Augmenting Neural Response Generation with Context-Aware Topical Attention
This work addresses the issue of generic responses in multi-turn conversational AI systems, though it appears incremental as it builds upon existing Seq2Seq models.
The authors tackled the problem of generating generic and acontextual responses in neural conversation models by introducing THRED, a model that incorporates topical concepts and previous interactions, resulting in more diverse and contextually relevant responses compared to strong baselines.
Sequence-to-Sequence (Seq2Seq) models have witnessed a notable success in generating natural conversational exchanges. Notwithstanding the syntactically well-formed responses generated by these neural network models, they are prone to be acontextual, short and generic. In this work, we introduce a Topical Hierarchical Recurrent Encoder Decoder (THRED), a novel, fully data-driven, multi-turn response generation system intended to produce contextual and topic-aware responses. Our model is built upon the basic Seq2Seq model by augmenting it with a hierarchical joint attention mechanism that incorporates topical concepts and previous interactions into the response generation. To train our model, we provide a clean and high-quality conversational dataset mined from Reddit comments. We evaluate THRED on two novel automated metrics, dubbed Semantic Similarity and Response Echo Index, as well as with human evaluation. Our experiments demonstrate that the proposed model is able to generate more diverse and contextually relevant responses compared to the strong baselines.