The Design and Implementation of XiaoIce, an Empathetic Social Chatbot
By Li Zhou, Jianfeng Gao, Di Li, Harry Shum. 2018
748 citations recorded by the source corpus.
This article describes the development of Microsoft XiaoIce, the most popular social chatbot in the world. XiaoIce is uniquely designed as an artifical intelligence companion with an emotional connection to satisfy the human need for communication, affection, and social belonging. We take into account both intelligent quotient and emotional quotient in system design, cast human–machine social chat as decision-making over Markov Decision Processes, and optimize XiaoIce for long-term user engagement, measured in expected Conversation-turns Per Session (CPS). We detail the system architecture and key components, including dialogue manager, core chat, skills, and an empathetic computing module. We show how XiaoIce dynamically recognizes human feelings and states, understands user intent, and responds to user needs throughout long conversations. Since the release in 2014, XiaoIce has communicated with over 660 million active users and succeeded in establishing long-term relationships with many of them. Analysis of large-scale online logs shows that XiaoIce has achieved an average CPS of 23, which is significantly higher than that of other chatbots and even human conversations.
Subfields: Artificial Intelligence, Sociology and Political Science, Natural Language Processing, LLM Agents & Tool Use
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Jason Aaron Edward Weston
- Mohit Bansal
- Jianfeng Gao
- Yejin Choi
- Luke S. Zettlemoyer
- Percy Liang
- Noah A. Smith
- Kyunghyun Cho
In Sociology and Political Science
- Yejin Choi
- Charu C. Aggarwal
- Maarten Sap
- Elizabeth Clark
- Philip S. Yu
- Douwe Kiela
- Myle Ott
- Endang Wahyu Pamungkas
In Natural Language Processing
- Jason Aaron Edward Weston
- Mohit Bansal
- Yejin Choi
- Jianfeng Gao
- Luke S. Zettlemoyer
- Graham Neubig
- Douwe Kiela
- Kyunghyun Cho
In LLM Agents & Tool Use
- Endang Wahyu Pamungkas
- Jason Aaron Edward Weston
- Jacob Andreas
- Jianfeng Gao
- Sungjin Lee
- Jiliang Tang
- Mirella Lapata
- Mohit Bansal
Citations in the corpus
- Experience Grounds Language (2020, background): “This is in stark contrast to the disembodied chat bots that are the focus of the current dialogue community (Adiwardana et al., 2020; Zhou et al., 2020; Chen et al., 2018; Serban et al., 2017), which often do not learn from individual experiences and are not given enough of a persistent environment…”
- Towards a Human-like Open-Domain Chatbot (2020, background): “Some open-domain chatbots such as MILABOT (Ser-ban et al., 2017), XiaoIce (Zhou et al., 2018) 1 , Gunrock (Chen et al., 2018), Mitsuku (Wor-swick, 2018) 2 and Cleverbot 3 (by Rollo Carpen-ter) display human-like attributes, but rely on complex frameworks, such as dialog managers with 1…”
- Open-Domain Conversational Agents: Current Progress, Open Problems, and Future Directions (2020, methodology): “commonly used wordoverlap metrics include F1 (Rajpurkar et al., 2016), BLEU (Papineni et al., 2002; Li et al., 2017c), ROUGE (Lin, 2004; Fan et al., 2019), CIDEr (Vedantam, Zitnick, and Parikh, 2015; Zhou et al., 2020), and METEOR (Banerjee and Lavie, 2005; Zhou et al., 2020). Each covers slightly d”
- Way Off-Policy Batch Deep Reinforcement Learning of Implicit Human Preferences in Dialog (2019, background): “We design several intrinsic reward functions based on the rich, interactive content of conversation, taking inspiration from the psychology of human conversation: 1) eliciting positive sentiment and transitions from negative to positive sentiment, due to the importance of emotion to creating a sense”
- Human-centric Dialog Training via Offline Reinforcement Learning (2020, methodology): “Based on prior work (Zhou et al., 2018), we use the number of turns in the conversation as an indicator of the quality of the bot’s performance.”
- BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage (2022, methodology): “Similarly proprietary models (Zhou et al., 2020) or data (Ram et al., 2018) from several other products have not been openly released.”
- Recipes for Building an Open-Domain Chatbot (2020, methodology): “, 2019), Mitsuku5, Cleverbot6, and XiaoIce (Shum et al., 2018; Zhou et al., 2020).”
- Neural Approaches to Conversational AI (2018, neutral): “.luis.ai 4https://docs.microsoft.com/en-us/azure/bot-service/?view=azure-bot-service-3.0 5https://dialogflow.com 6https://www.msxiaobing.com 69 Figure 6.7: XiaoIce system architecture. Figure credit: Zhou et al. (2018) Zo,7 and Ruuh8 focus almost entirely on chitchat. Since relatively little is publ”
- Can You Put it All Together: Evaluating Conversational Agents’ Ability to Blend Skills (2020, background): “While most commercial dialogue systems rely on hand-coded narrow skills (e.g., see Zhou et al. (2020); Ram et al. (2018)), typically focusing on separate task-oriented features such as alarm setting, calendar entries, etc., we are interested in models that display various qualities in open-domain…”
- Revealing Persona Biases in Dialogue Systems (2021, neutral): “, through automated customer assistant bots (Hardalov, Koychev, and Nakov 2018) and socialcompanion bots (Zhou et al. 2019).”
- Chat as Expected: Learning to Manipulate Black-box Neural Dialogue Models (2020, neutral): “Due to the simple implementation and the strong generalization ability of neural dialogue models, they are one of the most popular techniques to build practical chatbot services (Zhou et al., 2018).”
- Challenges in Building Intelligent Open-domain Dialog Systems (2019, methodology): “A well-known example of the former is Microsoft XiaoIce [192].”
- Evaluation of Text Generation: A Survey (2020, background): “However, the most commonly used human evaluations of dialog systems is still via crowd-sourcing platforms such as Amazon Mechanical Turk (AMT) (Serban et al., 2016a; Peng et al., 2020; Li et al., 2020; Zhou et al., 2020).”
- PLATO-2: Towards Building an Open-Domain Chatbot via Curriculum Learning (2020, background): “• Microsoft XiaoIce (Zhou et al., 2020) is a popular social chatbot in Chinese.”
- Towards Emotional Support Dialog Systems (2021, background): “on daily basis (Van der Zwaan et al., 2012; Zhou et al., 2020), particularly for settings that include social interactions (accompanying and cheering up the user), mental health support (comforting a frustrated help-seeker and helping identify the problem), customer service chats (appeasing an angry”
- ProsocialDialog: A Prosocial Backbone for Conversational Agents (2022, background): “Such overly agreeable characteristics of conversational systems comes from their exposure to predominantly positive or agreeable training data (Baheti et al., 2021; Zhou et al., 2020).”