Topic Aware Neural Response Generation

TL;DR

This paper introduces TA-Seq2Seq, a novel topic-aware model for chatbot response generation that incorporates topic information through attention mechanisms and biased probabilities, resulting in more informative and engaging responses.

Contribution

The paper proposes a new topic-aware sequence-to-sequence model that effectively integrates topic information into response generation using joint attention and biased probabilities.

Findings

TA-Seq2Seq outperforms existing models in automatic metrics.

The model generates more informative responses according to human evaluations.

Incorporating topic information improves response relevance and diversity.

Abstract

We consider incorporating topic information into the sequence-to-sequence framework to generate informative and interesting responses for chatbots. To this end, we propose a topic aware sequence-to-sequence (TA-Seq2Seq) model. The model utilizes topics to simulate prior knowledge of human that guides them to form informative and interesting responses in conversation, and leverages the topic information in generation by a joint attention mechanism and a biased generation probability. The joint attention mechanism summarizes the hidden vectors of an input message as context vectors by message attention, synthesizes topic vectors by topic attention from the topic words of the message obtained from a pre-trained LDA model, and let these vectors jointly affect the generation of words in decoding. To increase the possibility of topic words appearing in responses, the model modifies the generation probability of topic words by adding an extra probability item to bias the overall distribution. Empirical study on both automatic evaluation metrics and human annotations shows that TA-Seq2Seq can generate more informative and interesting responses, and significantly outperform the-state-of-the-art response generation models.