Towards User Modeling Chinese Emotional Conversation Generation Dataset

One of the long-term goals of data-driven Chinese dialogue model is to generate emotional responses with user attributes. At present, the dataset modeled by users in dialogue is realized by labeling user entity attributes. This method can ensure the relative consistency of the generated and replied users' identities, but it cannot guarantee the consistency of users' emotional expression, which hinders the further development of the Chinese emotional dialogue system. Based on the movie script of "Wu Lin Wai Zhuan", the research group has manually constructed a user oriented modeling Chinese emotional dialogue generation dataset, which contains more than 16000 groups of continuous dialogue data and personality tags of 9 main characters, aiming to improve the consistency of user emotional expression of the Chinese dialogue system. The dataset contains 2 folders (1) Format_ Data refers to the processed formal text data, with a data volume of 5.82MB, which is provided to users for direct use; （2）Raw_ Data refers to raw data that has not been processed, with a data volume of 4.53MB, and is provided for users to customize the data format and annotation specification according to their needs. The data set quality control is completed by manual correction from two aspects of annotation quality control and dialogue quality control. This data set can be applied to the research and evaluation of Chinese emotion dialogue generation, user emotion modeling, emotion analysis and emotional state generation of dialogue system.