Open Access

Yaguang Sun

• In the last years, Business Process Mining (BPMI) has become a very important research topic in academia. In the industry, also more and more big companies are starting to use such a technique to help them understand how their business processes are implemented in reality and locate the inefficient and noneffective part in their business processes. Traditional BPMI research topic can be classified into three sub-topics: Business Process Model Discovery (BPMD), conformance checking and process extension. However, as one of the most significant branch in the BPMI research area, the present BPMD techniques meet great challenges when mining process models from real-life event logs. "Spaghetti-like" process models are often generated. Such models are normally inaccurate and very complex. The main reason is that in the real world many businesses are often executed in highly flexible environments, e.g., healthcare, customer relationship management(CRM) and product development. As a result,In the last years, Business Process Mining (BPMI) has become a very important research topic in academia. In the industry, also more and more big companies are starting to use such a technique to help them understand how their business processes are implemented in reality and locate the inefficient and noneffective part in their business processes. Traditional BPMI research topic can be classified into three sub-topics: Business Process Model Discovery (BPMD), conformance checking and process extension. However, as one of the most significant branch in the BPMI research area, the present BPMD techniques meet great challenges when mining process models from real-life event logs. "Spaghetti-like" process models are often generated. Such models are normally inaccurate and very complex. The main reason is that in the real world many businesses are often executed in highly flexible environments, e.g., healthcare, customer relationship management(CRM) and product development. As a result, the event logs that stem from such flexible environments often contain dense distribution of cases with a high variety of complex behaviours. In this thesis, we explore the approaches and techniques to help existing BPMD techniques generate accurate and simple process models when mining real-life event logs. The approaches and techniques presented in this thesis mainly inherit the basic ideas of three classical strategies proposed in the literature for assisting the BPMD techniques in mining process models with high quality which are Mining Algorithm Enhancement-Based Strategy (MEBS), Model Division-Based Strategy(MDS) and Model Abstraction-Based Strategy (MAS). Moreover, the proposed techniques are also carefully designed so as to overcome the weaknesses of the current realisations of the three strategies. The main contributions of this thesis are as follows: 1. For the MEBS, we have developed a new technique named HIF which is able to help existing BPMD techniques overcome their limitations on their expressive ability. The working principle is that HIF can locate the inexpressible process behaviours in the given event logs and then transform them into expressible behaviours for the utilised BPMD techniques. 2. For the MDS, we have developed two trace clustering techniques named TDTC and CTC and one multi-label case classification technique named MLCC. The techniques TDTC and CTC are devised to optimise the accuracy and complexity of the potential sub-process models of each trace cluster during the runtime so as to assure the quality of the generated sub-models. The technique MLCC is able to combine the domain knowledge from the process experts so as to make a more meaningful division of the raw cases from a specific event log. 3. For the MAS, we have developed a mined model abstraction technique named GTCA which utilises a new model abstraction strategy proposed by us. Through this strategy, GTCA is capable of generating an abstraction process model with higher fitness and lower complexity which cannot be ensured by existing realisations of the MAS. Furthermore, trace clustering technique is employed by GTCA for optimising the quality of the found sub-process models.…