Explainable Detection of Implicit Influential Patterns in Conversations via Data Augmentation