SUMBot: Summarizing Context in Open-Domain Dialogue Systems

Anonymous

Keywords: dialogue summarization, dialogue context

TL;DR: This paper improves dialogue generation by including summaries of context as input.

Abstract:

In this paper, we investigate the problem of including relevant information as context in a dialogue system.
Most models struggle to identify and incorporate important knowledge from dialogues and simply use the entire turns as context, which increases the size of the input fed to the model with unnecessary information.
In order to surpass this problem, we substitute part of the context with a summary and increase the ability of models to keep track of all the previous utterances. We show that including a summary as input to a dialogue model increases the overall quality of generated responses and enhances the ability to capture information from the context long ago.