{"title":"Research on Multi-turn Dialogue Generation Strategy Guided by Topic","authors":"P. Zhang, Hongrong Wang, Jie Wang","doi":"10.1109/AINIT54228.2021.00043","DOIUrl":null,"url":null,"abstract":"The open-domain generated dialogue model relies on masssively the neural network model to generate sentences without grammatical errors, and does not consider effective mechanisms to manage chat topics, resulting in monotonous and incoherent conversation topics. Inspired by human’s dialogue strategy, this paper proposes a topic-guided multi-turn dialogue generation strategy, DATHRED, which with a knowledge topic smoothing technology. It uses HRED to model multiple turns of dialogue, and proposes the two-way confrontation model to improve the topic richness of multi-turn dialogue and the fluency of topic transition. The comparison with the baseline model on the KdConv dataset verifies the effectiveness of our method.","PeriodicalId":326400,"journal":{"name":"2021 2nd International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 2nd International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/AINIT54228.2021.00043","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
The open-domain generated dialogue model relies on masssively the neural network model to generate sentences without grammatical errors, and does not consider effective mechanisms to manage chat topics, resulting in monotonous and incoherent conversation topics. Inspired by human’s dialogue strategy, this paper proposes a topic-guided multi-turn dialogue generation strategy, DATHRED, which with a knowledge topic smoothing technology. It uses HRED to model multiple turns of dialogue, and proposes the two-way confrontation model to improve the topic richness of multi-turn dialogue and the fluency of topic transition. The comparison with the baseline model on the KdConv dataset verifies the effectiveness of our method.