Mean?” can sometimes increase the CPS of the ongoing human–machineĬonversation. 2017) show that encompassing blandīut interactive responses such as “I don’t understand, what do you The evaluation methodologyĮliminates many possibilities of gaming the metric. Human–XiaoIce conversations collected from millions of active users overĪ long period of time (typically 1–6 months). In ourĮvaluation, the expected CPS is approximated by averaging the CPS of That corresponds to long-term, rather than short-term, engagement. It is worth noting that we optimize XiaoIce for expected CPS As we are making XiaoIce an open social chatbotĭevelopment platform for third-parties, the XiaoIce persona will be configurableīased on specific user scenarios and cultures. (e.g., Session 20), and then skillfully shifts to new topics that are moreĬomfortable for both parties. As shown in Figure 1, XiaoIce responds sensibly to some sensitive questions Knowledge, XiaoIce never comes across as egotistical and only demonstrates her Despiteīeing extremely knowledgeable due to her access to large amounts of data and Reliable, sympathetic, affectionate, and has a wonderful sense of humor. Therefore, we design the XiaoIce persona as an 18-year-old girl who is always Majority of the “desired” users are young, female users. Responses that contain swearing, bullying, and so forth. We have collected human conversations of millions of users,Īnd labeled each user as having a “desired” persona or notĭepending on whether his or her conversations contain inappropriate requests or Take the XiaoIce persona designed for WeChat deployed inĬhina as an example. Thus, for different platforms deployed inĭifferent regions, we design different personas guided by large-scale analysis Long-term, emotional connections, but also take into account culture differencesĪnd many sensitive ethical questions as exemplified in Curry and Rieser ( 2018), Schmidt and Wiegand ( 2017), and Brahnam ( 2005). With the primary design goal of XiaoIce as an AI companion with which users form The design of the XiaoIce persona needs to not only align Right expectations for users in the conversation and gain their long-termĬonfidence and trust. A social chatbot needs to present a consistent personality to set the Personality is defined as the characteristic set of behaviors,Ĭognition, and emotional patterns that form an individual’s distinctiveĬharacter.
User herself is engaged in the conversation. As shown in Figure 2, XiaoIce demonstrates sufficient EQ as she generates sociallyĪcceptable responses (having a sense of humor, comforting, etc.), and canĭetermine whether to drive the conversation to a new topic when, e.g., theĬonversation has stalled, or whether or not to be actively listening when the That are emotionally appropriate, possibly encouraging and motivating, and fit To have the ability to personalize the responses (i.e., interpersonal responses) Users have differentīackgrounds, varied personal interests, and unique needs. Social chatbot must demonstrate enough social skills. Recognition, and dynamically tracking the mood of the user in a conversation. This requires query understanding, user profiling, emotion detection, sentiment The emotions evolve over time, and understand the user’s emotional needs. A social chatbot with empathy needs to have theĪbility to identify the user’s emotions from the conversation, detect how Within her frame of reference, that is, the ability to place oneself in the Empathy is theĬapability of understanding or feeling what another person is experiencing from
Higher than that of other chatbots and even human conversations.ĮQ has two key components, empathy and social skills. Shows that XiaoIce has achieved an average CPS of 23, which is significantly Long-term relationships with many of them. Since the release in 2014, XiaoIce hasĬommunicated with over 660 million active users and succeeded in establishing We show how XiaoIce dynamically recognizes humanįeelings and states, understands user intent, and responds to user needs We detail the system architectureĪnd key components, including dialogue manager, core chat, skills, and anĮmpathetic computing module. Processes, and optimize XiaoIce for long-term user engagement, measured inĮxpected Conversation-turns Per Session (CPS). Human–machine social chat as decision-making over Markov Decision We take intoĪccount both intelligent quotient and emotional quotient in system design, cast Human need for communication, affection, and social belonging. XiaoIce is uniquely designed as anĪrtifical intelligence companion with an emotional connection to satisfy the Most popular social chatbot in the world. This article describes the development of Microsoft XiaoIce, the