Social Inﬂuence Dialogue Systems A Survey of Datasets and Models For Social Inﬂuence Tasks Kushal Chawla1Weiyan Shi2Jingwen Zhang3

2025-04-24 0 0 1.06MB 17 页 10玖币

侵权投诉

Social Inﬂuence Dialogue Systems: A Survey of Datasets and Models For

Social Inﬂuence Tasks

Kushal Chawla1∗Weiyan Shi2∗Jingwen Zhang3∗∗

Gale Lucas1∗∗ Zhou Yu2∗∗ Jonathan Gratch1∗∗

1University of Southern California 2Columbia University

3University of California Davis

1{chawla, lucas, gratch}@ict.usc.edu 2{ws2634, zy2461}@columbia.edu

3jwzzhang@ucdavis.edu

Abstract

Dialogue systems capable of social inﬂuence

such as persuasion, negotiation, and therapy,

are essential for extending the use of technol-

ogy to numerous realistic scenarios. However,

existing research primarily focuses on either

task-oriented or open-domain scenarios, a cat-

egorization that has been inadequate for cap-

turing inﬂuence skills systematically. There

exists no formal deﬁnition or category for

dialogue systems with these skills and data-

driven efforts in this direction are highly lim-

ited. In this work, we formally deﬁne and

introduce the category of social inﬂuence di-

alogue systems that inﬂuence users’ cognitive

and emotional responses, leading to changes

in thoughts, opinions, and behaviors through

natural conversations. We present a survey of

various tasks, datasets, and methods, compil-

ing the progress across seven diverse domains.

We discuss the commonalities and differences

between the examined systems, identify limita-

tions, and recommend future directions. This

study serves as a comprehensive reference for

social inﬂuence dialogue systems to inspire

more dedicated research and discussion in this

emerging area.

1 Introduction

Consider a human user who signs up to interact

with a persuasive dialogue system that motivates

for engaging in physical exercise. The system: 1)

uses social cues like small talk and empathy, useful

for providing continued support, and 2) employs

persuasive strategies to convince the user who, at

least in the short-term, is reluctant to indulge in

exercise. Does such a system ﬁt the deﬁnition of a

task-oriented dialogue system that are traditionally

designed to assist users in completing their tasks

such as restaurant or ﬂight booking (Zhang et al.,

2020c)? Although the system is task-oriented or

goal-oriented per se, the task here goes beyond the

∗Equal contribution, ∗∗ Co-supervised project

traditional deﬁnition of assisting users, given the

possible misalignment between the goals of the sys-

tem and the user. Clearly, this system is also not

open-domain (Huang et al.,2020). Although con-

versations involve social open-ended interactions,

there is still a concrete goal of persuading the user

towards a healthier habit.

Scenarios similar to above are ubiquitous in ev-

eryday life, including games (Peskov et al.,2020),

social platforms (Tan et al.,2016), and therapeu-

tic interactions (Tanana et al.,2016). Dialogue

systems for these applications require a core func-

tion in human communication, that is, social in-

ﬂuence (Cialdini and Goldstein,2004;Cialdini,

2009), which involves inﬂuencing users’ cogni-

tive and emotional responses, leading to changes in

thoughts, opinions, and behaviors through natural

conversations. This goes beyond what is captured

by traditional task deﬁnitions in the dialogue com-

munity which primarily focus on task completion

and social companionship.

Despite numerous independent efforts in iden-

tifying and analyzing various social inﬂuence sce-

narios, there is a lack of common understanding

around social inﬂuence in AI research which in-

hibits a systematic study in this space. Further,

data-driven efforts for dialogue systems in this

space are highly limited. To this end, we introduce

the category of social inﬂuence dialogue systems

(Section 2), providing a comprehensive literature

review and discussing future directions.

Developing these systems holds importance in

AI research for multiple reasons. Tackling these

tasks not only involves AI but also aspects of game

theory, communication, linguistics, and social psy-

chology, making them an ideal testbed for inter-

disciplinary AI research. Most importantly, they

reﬂect AI’s general ability to consider their part-

ners’ inputs, tailor the communication strategies,

personalize the responses, and lead the conversa-

tion actively.

arXiv:2210.05664v2 [cs.CL] 24 Jan 2023

We design a taxonomy for existing social inﬂu-

ence dialogue datasets, studying their task struc-

ture (symmetric vs asymmetric) and context (lo-

cal vs global). We also organize them by their

domains: games, multi-issue bargaining, social

good, e-commerce, therapy and support, argumen-

tation, conversational recommendations, and mis-

cellaneous tasks (Section 3). We further design a

taxonomy of existing methods, assisting readers

to comprehend the progress and reﬂect on future

directions. We organize them based on the system

strategy, language generation, partner model, archi-

tecture, learning process, and the use of pretrained

language models (Section 4). Finally, we identify

key challenges and provide recommendations for

future work (Section 5).

