期刊名称：Romanian Conference on Human-Computer Interaction
印刷版ISSN：2344-1690
出版年度：2017
卷号：RoCHI 2017
页码：125-130
语种：English
出版社：Matrix ROM
摘要：This paper presents a preliminary study on how a deep neural architecture can be used to build a chatbot which simulates a conversation having the style of a given character, including famous personalities. Thus, we analyze how well a sequence to sequence (seq2seq) neural model can learn to generate meaningful responses for open domain dialogue and to imitate a persona. For the experiments, we trained the model to imitate two different personas: Romania's most famous poet, Mihai Eminescu, and Eric Cartman from the South Park series. The model was trained without using any handcrafted rules on an end-to-end dialog and is able to build new utterances to a wide number of user inputs using original distinguished expressions for both characters. However, the generated utterances still need to be improved for a meaningful dialogue.
