For over 12 years, machine learning is used to extract opinion-holder-target structures from text to answer the question: Who expressed what kind of sentiment towards what?. However, recent neural approaches do not outperform the state-of-the-art feature-based model for Opinion Role La