Learning Workflow Embeddings to Improve the Performance of Similarity-Based Retrieval for Process-Oriented Case-Based Reasoning

摘要

In process-oriented case-based reasoning, similarity-based retrieval of workilow cases from large case bases is still a difficult issue due to the computationally expensive similarity assessment. The two-phase MAC/FAG ("Many are called, but few are chosen") retrieval has been proven useful to reduce the retrieval time but comes at the cost of an additional modeling effort for implementing the MAC phase. In this paper, we present a new approach to implement the MAC phase for POCBR retrieval, which makes use of the StarSpace embedding algorithm to automatically learn a vector representation for workflows, which can be used to significantly speed-up the MAC retrieval phase. In an experimental evaluation in the domain of cooking workflows, we show that the presented approach outperforms two existing MAC/FAC approaches on the same data.