provide a starting point for the design of the original CMM. Third, we present two case studies followed by a cross case analysis to describe the changes that were made in the research process. We particularly regard to the according meta models and show how the changes were incorporated. We conclude summarizing our results and provide an outlook to future research activities. 86 The research methodology of this paper is illustrated in Figure 1. This research belongs to the design science paradigm [He04; MS95]. It strives for developing a practically relevant IT artifact in form of a domain independent, purpose-specific CMM. Figure 1: Research Method 87 The research process started with the awareness of a practical problem [Ta90]. The practical problem the researchers were concerned with and still are can be formulated by the following research question: How to specify a management information system on a conceptual level? The aim of the research process is to provide a CMM guiding the development of a management information system. Our comprehension of an architecture of a CMM is described in Figure 2. In this paper we focus on the language based aspects of the CMM. The practical need for such a CMM became apparent in discussions with IT managers about their experiences with management information systems (inductive reasoning) and from theoretical considerations about cost accounting theory (deductive reasoning) [Ho99]. From the inquiry it turned out that to solve the identified practical problem such a CMM must: (R1) Foster the company wide communication in order to identify relevant information needs and eliminate weaknesses in the current management information systems. The effective implementation of a management information system must be preceded by an in-depth information requirements analysis including an evaluation of the current management information system. This necessitates a sound communication between IT developers and system users. Hence, a CMM must support this process by providing a conceptual language that can be understood by both parties. $(R2)$ Prepare and assist the technical implementation of the management information system. From a technical perspective a management information system consists of a data warehouse, a corresponding online analytical processing (OLAP) tool as well as standard reporting tools. As the Entity Relationship approach [Ch76] prepares the implementation of a database structure, the CMM should provide the conceptual basis for an implementation of a data warehouse structure. Furthermore, the models of the CMM should serve as a template for the specification of the reports in OLAP and standard reporting tools. 88 Figure 2: Architecture of a Conceptual Modeling Method In the next step of the research process, a literature review has been performed in order to consider previous work on management information systems (cf. section 3 on related work). Based on (R1) and (R2) the study of literature resulted in a set of requirements which must be met by a CMM in order to be able to specify a management information system. Three major approaches could be identified which already aimed at the specification of management information systems. However, a detailed analysis of these approaches based on the comparison process of Song \& Osterweil [SO94] showed that they do not meet all relevant requirements. Therefore, the decision was made to develop a CMM for the specification of management information systems from scratch. The construction process of the CMM [Ho01; Ho99] is comparable to the procedures proposed by Greiffenberg [Gr04, p. 166 ff.] and Gupta \& Prakash [GP01, p. 143]. First, all relevant Modeling Language Concepts were identified. Afterwards, the resulting concepts were supplemented with Attributes and, thus, relations between the concepts were established [GP01, p. 143]. In the next step, the Modeling Language Constructs have been assigned to one or more Modeling Language Views in order to reduce the complexity of the resulting conceptual models. All elements of the conceptual modeling language as part of the CMM have then been consolidated in form of a Language Based Meta Model. To make all those design choices traceable all Decisions including their Reasons have been documented in form of a Method Rationale [Ro00, p. 6 ff.]. 89 As mentioned above, all elements of the conceptual modeling language as part of the CMM have been consolidated in form of a Language Based Meta Model. Whereas a model is an abstract representation of a real world object, a meta model is a model of a language for modeling that real world object [St96; N96]. If the abstraction takes place on a language based level the meta model is called a language based meta model [St96]. Holten shows the cohesion between model and language on the following levels: meta level, type level, and instance level [Ho00] (cf. Figure 3). Model M1 of a part of the real or perceived world is constructed in a language L1. Model M2 is a model of the language L1. Thus, it is a language based meta model of the object that is part of the real or perceived world. The model M2 is constructed in a meta modeling language L2. M2 represented in L2 meta level language based meta model of model of M1 represented in L1 type level model of part of the real or perceived world instance level Figure 3: Meta Modeling - Language Abstraction Levels [Ho00] The evaluation of the CMM has been performed in multiple case studies [Y03, p. 43 ff.]