用户体验对产品和服务至关重要。关注用户在使用产品和接受服务过程中的情绪，有助于帮助改进产品设计和服务流程设计。大数据时代为我们提供了研究用户情绪的新的可能，人工智能为我们预测用户情绪提供了技术和方法。本研究将根据客服和用户之间的聊天记录来预测用户的情绪。与以往的研究不同，本研究关注客服和用户交互的整个过程，基于心理学情绪理论和语言学谈话理论，我们从聊天记录中提取一些关键指标，并应用文本挖掘和机器学习方法来处理这些多源数据。本研究试图整合多来源指标提出比文本分析更为准确的预测用户情绪的方法，因此在基于文本的情感分析的基础上加入了用户的打字速度来预测用户情绪，然后尝试进一步加入对用户情绪具有影响的客服方面的指标（如客服反馈潜伏期等）对用户离开服务场景时的情绪状态进行预测。这种跨学科研究不仅可以为理论提供实践证据，而且可以为智能聊天机器人的设计以及其他人工智能交互研究提供一些帮助。

User experience is essential for product and service. Nowadays, society enters a stage of artificial intelligence and big data. Products based on this stage are aiming to step further in user experience improvement, for which user’s emotion is a key point. Different from previous researches, this research focus on the whole process of agent and user interaction and aims to investigate better ways to predict user’ emotion than text analysis alone and indeed predict user’s emotion according to the chat log between agent and user. Based on emotion and linguistic theory, we extract some key indicators from chat log and apply text mining and machine learning methods to deal with these multiple sources of data. We added user’s type rate as the behavior index and included agent’s feedback latency as well as language style similarity to predict user’s emotion. This interdisciplinary research can not only give practice evidence to the theory but also give some help to Chabot design as well as other artificial intelligence interaction research.