Over the years, research in task-oriented and

open-domain dialogues has beneﬁted from a myr-

iad of survey efforts (Huang et al.,2020;Zhang

et al.,2020c;Ni et al.,2021). We instead focus on

dialogue systems with social inﬂuence capabilities

and present a thorough review across various do-

mains. We hope that our work serves as a timely

entry point for interested researchers to take this

area further, inspiring dedicated effort and discus-

sion on social inﬂuence in the dialogue community.

2 Social Inﬂuence Dialogue Systems

“Social inﬂuence is a fact of everyday life” (Gass,

2015). It is the change in thoughts, feelings, atti-

tudes, or behaviors resulting from interaction with

an individual or a group (Rashotte,2007). Inﬂu-

ence is measured by quantiﬁable proxies of the ob-

served change, like the interest to indulge in phys-

ical exercise before or after the interaction with a

system, or the ﬁnal deal in a negotiation as opposed

to one person taking it all. Social inﬂuence dia-

logue systems act interactively and inﬂuence their

partners in decision-making and behavioral con-

texts (Zhang et al.,2020a;Lee et al.,2020). This

calls for an active role by the system, distinguish-

ing them from other well-studied scenarios, such

as purely task-oriented, where systems passively

assist their partners to complete tasks, and open-

domain, that target social companionship. Key

social inﬂuence tasks include persuasion (Wang

et al.,2019), aiming to change users’ attitudes or

behaviors, and negotiation, aiming to change the

users’ perspective to achieve a common ground

(Lewis et al.,2017).

Conceptual overview

: Figure 1distinguishes be-

Figure 1: A conceptual overview.

tween the kinds of conversational content in social

inﬂuence interactions. The task-oriented content

focuses on inﬂuencing for a domain-speciﬁc goal,

like persuading for donation, bargaining with trade-

offs, or encouraging healthier habits. These interac-

tions may also contain social content, such as small

talk, empathy, or self-disclosure. The task-oriented

content provides a context for social interactions.

Depending on the task, social content is optional,

but if present, can in turn build rapport and en-

hance user-system relationship for improved task

outcomes (Liao et al.,2021).

Connections with task-oriented and open-

domain systems

: Similar to a task-oriented or an

open-domain scenario, social inﬂuence dialogue

can also be seen as a sequential decision mak-

ing process with the goal of maximizing the ex-

pected reward (Huang et al.,2020;Gao et al.,2018).

Our proposed category is not meant to be disjoint

from these traditional categories. However, it still

uniquely brings together the tasks that capture so-

cial inﬂuence, which is fundamentally absent from

how we primarily deﬁne dialogue tasks in the com-

munity. Deﬁning a new category that captures so-

cial inﬂuence dialogue would foster a dedicated

effort towards this important aspect of real-world

conversations.

Task-oriented scenarios focus on collaborative

information exchange for a common goal of task

completion. In social inﬂuence tasks, the goals

of the system and the user can be different and

even conﬂicting, leading to collaborative or non-

collaborative interactions. Further, the goals can

go beyond the current task (e.g. multiple therapy

interactions, repeated negotiations), leading to so-

cial interactions for long-term relationships. If a

scenario involves the system’s goal to inﬂuence its

partner, we consider it under social inﬂuence in this

paper. For instance, He et al. (2018) studied buyer-

seller price negotiations. The task of the buyer is to

negotiate for a reasonable price (arguably making it

task-oriented), but achieving it requires social inﬂu-

ence skills of engaging in trade-offs and building a

rapport with the seller so as to reach an agreement.

Measures of Success

: The above discussion indi-

cates that a comprehensive evaluation of social in-

ﬂuence systems must draw from both task-oriented

and open-domain dialogue research. Since there

exist surveys that discuss the evaluation in these set-

tings (Deriu et al.,2021;Li et al.,2021), we don’t

cover them here in detail. However, we deﬁne

three essential axes for evaluation: 1) Linguistic

Performance, or the system’s linguistic sophistica-

tion based on automatic (e.g. perplexity, BLEU)

and human (e.g. ﬂuency, consistency, coherency)

evaluation. 2) Inﬂuence Outcome, or the ability to

inﬂuence deﬁned by objective goals like the nego-

tiated price or weight loss after therapy. 3) Partner

Perception, or the subjective evaluation of the user,

for instance, the user’s satisfaction, likeness to-

wards the system, and interest in interacting again.

