Apeak-CG: Automatically predicting emotion based dynamic multi-form knowledge fusion conversation generation

•To afford the machine to have the empathetic ability of emotional feedback, we design an Emotion Auto-predictor in the emotion branch. This predictor can analyze the dialogue context and historical emotion lines of them to automatically allocate appropriate sentiment feedback for responses.•We construct a new virtual knowledge base using a commonsense story dataset-ROCStory, which is more in line with the natural dialogues.•To take advantage of historical virtual knowledge, we propose a new delay updating algorithm to update our designed knowledge memory bank dynamically.•We dynamically integrate Virtual KB (unstructured knowledge) with KGs (structured knowledge) to enrich and expand the informativeness of dialogue.•We propose a novel knowledge-based emotion dialogue generation model, Apeak-CG. To our best knowledge, our work is the first attempt to automatically predict emotion on knowledge-based dialogue models. The experimental results and cases show the superior performance of our model.