Proactive Human-Machine Conversation with Explicit Conversation Goal

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3794–3804 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 3794 Proactive Human-Machine Conversation with Explicit Conversation Goals Wenquan Wu, Zhen Guo, Xiangyang Zhou, Hua Wu, Xiyuan Zhang, Rongzhong Lian and Haifeng Wang Baidu Inc., Beijing, China {wuwenquan01,guozhenguozhen,zhouxiangyang,wu hua}@baidu.com {zhangxiyuan01,lianrongzhong,wanghaifeng}@baidu.com Abstract Though great progress has been made for human-machine conversation, current dia- logue system is still in its infancy: it usually converses passively and utters words more as a matter of response, rather than on its own initiatives. In this paper, we take a radical step towards building a human-like conversa- tional agent: endowing it with the ability of proactively leading the conversation (introduc- ing a new topic or maintaining the current topic). To facilitate the development of such conversation systems, we create a new dataset named DuConv where one acts as a conversa- tion leader and the other acts as the follower. The leader is provided with a knowledge graph and asked to sequentially change the discus- sion topics, following the given conversation goal, and meanwhile keep the dialogue as nat- ural and engaging as possible. DuConv en- ables a very challenging task as the model needs to both understand dialogue and plan over the given knowledge graph. We establish baseline results on this dataset (about 270K utterances and 30k dialogues) using several state-of-the-art models. Experimental results show that dialogue models that plan over the knowledge graph can make full use of related knowledge to generate more diverse multi-turn conversations. The baseline systems along with the dataset are publicly available 1. 1 Introduction Building a human-like conversational agent is one of long-cherished goals in Artificial Intelligence (AI) (Turing, 2009). Typical conversations involve exchanging information (Zhang et al., 2018), rec- ommending something (Li et al., 2018), and com- pleting tasks (Bordes et al., 2016), most of which rely on background knowledge. However, many 1 https://github.com/PaddlePaddle/models/tree/develop/ PaddleNLP/Research/ACL2019-DuConv dialogue systems only rely on utterances and re- sponses as training data, without explicitly ex- ploiting knowledge associated with them, which sometimes results in uninformative and inappro- priate responses (Wang et al., 2018). Although there exist some work that use external back- ground knowledge to generate more informative responses (Liu et al., 2018; Yin et al., 2015; Zhu et al., 2017), these systems usually generate re- sponses to answer questions instead of asking questions or leading the conversation. In order to solve the above problems, some new datasets have been created, where external background knowl- edge is explicitly linked to utterances (Dinan et al., 2019; Moghe et al., 2018), to facilitate the de- velopment of knowledge aware conversation mod- els. With these datasets, conversation systems can be built to talk with humans given a topic based on the provided external knowledge. Unlike task- oriented systems (Bordes et al., 2016), these con- versation systems don’t have an explicit goal to achieve, thereof not able to plan over the back- ground knowledge. In this paper, we take a radical …

Content truncated to 500 words