. While setting up these case studies the following four hypotheses were generated based on (R1) and (R2) as well as on considerations from evolutionary method engineering: (H1) An application of the CMM improves company wide communication and supports the identification as well as the elimination of weaknesses in the current management information system. $(H2)$ An application of the CMM is useful for the technical implementation of a management information system. $(H3)$ Method engineering is a continuous, evolutionary process which does not converge to a universal method which meets all technical and organizational contingencies [Ro04, p. 358]. $(H4)$ Meta Modeling is an appropriate means to support continuous, evolutionary method engineering. 90 Using theoretical sampling [Ei89, p. 537; GS67, p. 45 ff.] in each of the two iterations an organization was identified which faced the problem of inaccuracies and inconsistencies in its management information system. Since the problem of these organizations and the scope of the CMM corresponded, these companies were selected for an application of the CMM. As data collection techniques, interviews and document analyses have been used. The names of the companies, facts on performed interviews, and the examined documents are listed in Table 1. The interview partners in the organizations were chosen according to the corresponding roles of the procedure model of the CMM. As relevant roles persons in charge of the reporting, report recipients, and system administrators were identified. Moreover, relevant documents were selected according to the procedure model of the CMM. Strategy documents, reports, specifications of the IT infrastructure, documents on the organizational structure, and process models were considered. Guided by the procedure model of the CMM, in a series of semi-structured open-ended questions the participants were asked to explicate their information needs and to specify the data they can provide to satisfy the information requirements of the company. The interviewer documented the responses of the interview partners by applying the modeling language of the CMM. As second data collection technique the documents of the organizations were analyzed according to the information needs and the available information stock. The resulting data was described by using the same modeling language of the CMM. During the application of the CMM the interviewer noted the reactions of the interview partners on the procedure of the CMM. The interviewer (method expert) also recorded statements from interview partners as well as described facts extracted from documents, which were relevant from his perspective for the specification of the management information system, but not describable by means of the CMM. Organization Number of Interview Partners Examined Interviews Documents Swiss Re 7 Financial Service Paper based reports, Manager, Risk contracts Manager, IT Project Manager Christ Juweliere und 21 CIO, Controlling, Top Standard reports Uhrmacher seit 1863 Management, (daily, monthly), GmbH Managers Purchasing, purchasing data Managers Sales, warehouse Managers Logistics, IT Department Table 1: Facts on the case studies 91 The experiences from the interviews and the document analyses were then compared to the hypotheses. New requirements on the CMM were derived, deficiencies in the CMM were identified, and suggestions for improvement were specified. Depending on the number and the importance of the proposed changes, the CMM has then been adapted via meta modeling. Modeling Language Constructs were introduced, were omitted or were modified. All Decisions about adaptations of the CMM including their Reasons were documented in form of a Method Rationale. After the adaptation of the CMM the interviews and the document analysis continued with the new version of the CMM. This cycle of interviews, document analyses, and method adaptations has been continued until the management information system of the organization has been specified completely. Then it has been decided whether a new case study should be started. The evolutionary development of the CMM followed guiding principles that have been derived from the Guidelines of Modelling (GoM). These are correctness, relevance, economic efficiency, clarity, comparability, and systematic design. These guidelines aim to “increase the quality of models beyond the fulfillment of syntactic rules”[BRU00]. For a detailed description cf. [BRS95; BRU00]. In analogy, we used the same guidelines to evaluate our CMM by questioning if models constructed using the CMM meet the guidelines. Consequently, the guidelines were applied after every change that was made to the CMM. All changes we describe here were found to meet the guidelines. After two iterations the findings of the case studies were evaluated in a cross case analysis. This resulted in a further adaptation of the CMM. 3 Related Work In order to exchange thoughts, opinions and beliefs about the development process of management information systems and its objectives, representation forms of the object system have to be created. Conceptual modeling is considered to be a suitable tool for creating such representation forms respectively conceptual models [Fr99]. As mentioned above, three modeling approaches were identified that are of major importance to the design of management information systems: the Multidimensional Entity Relationship Model (ME/RM) [Sa98], the Application Design for Analytical