Customer profiles: extracting usage models from log files

The project "Customer Profiles" is executed under supervision of the Embedded System Innovation by TNO (TNO-ESI) at ASML. The project was a full-time, nine-month graduation assignment in the context of a post-master program in Software Technology offered by the Eindhoven University of Technology. The project goal was to obtain insight into the actual usage of systems by analyzing log files. The project resulted with a prototype, a portable architecture, domain analysis, and suggestions how to improve the process of extracting customer profiles. The most important project artifact is the prototype that shows the feasibility of applying process mining and resources tracing techniques to obtain insight into the actual usage of a system by analyzing log files. The prototype supports set of different activities such as: data collection, data preprocessing, information extraction, and information aggregation that work together to obtain a customer profile model that express the typical and atypical behavior of the participants in production environment as captured in the log files, which defines the prototype output. The validation phase has shown that the prototype output exceeds the stakeholders' expectations. ASML profited from the prototype output and TNO-ESI will reuse the approach for different customers. The success of the prototype output lead to a new requirement: a portable system architecture. Therefore, as a part of the project a portable system architecture that supports extracting customer profiles was designed. The architecture is based on the Pipes and Filters architectural pattern. The system architecture and design are a result of a broad architectural and system analysis, which balances between the stakeholder requirements and the most common practices in the software architecture and software development. As a part of the architecture, components that support different functionalities such as: Data Source, Event Parser, Event Enricher, and Event Combiner were designed. A lot of domain knowledge was gained during the project. The domain knowledge was transformed into a comprehensive domain analysis. The domain analysis contains the most common aspects of applying process mining for extracting customer profiles such us: mapping issues, missing information, and the minimal log data requirements. As a part of the domain analysis an evaluation of the process mining algorithms was performed. The evaluation showed that the heuristics miner and the genetic miner are the most appropriate process mining algorithms for extracting customer profiles. In order to improve the process of extracting customer profiles a list of suggestions was created. The suggestions focus on the most common problems in the logging infrastructures and in the process mining techniques. One of the suggestions is conscious manufacturer decision on the log file content. The manufacturer should define the ratio, the context (based on the minimal log data requirements), and the scope of the logging infrastructure. Another important suggestion for the logging infrastructure is having unique identifiers across the entire logging domain. The next suggestion advocates logging infrastructure on use case (end user activity) level. The last, but not the least suggestion is consistent accurate and standardized timestamp in the logging infrastructure. During the project experiments it was detected that the maturity level of the process mining tools is not on an appropriate level for industrial usage.