Long Time No See! Open-Domain Conversation with Long-Term Persona Memory

Findings of the Association for Computational Linguistics: ACL 2022, pages 2639 - 2650 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Long Time No See! Open-Domain Conversation with Long-Term Persona Memory Xinchao Xu1∗, Zhibin Gou1,2∗, Wenquan Wu1, Zheng-Yu Niu1, Hua Wu1, Haifeng Wang1 and Shihang Wang3 1Baidu Inc., China 2School of Computer Science, Beijing University of Posts and Telecommunications 3Columbia University {xinchaoxu,wuwenquan01,niuzhengyu,wu_hua,wanghaifeng}@baidu.com zebgou@gmail.com, sw3275@columbia.edu Abstract Most of the open-domain dialogue models tend to perform poorly in the setting of long-term human-bot conversations. The possible reason is that they lack the capability of understanding and memorizing long-term dialogue history in- formation. To address this issue, we present a novel task of Long-term Memory Conversation (LeMon) and then build a new dialogue dataset DuLeMon and a dialogue generation frame- work PLATO-LTM with a Long-Term Mem- ory (LTM) mechanism. This LTM mechanism enables our system to accurately extract and continuously update long-term persona mem- ory without requiring multiple-session dialogue datasets for model training. To our knowledge, this is the first attempt to conduct real-time dy- namic management of persona information of both parties, including the user and the bot. Re- sults on DuLeMon indicate that PLATO-LTM can significantly outperform baselines in terms of long-term dialogue consistency, leading to better dialogue engagingness 1. 1 Introduction Persona is crucial for open-domain dialogue sys- tems to establish long-term intimacy with users (Huang et al., 2020). Existing persona dialogue datasets such as PersonaChat (Zhang et al., 2018; Dinan et al., 2019) and models (Li et al., 2016a; Zhang et al., 2017; Qian et al., 2018) have greatly facilitated the chatbot with configurable and persis- tent personalities. Nevertheless, current open-domain dialogue sys- tems still cannot build a long-term connection with humans. The possible reason is that they lack the capability of understanding and memorizing long- term dialogue history information, which we called ∗ Equal contribution. The work was done when Zhibin Gou and Shihang Wang were doing internship at Baidu. 1Our data and codes are released at https: //github.com/PaddlePaddle/Research/tree/ master/NLP/ACL2022-DuLeMon 最近又看电视剧了吗 (Have you watched TV series recently? ) 小张，好久不见！ (Long time No See, Xiaozhang!) 叫我小张就行 (Just call me Xiaozhang) 我叫小明，你在干啥呢 (I’m Xiaoming, what are you doing) 我在看电视呢，我特别喜欢电视剧 (I’m watching TV, I really like TV series) After multiple sessions …… User Chatbot User Persona Memory Chatbot Persona Memory • 我叫小明 (I’m Xiaoming) • …… Remember Trigger Trigger Remember 你好 (Hello) 你好，怎么称呼？ (Hello, what’s your name?) 是啊，好久不见！ (Yeah, long time!) • 叫我小张就行 (Just call me Xiaozhang) • 我特别喜欢电视剧 (I really like TV series) Figure 1: A sample of long-term conversation with memory. At first, the chat partner is not familiar with each other, so the goal is to get to know each other; Then, after multiple sessions, the chatbot already has a certain understanding and memory of the user’s persona and its own persona, making the deep chat possible. long-term persona ability. Remembering and ac- tively utilizing the user’s persona increases engag- ingness and contributes to long-term friendships be- tween chatbot and user (Campos et al., 2018). With- out this …

Content truncated to 500 words