Processing Technologies (ADAPT) [Bu96; Bu98], and the Dimensional Fact Model (DFM) [GMR98]. Generic approaches to system design, such as UML/OO, Structured Analysis and Design Technique (SADT) or System Dynamics, were not taken into account. They do not contain explicit model constructs for the design of management information systems (see below). This, however, is the goal of this CMM. ME/RM: Since the Entity Relationship (ER) Model of Chen [Ch76] does not provide sufficient support in the design of multidimensional structures the Multidimensional Entity Relationship Model is proposed. The aim is to only slightly enhance the ER language to ensure the flexibility and the simplicity of the ER notation but to allow the definition of hierarchies with qualifying and quantifying data and the hierarchical structure of the qualifying data. New constructs are a fact relationship set, a dimension level set, and a roll-up relationship set. The former connects atomic dimension levels 92 (i. e. the dimension hierarchy) while the latter connects different dimension levels. Attributes that are assigned to the fact relationship set are regarded to as ratios. An obvious drawback of ME/RM is that alternative hierarchies are complicated to model. ADAPT: The Application Design for Analytical Processing Technologies approach is independent of any prior existing modeling language. The core elements of the language are hypercubes and dimensions. Each cube can have multiple dimensions. Each dimension consists of one or more hierarchies which in turn consist of levels. Each dimension can be associated with members, scopes, and attributes. Members are singular objects of the dimension, scopes are collections of members, and attributes are descriptive information about members of a dimension. Furthermore, models and contexts can be added. Models in this case are algebraic calculation of derived data and a context is a section of a hypercube. Ratios are associated with a cube via a ratio dimension. ADAPT covers most of the required modeling language constructs specified below. However, a comparable construct to dimension scope combinations (cf. Table 2 and section 5.1) is not available. Furthermore, ADAPT does not support modeling on instance level (i. e. “Audi, BMW” instead of “Car”) what we consider to be relevant for information requirements engineering. Modeling Language ME/RM ADAPT DFM Construct Dimension Object - Member - Dimension - Hierarchy - Hierarchy Level Dimension Level Dimension, Level Set Dimension Attribute Dimension Scope - Scope - Dimension Grouping - Dimension Hierarchy Dimension Scope - - - Combination Ratio Attribute Member Fact Attribute (Measure) Ratio System - Measure - Dimension Information Object Fact Hypercube/ Fact Relationship Set Cube Table 2: Comparison of Modeling Constructs 93 DFM: The Dimensional Fact Model provides so-called dimensional schemes which consist of a set of fact schemes whose basic elements are facts, dimensions, and hierarchies. The DFM, too, was developed to fill the conceptual gap between the enduser's requirements and the logical or physical design of the data warehouse. The center of any DFM is the fact. It is usually provided with fact attributes, i. e. ratios that measure the fact. Hierarchies are ordered around the fact and provide aggregation paths of dimensions and their attributes that are situated along the path. Dimensions are the finest level of information. Apart from dimension attributes there may be non-dimensional attributes which cannot be used for aggregation. A drawback hindering the use of DF models to foster communication with non-IT staff is that DFM abstracts from the actual object instances and only shows hierarchy levels. In a comprehensive analysis, required modeling language constructs for the specification of management information systems were defined [Ho99; Ho01]. For an explanation of these constructs cf. section 5.1 of this paper. Table 2 shows the required modeling language constructs and how they are supported by existing CMMs [Kn04]. The table lists semantically similar constructs. If a modeling language construct is not available in a CMM this is denoted by “-“. This analysis led to the conclusion to develop a CMM from scratch that contains all necessary constructs and fosters modeling on instance level. Since 1999 when the final decision was made to design the new CMM, other approaches have been proposed to model data warehouse structures. Apart from the Common Warehouse Metamodel [OMG01] several efforts to create other modeling methods for data warehousing semantics include but are not limited to the extension of the UML as proposed by Totok [T00] as well as another ER based approach, starER [TBC99], and the Multidimensional Aggregation Cube (MAC) [TKS01]. For other comparison efforts besides this cf. [TKS01; ASS00; GG98; HPN03]. 5 Method Construction and Evaluation 5.1 Method Construction: MetaMIS As mentioned above, the MetaMIS approach has been developed to conceptually specify information needs. It comprises a conceptual language that facilitates the communication between management and IT analysts. Besides, MetaMIS models can be used to derive data structures for the development of data warehouses and management information systems [Ho02]. The conceptual language is used within a procedure model that comprises both as-is-analysis and to-be-modelling [Ho99]. For a more comprehensive


Full Text: PDF

ISBN 978-3-88579-190-4


Last changed 24.01.2012 21:56:40