In a buyer-seller negotiation, if the seller hates the

buyer in the end, no matter how favorable the deal

is for the buyer, one might argue that this is still a

failed negotiation for the buyer. Hence, we encour-

age future work to take all three dimensions into

account collectively.

3 Social Inﬂuence Across Diverse

Application Areas

We now illustrate social inﬂuence across numer-

ous domains and application areas. In total, we

curated

datasets from prior work that capture

social inﬂuence in various forms, spanning

pub-

lication venues,

languages, and

application do-

mains (see Appendix Afor details on the compi-

lation process). In general, the datasets capture

the following information about an interaction: the

non-conversational context for the participants (e.g.

negotiation preferences or other role-speciﬁc infor-

mation), the conversation between them, and out-

come assessment. Optionally, some datasets also

gather participant demographics and personality

traits, utterance-level annotations, and subjective

evaluations via post-surveys.

To understand the structural similarities and dif-

ferences between these datasets, we design a tax-

onomy with two primary dimensions:

Task Struc-

ture

(Symmetric vs Asymmetric), and

Context Def-

inition

(Global vs Local).

Task Structure

cap-

tures whether the participant roles are deﬁned in a

symmetric or an asymmetric manner. For instance,

a typical multi-issue negotiation is symmetric, in

the sense that both parties have their own prefer-

ences and goals based on which they actively try

to reach a favorable agreement (Lewis et al.,2017).

On the other hand, a counseling session between

a therapist and a patient is asymmetric, where the

therapist attempts to emotionally support the pa-

tient by employing social inﬂuence skills (Althoff

et al.,2016).

Context Deﬁnition

relates to whether

the input context before each interaction is deﬁned

globally or locally. For instance, the Persuasion-

ForGood dataset globally deﬁnes the context of

persuasion for charity donation, which is kept the

same throughout (Wang et al.,2019). On the con-

trary, in a typical debate, although the rules are

deﬁned globally, the conversation topic and argu-

ments are local and can vary for each conversa-

tion (Durmus and Cardie,2019). We present this

categorization in Table 1. We further categorize the

datasets according to their

Domain

Source

, and

the

# of parties

. We provide key statistics and the

available metadata in Appendix B. We now brieﬂy

discuss the datasets in each domain.

Games

: Strategy games involve social inﬂuence

dynamics of trust and deception. Diplomacy cap-

tures deception in long-lasting relationships, where

players forge and break alliances to dominate Eu-

rope (Peskov et al.,2020). Catan revolves around

the trade of resources for acquiring roads, settle-

ments, and cities (Asher et al.,2016;Boritchev and

Amblard,2021). The players have access to only

a subset of resources that they would need, which

encourages strategic inﬂuence and trade.

Multi-Issue Bargaining Tasks (MIBT)

: MIBT is

a tractable closed-domain abstraction of a typical

negotiation (Fershtman,1990). It is based on a

ﬁxed set of issues each with a predeﬁned priority

for each player, which essentially governs the goals

of the players. If the priorities of the players align,

this leads to competitive negotiations, where each

party attempts to convince their partner with trade-

offs and persuasive arguments. If they don’t, this al-

lows cooperative interactions where the negotiators

try to ﬁnd optimal divisions that beneﬁt everyone.

DealOrNoDeal (Lewis et al.,2017) involves nego-

tiations over three issues: books,balls, and hats.

Other datasets deﬁne a more grounded scenario,

such as symmetric CaSiNo (Chawla et al.,2021b)

negotiations between two campsite neighbors and

Name (Citation) Domain Source Structure Context # of Parties

STAC (As16)Games Crowdsource Symmetric Global Multiparty

Diplomacy (Pe20)Games Crowdsource Asymmetric Global Multiparty

DinG (Bo21)Games University game night logs Symmetric Global Multiparty

Tabletop (De15)MIBT Face-to-face, Wizard-of-Oz Symmetric Local Bilateral

DealOrNoDeal (Le17)MIBT Crowdsource Symmetric Local Bilateral

CaSiNo (Ch21)MIBT Crowdsource Symmetric Local Bilateral

JobInterview (YaD21)MIBT Crowdsource Asymmetric Local Bilateral

PersuasionforGood (Wa19)Social Good Crowdsource Asymmetric Global Bilateral

CraigslistBargain (He18)E-commerce Crowdsource Asymmetric Local Bilateral

AntiScam (Li20)E-commerce Crowdsource Asymmetric Global Bilateral

MI (TaC16)Therapy & Support Psychotherapy session logs Asymmetric Global Bilateral

SMS Counseling (Al16)Therapy & Support SMS chat logs Asymmetric Global Bilateral

EmpatheticDialogues (Ra19)Therapy & Support Crowdsource Asymmetric Global Bilateral

