Interactive learning is becoming increasingly important in the modern educational system. Ideally students should be able to expand on their knowledge, assess their progress and receive feedback from a remote location, outside the classroom. This research presents a graphically-based methodology to model the semantic structure of textual exchanges in the form of question and answer (Q/A). A machine learning approach is then presented which classifies questions and answers based on the similarities of their semantic structures. Because the methodology is graphically-based, similarities between graphs can be identified to establish context-free relationships/ associations between answers, or between questions and possible answers. By these means the relevant textual exchanges can be systematically analyzed and classified.