Since 2015 a set of preliminary design studies were started on how to promote the stewardship of research data at RWTH Aachen University. This has resulted in a bottom-up software architecture approach that has created fundamentals for interconnection of Research Data Management (RDM) services. It has facilitated the development of essential services for the collection of structured data-sets with unique persistent identi ers. However, this service-oriented architecture has to be complemented by a set of web technologies to support the exploration and discovery of relevant data or, track and trace data within the research life cycle. With respect to lessons learned from the RDM project and literature reviews, besides technical improvements, investigation on scientists' research process and providing means for operational support ( data detection, prediction, and recommendations) are essential. Thus, this research project plans to enable operational support for RDM services across the research data life cycle while at the same time, keeping an eye on data privacy concerns. The goal is to build control- ow models, predict deviations and recommend personalized solutions by analyzing and discovering users' process model with the help of process intelligence techniques.
Scienti c research is one of the core university processes. Therefore, RWTH Aachen University is investing in research and technical development of eScience to support access to research data and its processes. eScience and, more specifically, RDM o ers opportunities to improve research processes such as reproducibility and reliability of scienti c experiments and as well as of further secondary data analysis. Despite researchers' rising demand for IT infrastructures to deliver business value, still supporting technology for scienti c research processes lacks necessary tools for Research Data Life Cycle (RDLC) services. Thus, further investigations are required to identify and implement exible services that researchers wish to use within their research processes. Through the course of the RDM project at RWTH University, a RDLC model has been presumed as it is shown in Figure 1. This RDLC model consists of six sections: (a) Data planning to determine data type, format and standards. (b) Production of data along side metadata management. (c) Data analysis to identify patterns and explore data through data mining. (d) Storage and archiving, provide means to facilitate data backup. (e) Accessing Data and data sharing to ensure availability for further publications. (f) Discovery and reuse of data to prepare it for future use and identifying relevant data.
Currently, there are number of services that support researchers within planning, Production, Storage and Accessing phases of research, but this RDLC lacks support for data analysis and reuse. Therefore, this research project is planning to employ data science and process mining techniques to deal with the shortcomings. 2
RDM involves a large number of researchers with diverse expertise and roles from many disciplines. This diversity has resulted in the development of a infrastructure for RDM with number of services for researchers. Currently, users lack a system that allows them the dynamic exploration and discovery of relevant data that originated from across scienti c disciplines. Moreover, there is no clear process model of researchers to identify most common obstacles and potentials for improving their research process. To ll into the gaps, the following research questions are proposed.
RQ1: How can process intelligence provide means in RDLC to incorporate existing heterogeneous and distributed services?
Currently, distributed services for data management are trending and come with advantages such as autonomy, ease of local data management, privacy or data protection. However, despite the advantages of distributed services, some degree of centralization is essential. At IT Center of the University, there are a number of tools and services provided to handle research data based on users' requirements. In order to o er transparency in current research practices, scholars can be supported with a Computer-Supported Cooperative Work (CSCW) platform to integrate decentralized and centralized services alike, that are involved in RDLC. Through such an integration platform, user interactions can be recorded and collected in the format of event logs (Case ID, timestamp, Activity name). Event logs are then prepared for further analytical investigations. The process mining can e ciently discover non-trivial process models, locate and extract patterns of use. By incorporating and structuring event logs and running process mining algorithms (such as Fuzzy miner, Multi-phase miner, Clustering, Decision rule mining, -algorithm, etc.) Across available services, a control- ow model for users' RDLC can be generated. These techniques provide means to monitor data circulations, analyze research habits, discover bottlenecks and improve key performance indicators.
RQ2: How to develop a self-steering tool for researchers in a complex research network to explore relevant data and opportunities for research collaborations?
Across the RDLC, it is essential for researchers to monitor and steer their research throughout the period of their investigations. Lack of research environment awareness and di culty of networking within the academic world, increases the potentials for research redundancies or missing alluring research opportunities. Therefore, a tool has to be implemented to trace and collect users' scienti c contributions and allows users to con gure their own metrics for self-evaluations or key performance indicators. By extending this system with in-depth social network analysis based on previous cooperations, users should be able to explore their network for relevant data and to identify potentials for research collaborations. Further, user experience and usability evaluations should be the pillars to development of such a system to guarantee technology acceptance and success of this tool.
RQ3: How to o er operational support in the RDLC while predicting research models and carrying out personalized recommendations? In the academic world, scientists are involved with many heterogeneous services that often discourage researchers from active participation within RDM. Hence, a successful RDM system should motivate researchers to get involved with the required tools through a seamless integration of the RDM services in an integration platform. After identi cation of the RDLC model using current data, it would be possible to o er on-the- y operational support for researchers. Thus, a web service has to collect and analyze researchers current event logs and check the compliance of the users' process model iteratively, correspondingly detecting and predicting its deviations. Additionally, besides empowering an e cient data discovery, this tool would allow monitoring of researchers performance and contributions. Furthermore, the impact and validity of the developed service has to be evaluated by experimental analysis to ascertain the usability of the system in interdisciplinary and disciplinary research environments.