Hotline Counseling (De19)Therapy & Support Synthetic Transcripts Asymmetric Global Bilateral

mPED (LiE21)Therapy & Support Physical activity clinical trials Asymmetric Global Bilateral

Congressional Debates (Th06)Argumentation U.S. Congressional transcripts Asymmetric Local Multiparty

Supreme Court (Da12)Argumentation Oyez.org transcripts Asymmetric Local Multiparty

DeCour (Fo12)Argumentation Italian court hearings Asymmetric Local Multiparty

ChangeMyView (Ta16)Argumentation Reddit Asymmetric Local Multiparty

DDO Debates (Du19)Argumentation debate.org logs Symmetric Local Bilateral

Court Debates (Ji20)Argumentation China Court transcripts Asymmetric Local Multiparty

Target-Guided (Ta19)Miscellaneous Crowdsource Symmetric Local Bilateral

Table 1: Categorization of social inﬂuence dialogue corpora. This list is non-exhaustive, and also covers the

datasets that have enabled research into various sub-tasks and analyses that can eventually be useful for dialogue

systems in respective domains. MIBT: Multi-Issue Bargaining Task. Key statistics and associated metadata are in

Appendix 3.

asymmetric JobInterview (Yamaguchi et al.,2021)

negotiations between recruiters and applicants.

Social Good

: Social inﬂuence is critical for social

good applications. The tactics must be person-

alized using knowledge that is both relevant and

appealing. PersuasionForGood (Wang et al.,2019)

involves asymmetric interactions led by a persuader

who attempts to convince the other participant for

charity donation by employing a variety of tactics.

For instance, Logical Appeal uses reason and ev-

idence to support the argument, while Emotional

Appeal elicits speciﬁc emotions.

E-commerce

: These tasks are typically asymmet-

ric. A buyer inﬂuences the seller towards a rea-

sonable price, while the seller tries to maximize

their own proﬁt. An effective system must combine

price-related reasoning with language realization.

CraigslistBargain (He et al.,2018) involves open-

ended price negotiations with rich inﬂuence strate-

gies like embellishments, side offers, emotional

appeals, and using world knowledge. Another ex-

ample is customer support interactions in AntiScam

dataset (Li et al.,2020), where users defend them-

selves against attackers who try to steal sensitive

personal information with convincing arguments.

Therapy & Support

: Effective therapy using so-

cial inﬂuence aids in the treatment of mental dis-

orders, and substance use disorders, along with

changing undesirable behaviors like unhealthy di-

ets. A counselor needs to be adaptive, personalized,

should understand the core issues, and should facil-

itate a change in patient’s perspective (Althoff et al.,

2016). In SMS counseling, Althoff et al. (2016)

found that linguistic inﬂuence like pushing the

conversation in the desired direction is associated

with perspective change. Similar scenarios were

captured in other datasets as well (Demasi et al.,

2019;Liang et al.,2021). Tanana et al. (2016) col-

lected the Motivational Interviewing dataset where

the goal is to elicit and explore the patient’s own

motivations for behavior change. EmpatheticDia-

logues (Rashkin et al.,2019) captured empathetic

support interactions, which has been associated

with rapport and better task outcomes (Kim et al.,

2004;Norfolk et al.,2007;Fraser et al.,2018).

Argumentation

: In addition to factuality and so-

cial proof, a convincing argument must also con-

sider the intensity, valence, authoritativeness, and

framing (Chaiken,1987;Althoff et al.,2014). Tan

et al. (2016) released the ChangeMyView logs from

Reddit, involving discussions on numerous contro-

versial topics. Other datasets include Debate Dot

Org (DDO) debates on diverse topics (Durmus and

Cardie,2019), congressional proceedings (Thomas

et al.,2006), and court hearings (Fornaciari and

Poesio,2012;D.-N.-M. et al.,2012;Ji et al.,2020).

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

SocialInuenceDialogueSystems:ASurveyofDatasetsandModelsForSocialInuenceTasksKushalChawla1WeiyanShi2JingwenZhang3GaleLucas1ZhouYu2JonathanGratch11UniversityofSouthernCalifornia2ColumbiaUniversity3UniversityofCaliforniaDavis1{chawla,lucas,gratch}@ict.usc.edu2{ws2634,zy2461}@columbia.edu3jw...

收起<<

Social Inﬂuence Dialogue Systems A Survey of Datasets and Models For Social Inﬂuence Tasks Kushal Chawla1Weiyan Shi2Jingwen Zhang3.pdf

共17页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Social Inﬂuence Dialogue Systems A Survey of Datasets and Models For Social Inﬂuence Tasks Kushal Chawla1Weiyan Shi2Jingwen Zhang3

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: