Business process provisioning deals with the execution of a process by providing su cient resources (people, technology, and information), to realize its goals. To ensure e ective execution of the business process, process execution logs or event logs are analyzed considering multiple perspectives: i) ordering of activities (process), ii) completion time of tasks and process (timing), iii) variances based on case attributes (case), and iv) capabilities, roles of resources and their allocation (organization). A large part of existing work has focused on these perspectives in an isolated manner. This thesis is mainly concerned with analyzing variance in resource e ciency based on case attributes and its impact on process performance, thus considering organizational, case, and timing perspective together. Further, an important dimension of operational context is considered as event log data suggests that human workers or resources with the same organizational role and capabilities can have heterogeneous e ciencies based on their operational context. In practice, experienced managers use this knowledge and consider these heterogeneous resource 1 e ciency, when allocating tasks to meet the required process performance. The thesis is mainly concerned with answering operational questions such as, \How does resource e ciency vary with case attributes, resource behavior (that manifests as context), further impacting task allocation?", \How do we support task allocation based on process context, for pull-based and pull-based dispatching scenarios?", and \How do we identify useful contextual information from process data?" Hence, we refer to this work as context-aware process provisioning.
The importance of context has been recognized by many researchers in
disciplines such as mobile computing, ubiquitous computing, information retrieval, as
well as business process management [
The in uence of individual contextual parameters such as workload and
cooperation on resource performance have been studied [
The overall contribution of this thesis is to validate the in uence of multiple factors related to case attributes and the context on resource performance, develop solutions that learn the in uence of the factors from event data using supervised machine learning algorithms, and explore the possibility of mining contextual factors from additional process data sources such as textual information to support process analysis. In this thesis, in line with other de nitions of context, we consider context as exogenous knowledge potentially relevant to the execution of the process that is available at the start of the execution of the process, and that is not impacted/modi ed via the execution of the process
The key contributions to context-aware resource performance analysis and
allocation are as follows:
{ Evaluate the in uence of context and case attributes on resource performance
using event logs which leads to a context model comprising resource behavior
indicators such as experience, cooperation, preference [
In a business process where contractual service levels have to be maintained, the time taken (completion time) to complete a task is a critical performance metric. The completion time comprises of: (a) queue waiting time in the process, and (b) the service time of the resource (time required to complete a single unit of work). The queue waiting time depends on the amount of work that exists in the system and the resources available for doing that work. The service time of a worker (or resource) is in uenced by several factors. In this work multiple factors are Data-driven and Context-aware Process Provisioning considered: (a) complexity of work (b) importance or priority of the work, and (c) contextual factor such as expertise level of the worker required for a work. It is observed that, while experts have lower service time than novices for complex work and important work, they tend to have the same e ciency as novices for less important work. The study illustrates that the e ciency of human resources in any process, depends on multiple factors that go beyond the role or availability of the workers. A simulation model is built to account for the behavior of experts and novices for varying work complexity and priority. A search based optimizer uses the simulation model to arrive at an optimal sta ng. The simulation model results in di erent sta ng solutions based on the contextual factors considered. Hence, the results presented serve as the basis for guidelines on using data-driven analysis taking into consideration the operational context when allocating task to resources. 2.2
In a process, where tasks allocation follows a pull-based dispatching policy, the ownership of selecting the right task to work on, lies with the resource. The proposed approach involves recommendation of tasks to resources taking into consideration the context of the resource and the task. Such a decision support is useful to novice service workers. To this end, we build a context-aware recommender system (CARS). The event logs are used to infer context and the outcome of the execution. The outcome is the performance indicator de ned for the task. The recommender predicts the suitability of a task for a resource, by providing a rating. Prediction is based on the assumption that resources who have similar ratings on certain tasks are likely to have similar ratings towards other tasks. The rating of a task is predicted, by identifying resources who have had similar ratings on tasks under similar context, hence considering the notion of context). A context model based on resource characteristics or behavior and task context such as `time of the day' is de ned. In order to conduct the evaluation, two Business Process Intelligence Challenge (BPIC) event logs are used (BPIC 2012 and BPIC 2013). Based on the identi ed performance outcome (completion time, quality), ratings are computed for each resource-task pair. The event logs are enriched by inferring contextual factors such as resource workload, experience and preference. The rating of a task for a resource is predicted and compared with the actual rating. Prediction with and without contextual information are compared to evaluate the importance of context in deriving a suitable rating of a resource for a task. Results indicate that the mean absolute error (MAE) considering contextual factors is lower than MAE without considering context 2. Further, the improvement in MAE is studied on individual contextual factors. The factors that in uence rating vary for each business process. 2 https://github.com/renuka98/context_aware_allocation
Sindhgatta et al.
In push-based dispatching, a system or a person, is responsible for allocating the tasks of the process to relevant human resources. To study the in uence of context on resource allocation, two common explainable machine learning algorithms are used: (1) Decision tree learning, and (2) the k-nearest neighbor (k-NN). With the former, context inferred from event logs are used as input features and the process outcome is used as a target variable to derive a decision tree, predicting the performance of a process instance. The decision tree thus obtained can also be used to extract rules correlating contextual knowledge with process data when the intent is to guarantee a certain set of outcomes (in other words, a certain performance pro le). With the k-NN approach, k-NN regression is used to determine from the nearest neighbors of a process instance, those values of the process context (and particularly those that characterize resources) that would likely lead to the desired outcomes. The evaluation of the approach is presented using both a real-world dataset (BPIC 2013) and a synthetic dataset. Evaluation of the synthetic dataset aims to verify the ability of using the approach to discover context dependent task allocation rules, that are synthetically introduced in the log. The real-world data is used to validate in uence of context when allocating tasks. The decision tree model helps explain the in uence of contextual factors on process outcome (and hence resource e ciency). The model predicts process outcome with an F1-score of 67.13% on the hold-out test data. 2.4
Observing and analyzing impact of the context of a process or the
environmental factors, on its execution outcome helps adapting and improving the process.
There can be two views of context [
However, a key challenge in the evaluation of this study is the use of unsupervised learning algorithm on business process logs that do not contain a gold-standard (or ground truth), that other natural language processing tasks utilize for evaluation. Lack of process event logs containing textual information limits the ability to evaluate on multiple process logs and assess the repeatability of results. Hence, we consider this work as a rst step in mining contextual data from process logs that could be further improved with the use of supervised learning on an expert annotated or labeled dataset.
In conclusion, this thesis compiles a set of methods that are data-driven and context-aware to help managers and resources allocate suitable tasks and help improve the overall process outcome.