This article introduces a customized approach to extracting contexts and answers from online discussion forum posts, addressing limitations in previous general-purpose graphical models. The proposed method utilizes a tailored structural support vector machine (S-SVM) approach, which explores sentence relations and thread structures specific to forum posts. New inference algorithms are developed to efficiently find the most violated constraints and optimize performance measures by adjusting loss functions. Experimental results demonstrate that this customized approach is both promising and flexible in improving context and answer extraction.