Mitigating the negative impact of over-association for conversational query production

Ante Wang, Linfeng Song, Zijun Min, Ge Xu, Xiaoli Wang, Junfeng Yao, Jinsong Su

Abstract: Highlights•Over-association is a common phenomenon in existing query production datasets.•Training on queries with high over-association degrees leads to performance decline.•The over-association degree can be measured by the input and output word overlap.•A trained model prefers to generate outputs with a lower over-association degree.•Applying weighting strategies eases the negative impacts of over-association.