Model-Inference

Final project for CS520, please kindly refer to the technical report for details.

Model inference is a powerful technique to inspect the process and gain insights of what the process is capable of by analyzing the traces generated by this process. In order to test the sensitivity and generalization of model inference, we generate a large amount of traces from intents flow in multi-turn question answering conversations. We perform both qualitative and quantitative analysis of the inferred models, and manage to make some intuitive observations from them. The experiments indicate that given a large of amount of traces, model inference can handle the complexity of human conversations, and provide insights on several patterns of QA dialogues.