Parallelized root cause analysis using cause-related aspect formulation technique (CRAFT)

Aspect-based opinion mining aims to provide results that aid in effective business decision making. Identifying the aspects, their major and minor causes proves to be the major challenge in this domain. This paper presents a cause-related aspect formulation technique (CRAFT) to perform opinion mining. The CRAFT model incorporates an enhanced aspect extraction module, ontology creation based on aspect and aspect categories, aspect and aspect category metadata repository creation and maintenance and a decision tree-based parallelized boosted ensemble. The proposed CRAFT model is implemented in Spark to incorporate parallelism in the architecture. The process of ontology creation and metadata repository creation aids in effective identification of both implicit and explicit aspects. Experiments were conducted using a customer review benchmark dataset incorporating reviews about five varied products. Comparisons were performed with state-of-the-art models CNN+LP, Popscu and TF-RBM. Comparisons indicate improved performances ranging up to 4% in terms of precision, up to 18% in terms of recall and up to 11% on F1 Scores, indicating the effectiveness of the proposed CRAFT model.