Generating Responses that Reflect Meta Information in User-Generated Question Answer Pairs - LREC 2020 Proceedings Paper

Aug 29, 2026

Uploaded by Ivy

This paper addresses consistent-personality neural conversational modeling using user-generated single-turn question-answer pairs as training data. Role play-based question-answering is employed to collect character-specific pairs from online users, while additional meta information is gathered, including emotion and intimacy levels. Data quality is verified, and subjective evaluation confirms the usefulness of the collected dataset for training conversational models that generate utterances reflecting the targeted meta information, particularly emotional tone.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Abstract

1. Introduction

2. Data collection using role play-based question-answering

2.1. Data collection including meta information

3. Training conversational models considering meta information

4. Experiments and evaluations

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020) , pages 5433–5441 Marseille, 11–16 May 2020 􀀍c European Language Resources Association (ELRA), licensed under CC-BY-NC Generating Responses that Reﬂect Meta Information in User-Generated Question Answer Pairs Takashi Kodama 1􀀃, Ryuichiro Higashinaka2 , Koh Mitsuda2 , Ryo Masumura2 , Yushi Aono2 , Ryuta Nakamura3 , Noritake Adachi3 and Hidetoshi Kawabata3 1 Kyoto University, Japan 2NTT Corporation, Japan 3DWANGO Co., Ltd, Japan [kodama@nlp.ist.i.kyoto-u.ac.jp](kodama@nlp.ist.i.kyoto-u.ac.jp) {[ryuichiro.higashinaka.tp](ryuichiro.higashinaka.tp), [koh.mitsuda.td](koh.mitsuda.td), [ryou.masumura.ba](ryou.masumura.ba), [yushi.aono.dy}@hco.ntt.co.jp](yushi.aono.dy}@hco.ntt.co.jp) {ryuuta_nakamura, noritake_adachi, [hidetoshi_kawabata}@dwango.co.jp](hidetoshi_kawabata}@dwango.co.jp) Abstract This paper concerns the problem of realizing consistent personalities in neural conversational modeling by using user generated question-answer pairs as training data. Using the framework of role play-based question-answering, we collected single-turn question-answer pairs for particular characters from online users. Meta information was also collected such as emotion and intimacy related to question-answer pairs. We veriﬁed the quality of the collected data and, by subjective evaluation, we also veriﬁed their usefulness in training neural conversational models for generating utterances reﬂecting the meta information, especially emotion. Keywords: data collection, meta information, user-generated content, utterance generation, question-answering 1. Introduction Much attention has been paid to non-task oriented dialogue systems from their social and entertainment aspects (Wallace, 2009; Banchs and Li, 2012; Higashinaka et al., 2014) . To make such systems more engaging, it is necessary that they exhibit consistent personalities. In neural conversational modeling (Sordoni et al., 2015; Vinyals and Le, 2015; Shang et al., 2015; Serban et al., 2016), various architectures have been proposed to realize consistent personalities, such as incorporating user IDs (Li et al., 2016), personal attributes (Qian et al., 2018), and proﬁle text (Zhang et al., 2018) . However, the problem is that it is still costly to collect dialogue data for training such neural models. For eﬃciently collecting dialogue data (we deal with single question-answer pairs as dialogue data in this paper) for a particular character, a data collection method called“role play-based question-answering” has been proposed (Higashinaka et al., 2018) (See Section 2 . for details) . In this method, fans of a particular character voluntarily provide question-answer pairs by playing the role of that character. It has been demonstrated that high-quality question-answer pairs can be eﬃciently collected and that dialogue systems that exhibit consistent personalities can be realized (Higashinaka et al., 2018) . This paper extends this work and aims to collect questionanswer pairs for particular characters together with other pieces of information (called “meta information”), such as emotion and intimacy levels. The aim of collecting this additional data is to realize dialogue systems whose utterances can be controlled to reﬂect such meta information (Zhou et al., 2018; Zhou and Wang, 2018; Song et al., 2019) . This is a useful feature when we want to realize systems that are aﬀective and can become more intimate asan interaction progresses (Zhou and Wang, 2018) . We verify 􀀃 Work carried out during an internship at NTT Corporation. the quality of the collected data and empirically show that conversational models that exhibit consistent personalities as well as meta information, especially emotion, can be successfully realized by using voluntarily provided usergenerated question-answer pairs. In what follows, we ﬁrst describe the idea of role playbased question-answering followed

FAQs

What is the main goal of the paper in neural conversational modeling?

The paper aims to realize consistent personalities in neural conversational modeling by training on user-generated question-answer pairs and related meta information.

How is the training data collected for specific characters?

It uses role play-based question-answering, where fans play the role of particular characters and respond to questions from online users.

What meta information is collected and used for response generation?

The study collects emotion and intimacy levels in addition to single-turn question-answer pairs, enabling models to generate utterances that reflect these signals, especially emotion.

Aug 29, 2026

Uploaded by Ivy