RQ4: How to address technical concerns toward research data security and privacy protection throughout the RDLC? Digitalization poses a continuous risk of breach of private data or secret information. In the context of research data management, researchers are constantly concerned with losing their control over their data that being used by di erent services or shared between colleagues. Therefore, very strict privacy policies have been enforced to avoid a data breach, but often this comes with the price of losing atomization and knowledge discovery. Of course, with every digitalized service, there is a tradeo between anonymity versus publicity. However, protection of data and privacy is considered as one of the main challenges to the involvement of users in any system. Hence, it is essential to infer which information is going to be utilized and for what purpose. To protect the rights and ethics of generated or shared data, it is particularly important to address these issues by initially identifying the concerns throughout the RDLC and later discover potentials for improvements. Further, the ndings and possible guidelines should be prescribed as requirements to implement any kind of tool within RDLC to gain users' consent and hence, increase trust and usage. 3
To tackle the lack of su cient services for RDLC, above all, it is required to have a brief overview of the current state of the art and related research in this eld. Moreover, the set of challenges in this project is formed by the incorporation of multiple computer science disciplines. Namely those are: Human-computer interaction, data and process science, and data privacy. Therefore, there are three main areas that contribute to the complex. Privacy and data protection concerns on RDLC, the role of users in recommender systems and business process intelligence. 3.1
Obviously, privacy and data protection concerns are two main challenges that hinder any tool from gaining users' attention and in uence on users participation. In this regard, one needs to understand the concerns in the eld and additionally, utilize the ndings on the design of the system.
Kokolakis [
Principally, a recommender system \is a subclass of information ltering
system that seeks to predict the rating a user would give to an item" [
User-orientated design includes an understanding of how the system is being used and how our users would like to extract information from a large set of data. Business Process Intelligence (BPI) assists with the general modeling of processes and software architecture. By comparing user interaction models and expected process models, many potentials for improvement of our systems can be identi ed.
As is described by Donoho [
Researchers in the eld have also emphasized on the suitability of o ering
operational support only for structured processes [
In order to answer the research questions, this research project is inspired from
PMPL [
The stages within this research methodology are planned to gradually shape the research project and answer the research problems as it gets mature and evolves.
Stage 1) Planning and business understanding: Dedicated to the understanding of the domain knowledge, available services, and its respective processes. The results of this stage is a set of research questions and an awareness of software architecture limitations in place. Respectively, a set of \questiondriven" and \goal-driven" research questions have been determined to ful ll the aforementioned research objectives. Stage 2) Data preparation and extraction: To proceed with the determined research questions, this stage should locate and enable exploration of data. Also, select and prepare data to create event logs, then, extract and remove the noise in the data. Additionally, it has to be veri ed with respect to the research goals. Further, if necessary, renew or re ne the main research questions to t the real-world problems. Stage 3) Process mining and analysis: To acquire a suitable event log structure and produce a control ow model, it is required to aggregate events, create views, enrich and lter logs.
Prior to running any process mining analysis, structured event data has to be obtained in order to enable process discovery techniques, conformance checking and process enhancement. Stage 4) Data evaluation and process enhancement: The ndings from the previous step within a control ow model have to get enriched and integrated with additional perspectives to better help understand the as-is process. This stage supports us to verify the conformance of the process and identify bottlenecks and further suggest for actions for enhancement. Stage 5) Operational support: Operational support is the state when a system is capable of detecting deviations, predict resulting events and recommend actions on the y using pre-mortem data(live/current event data). However, active operational support requires structured processes.
Presently, at the current research phase, the feasibility of an active operational support for RDM is unknown and is debatable. 5
The resulting research should generate a control- ow model of researchers across RDLC and facilitate a personalized operational support to cope with the high demand for RDM and knowledge exchange. During the course of this research, use case studies should be carried out to ensure the validity and orientation of the research. Moreover, apart from continuous development, the ndings from research questions will be re ected in iterative design cycles and will be utilized as a proof of concept.
Despite the unpredictability nature of the process mining, it is expected to extract insightful information and build interaction models from event logs within RDLC. Further, this model has to get extended to integrate other relevant perspectives that typical researchers utilize with their research life cycle. Furthermore, this research project should empower the data provenance and exploration of research data. By running the analysis on involved software components, it is expected to discover bottlenecks in semi-distributed systems and enhance the productivity of researchers by identifying key performance indicators. Using this methodology, we should be able to extract usability issues and further identify requirements for further development for the scope users.
Additionally, by tracking and tracing event logs within the RDLC, we should manage to obtain pre-mortem event data and enable on-the- y events' prediction, and data/process recommendations. Finally, as the user interaction and his data are the centers of the system, it is important to have a realistic understanding of the privacy issues in the eld. Throughout the evolution of this project, we should obtain and provide technical solutions that elaborate privacy concerns and allow a user to achieve complete control over his data.