Representation Learning for Conversational Data using Discourse Mutual Information Maximization
This work addresses the need for better conversational data representations in natural language processing, offering a novel approach that improves performance on dialog understanding tasks, though it is incremental in building upon existing representation learning methods.
The paper tackled the problem of training representations for dialog understanding by proposing a structure-aware mutual information loss function (DMI) that captures conversational structure and uncertainty, resulting in significant performance improvements over baselines across nine diverse dialog modeling tasks.
Although many pretrained models exist for text or images, there have been relatively fewer attempts to train representations specifically for dialog understanding. Prior works usually relied on finetuned representations based on generic text representation models like BERT or GPT-2. But such language modeling pretraining objectives do not take the structural information of conversational text into consideration. Although generative dialog models can learn structural features too, we argue that the structure-unaware word-by-word generation is not suitable for effective conversation modeling. We empirically demonstrate that such representations do not perform consistently across various dialog understanding tasks. Hence, we propose a structure-aware Mutual Information based loss-function DMI (Discourse Mutual Information) for training dialog-representation models, that additionally captures the inherent uncertainty in response prediction. Extensive evaluation on nine diverse dialog modeling tasks shows that our proposed DMI-based models outperform strong baselines by significant margins.