RECAP: Retrieval-Enhanced Context-Aware Prefix Encoder for Personalized Dialogue Response Generation

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 8404–8419 July 9-14, 2023 ©2023 Association for Computational Linguistics RECAP: Retrieval-Enhanced Context-Aware Prefix Encoder for Personalized Dialogue Response Generation Shuai Liu Hyundong J. Cho Marjorie Freedman Xuezhe Ma Jonathan May Information Sciences Institute University of Southern California {liushuai, jcho, mrf, xuezhema, jonmay}@isi.edu Abstract Endowing chatbots with a consistent persona is essential to an engaging conversation, yet it remains an unresolved challenge. In this work, we propose a new retrieval-enhanced approach for personalized response generation. Specifi- cally, we design a hierarchical transformer re- triever trained on dialogue domain data to per- form personalized retrieval and a context-aware prefix encoder that fuses the retrieved informa- tion to the decoder more effectively. Extensive experiments on a real-world dataset demon- strate the effectiveness of our model at gen- erating more fluent and personalized responses. We quantitatively evaluate our model’s perfor- mance under a suite of human and automatic metrics and find it to be superior compared to state-of-the-art baselines on English Reddit conversations.1 1 Introduction As tremendous successes have been achieved on open-domain dialogue generation (Zhang et al., 2020b; Cho and May, 2020; Roller et al., 2021; Shuster et al., 2022), personalized dialogue mod- els have started to draw attention because of their ability to generate consistent and engaging con- versations and their potential time-saving utility in on-message predictive generation (Wu et al., 2021; Ma et al., 2021b; Zhong et al., 2022). To generate persona-consistent responses, these models condi- tion on not only dialogue context but user personas, which can be either explicitly given or implicitly learned from the user conversations. Early works mostly focus on modeling explicit personas (Zhang et al., 2018; Zheng et al., 2019; Song et al., 2019, 2021). These methods rely on dialogue data paired with user traits, profiles or persona description sen- tences, which are difficult to collect in practice. Moreover, explicit personas usually only contain 1Our code and data are publicly available at https:// github.com/isi-nlp/RECAP. a few user traits (e.g. age, gender, and location) or a few profile sentences, so the amount of in- formation carried with them is limited, which re- stricts the models’ capability to capture and then express more nuanced personalization. Later works develop methods for automatically extracting per- sonas (Mazaré et al., 2018; Wu et al., 2020), in order to help improve content diversity, as com- pared to that seen in explicit personas. However, these extraction methods still cannot fully use all information from user history conversations. Recent works address these issues by incorpo- rating user dialogue history as their implicit pro- files (Wu et al., 2021; Ma et al., 2021b; Zhong et al., 2022). These methods generate personal- ized responses in two phases, retrieving relevant conversations from the user history and fusing the retrieved information to the generator. In the first phase, these methods retrieve a subset of conversa- tions from a user’s conversation history (Ma et al., 2021b; Zhong et al., 2022). In the second …

Content truncated to 500 words