BoB: BERT Over BERT for Training Persona-based Dialogue Models from Limited Personalized Data

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 167–177 August 1–6, 2021. ©2021 Association for Computational Linguistics 167 BoB: BERT Over BERT for Training Persona-based Dialogue Models from Limited Personalized Data Haoyu Song1, Yan Wang, Kaiyan Zhang1, Wei-Nan Zhang1∗, Ting Liu1 1Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, Heilongjiang, China {hysong,kyzhang,wnzhang,tliu}@ir.hit.edu.cn yanwang.branden@gmail.com Abstract Maintaining consistent personas is essential for dialogue agents. Although tremendous ad- vancements have been brought, the limited- scale of annotated persona-dense data are still barriers towards training robust and consistent persona-based dialogue models. In this work, we show how the challenges can be addressed by disentangling persona-based dialogue gen- eration into two sub-tasks with a novel BERT- over-BERT (BoB) model. Specifically, the model consists of a BERT-based encoder and two BERT-based decoders, where one decoder is for response generation, and another is for consistency understanding. In particular, to learn the ability of consistency understanding from large-scale non-dialogue inference data, we train the second decoder in an unlikeli- hood manner. Under different limited data set- tings, both automatic and human evaluations demonstrate that the proposed model outper- forms strong baselines in response quality and persona consistency. 1 Introduction Various approaches have been explored to intro- duce explicit personas in dialogue models (Qian et al., 2018; Song et al., 2019; Zheng et al., 2020; Liu et al., 2020). The PERSONA can be defined as a composite of elements of identity, such as profiles and background personal facts. In persona-based dialogues, the generated responses are conditioned not only on the dialogue context but also on some predefined personas, so the presenting personality could be more consistent. Existing persona-based dialogue models heavily utilize a set of persona-related dialogue data (Wolf et al., 2019; Golovanov et al., 2019), such as the PersonaChat (Zhang et al., 2018). This kind of crowd-sourced dataset covers rich persona features, ∗Wei-Nan Zhang is the corresponding author. Persona: I've a son who is in junior high school Query: You have any children? GPT-2: No kids. I work at home depot so I’m busy. Figure 1: A 12-layer GPT2 finetuned on PersonaChat dataset still generates an inconsistent response. namely “persona-dense”. Nevertheless, the scale of such crowd-sourced datasets is limited by the expensive costs: two annotators are asked to act the part of a given provided persona and chat natu- rally to get to know each other during the conversa- tion. On the other hand, conversations in daily life are not always persona-related. According to Twitter content analysis, less than 10% mes- sages on Twitter reveal personal anecdote or ac- tivities at home or work and even less for person- ally identifiable information (Naaman et al., 2010; Humphreys et al., 2014). As a result, the large-scale data collected from social media would only con- tain a limited amount of persona-related dialogues, which is “persona-sparse”. The limited-scale of crowd-sourced data and the persona-sparsity in large-scale data present one common challenge: a …

Content truncated to 500 words