Learning Retrieval Augmentation for Personalized Dialogue Generation

Personalized dialogue generation, focusing on generating highly tailored responses by leveraging persona profiles and dialogue context, has gained significant attention in conversational AI applications. However, persona profiles, a prevalent setting in current personalized dialogue datasets, typically composed of merely four to five sentences, may not offer comprehensive descriptions of the persona about the agent, posing a challenge to generate truly personalized dialogues. To handle this problem, we propose Learning Retrieval Augmentation for Personalized DialOgue Generation (LAPDOG), which studies the potential of leveraging external knowledge for persona dialogue generation. Specifically, the proposed LAPDOG model consists of a story retriever and a dialogue generator. The story retriever uses a given persona profile as queries to retrieve relevant information from the story document, which serves as a supplementary context to augment the persona profile. The dialogue generator utilizes both the dialogue history and the augmented persona profile to generate personalized responses.

Introduction. Personalized dialogue generation (Zhang et al., 2018; Dinan et al., 2019), which prompts an agent to generate consistent responses based on historical dialogue context and given persona profiles, has recently drawn substantial attention in many applications. For instance, such an agent could effectively adapt to different roles such as a customer service representative by tailoring its responses to specific customer needs based on its persona and improving customer interaction and satisfaction. Besides, personalized responses can foster a sense of human-like interaction in social platforms, thereby enriching the user experience. The persona profiles contain background sentences describing the agent (e.g., I like to go hunting.) and play a crucial role in customizing the dialogue. Ideally, a persona profile should be as comprehensive as possible, containing diverse and detailed descriptions of an agent. However, these persona profiles, typically consisting of only four to five sentences, do not provide comprehensive descriptions for the persona of the agent.

Discussion / Conclusion. In this paper, we introduced LAPDOG, an endto-end learnable retrieval augmentation personalized dialogue generation framework. We show that LAPDOG jointly tunes the retriever with the generator to retrieve useful stories from the ROCStory dataset for enhancing the desired performance over the CONVAI2 dataset. LAPDOG gains consistent performance enhancement over language models with varying sizes.