A data-driven two-layer software process mining method and system

Abstract

The invention belongs to the technical field of error detection, error correction, monitoring, and discloses a data-driven two-layer software process mining method and system. Activity information isfound from the process log by clustering; the activity is converted into a single trigger sequence in the order in which the activity occurs; the single trigger sequence is divided into several loop instances by loop; the activity layer mining module is used for discovering the activity information from the process log, the input is the process log, and the output is a single trigger sequence formed by the activities associated with each event in the log, and the sequence is used as the input of the process layer; the process layer mining module is used for discovering loops in a single trigger sequence and mining each loop instance as case information. The invention divides the single trigger sequence into cyclic instances so as to reach the case information of the discovery cycle part, which is used for process mining, and perfects the current process mining method, and has obvious advantages in processing the single trigger sequence compared with other algorithms.

Description

technical field [0001] The invention belongs to the field of error detection; error correction; monitoring technology, in particular to a data-driven two-level software process mining method and system. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] Software process mining; basic Petri net system (EN system); software process data management system (SCM system). The traditional software process is mainly divided into two categories: one is the software process evaluation and improvement model represented by the capability maturity integration model CMMI (Capability Maturity Model Integration); the other is the software process modeling, which is mainly through a specific method Abstract, represent and analyze the software process to increase the understanding of the software process, and guide the actual software development activities directly or indirectly. The establishment of the traditional so...

AI Technical Summary

Problems solved by technology

Theoretically, genetic process mining can discover any process model through the log, but in fact, the algorithm convergence time is too long, and in practice, a model that meets the desired fitness is often not found, so an end criterion is often added (for example, 10 consecutive generations cannot generate ends when better individual)

Therefore, although genetic process mining can discover cyclic structures in single-trigger sequences when the number of activities is small, it cannot be well adapted to logs containing a large number of activities.

(4) The current heuristic single-trigger sequence loop instance division is not mature enough, the current heuristic-based single-trigger sequence mining algorithm, the idea is to divide the loops in the single-trigger sequence to form multiple loops Instances, based on multiple loop instances and reusing the existing process mining algorithm, the loop can be accurately mined in the end

[0005] (1) The design of the software process model is extremely complex, error-prone, and the life cycle of the model is short. People are not sensitive to the difference between the actual process and the process model, and the requirements for process engineers are high. At the same time, the software process is still evolving dynamically.

[0006] (2) Learning from mining methods in the business process field, failure to conduct targeted analysis of software process data, lack of a unified framework, so that traditional software process methods cannot be better applied; unable to establish a complete theoretical basis for software process mining; As a result, the mining results cannot meet the expected standards, and cannot support the process mining of missing task activity information and case information while ensuring the correctness and accuracy of the mining results, and cannot effectively solve the single-trigger sequence problem. When the number of activities is too large, the algorithm converges If the time is too long, a model that achieves the desired fitness often cannot appear in practice

Images