Zero-Shot Prompting for Implicit Intent Prediction and Recommendation with Commonsense Reasoning Hui-Chi Kuo Yun-Nung Chen

2025-04-29 0 0 488.74KB 8 页 10玖币

侵权投诉

Zero-Shot Prompting for Implicit Intent Prediction and Recommendation

with Commonsense Reasoning

Hui-Chi Kuo Yun-Nung Chen

National Taiwan University, Taipei, Taiwan

r09922a21@csie.ntu.edu.tw y.v.chen@ieee.org

Abstract

The current generation of intelligent assistants

require explicit user requests to perform tasks

or services, often leading to lengthy and com-

plex conversations. In contrast, human assis-

tants can infer multiple implicit intents from

utterances via their commonsense knowledge,

thereby simplifying interactions. To bridge this

gap, this paper proposes a framework for multi-

domain dialogue systems. This framework au-

tomatically infers implicit intents from user ut-

terances, and prompts a large pre-trained lan-

guage model to suggest suitable task-oriented

bots. By leveraging commonsense knowledge,

our framework recommends associated bots in

a zero-shot manner, enhancing interaction ef-

ﬁciency and effectiveness. This approach sub-

stantially reduces interaction complexity, seam-

lessly integrates various domains and tasks, and

represents a signiﬁcant step towards creating

more human-like intelligent assistants that can

reason about implicit intents, offering a supe-

rior user experience.1

1 Introduction

Intelligent assistants have become increasingly pop-

ular in recent years, but they require users to explic-

itly describe their tasks within a single domain. Yet,

the exploration of gradually guiding users through

individual task-oriented dialogues has been rela-

tively limited (Chiu et al.,2022). This limitation

is ampliﬁed when tasks extend across multiple do-

mains, compelling users to interact with numerous

bots to accomplish their goals (Sun et al.,2016).

For instance, planning a trip might involve inter-

acting with one bot for ﬂight booking and another

for hotel reservation, each requiring distinct, task-

speciﬁc intentions like “Book a ﬂight ticket” to

activate the corresponding bot, such as an airline

bot. In contrast, human assistants can manage high-

level intentions spanning multiple domains, utiliz-

1Code: http://github.com/MiuLab/ImplicitBot.

ing commonsense knowledge. This approach ren-

ders conversations more pragmatic and efﬁcient, re-

ducing the user’s need to deliberate over each task

separately. To overcome this limitation of current

intelligent assistants, we present a ﬂexible frame-

work capable of recommending task-oriented bots

within a multi-domain dialogue system, leveraging

commonsense-inferred implicit intents as depicted

in Figure 1.

Multi-Domain Realization Sun et al. (2016) pin-

pointed the challenges associated with a multi-

domain dialogue system, such as 1) comprehend-

ing single-app and multi-app language descriptions,

and 2) conveying task-level functionality to users.

They also gathered multi-app data to encourage

research in these directions. The HELPR frame-

work (Sun et al.,2017) was the pioneering attempt

to grasp users’ multi-app intentions and conse-

quently suggest appropriate individual apps. Nev-

ertheless, previous work focused on understanding

individual apps based on high-level descriptions

exclusively through user behaviors, necessitating a

massive accumulation of personalized data. Due to

the lack of paired data for training, our work lever-

ages external commonsense knowledge to bridge

the gap between high-level utterances and their

task-speciﬁc bots. This approach enables us to

consider a broad range of intents for improved gen-

eralizability and scalability.

Commonsense Reasoning Commonsense rea-

soning involves making assumptions about the

nature and essence of typical situations humans

encounter daily. These assumptions encompass

judgments about the attributes of physical objects,

taxonomic properties, and individuals’ intentions.

Existing commonsense knowledge graphs such as

ConceptNet (Bosselut et al.,2019), ATOMIC (Sap

et al.,2019), and TransOMCS (Zhang et al.,2021)

facilitate models to reason over human-annotated

commonsense knowledge. This paper utilizes a

arXiv:2210.05901v2 [cs.CL] 6 Jun 2023

I want to plan a trip to Seattle.

User High-Level Intention Utterance

Sure. I want to fly on July 1st and back on July 15th.

[AirlineBot] can book a flight ticket for you. Do you want to set the travel dates?

[HotelBot] can reserve a room for you. Do you want to specify your preference?

Here are the flights: …

Oh, yes. Search the 5-star hotel for 2 people at that period.

Here are the 5-star hotels with available rooms: …

AirlineBot (Flight Ticket Booking), HotelBot (Room Reservation)

Implicit Tasks/Intents

AirlineBot

HotelBot

Figure 1: Illustration of a multi-task dialogue example.

generative model trained on ATOMIC

(Hwang

et al.,2021) to predict potential intents linking

given user high-level utterances with corresponding

task-oriented bots. The inferred intents can activate

the relevant task-oriented bots and also serve as

justiﬁcation for recommendations, thereby enhanc-

ing explainability. This work is the ﬁrst attempt

to integrate external commonsense relations with

task-oriented dialogue systems.

Zero-Shot Prompting Recent research has re-

vealed that large language models (Radford et al.,

2019;Brown et al.,2020) have acquired an as-

tounding ability to perform few-shot tasks by us-

ing a natural-language prompt and a handful of

task demonstrations as input context (Brown et al.,

2020). Guiding the model with interventions via an

input can render many downstream tasks remark-

ably easier if those tasks can be naturally framed as

a cloze test problem through language models. As

a result, the technique of prompting, which trans-

poses tasks into a language model format, is in-

creasingly being adopted for different tasks (Zhao

et al.,2021;Schick and Schütze,2021). Without

available data for prompt engineering (Shin et al.,

2020), we exploit the potential of prompting for

bot recommendation in a zero-shot manner. This

strategy further extends the applicability of our

proposed framework and enables it to accommo-

date a wider variety of user intents and tasks, thus

contributing to a more versatile and efﬁcient multi-

domain dialogue system.

2 Framework

Figure 2illustrates our proposed two-stage frame-

work, which consists of: 1) a commonsense-

inferred intent generator, and 2) a zero-shot bot

recommender. Given a user’s high-level intention

utterance, the ﬁrst component focuses on gener-

ating implicit task-oriented intents. The second

component then utilizes these task-speciﬁc intents

to recommend appropriate task-oriented bots, con-

sidering the bots’ functionality through a large pre-

trained language model.

2.1

Commonsense-Inferred Intent Generation

The commonsense-inferred implicit intents func-

tion not only as prompts for bot recommendation

but also as rationales for the suggested bots, thereby

establishing a solid connection between the high-

level intention and task-oriented bots throughout

the conversation. For instance, the multi-domain

system shown in Figure 1recommends not only

the AirlineBot but also describes its functionality—

“can book a ﬂight ticket”—to better convince the

user about the recommendation.

2.1.1 Relation Trigger Selection

ATOMIC

is a commonsense knowledge graph

featuring commonsense relations across three cat-

egories: social-interaction, event-centered, and

physical-entity relations, all of which concern sit-

uations surrounding a speciﬁed event of interest.

Following Hwang et al. (2021), we employ a

BART model (Lewis et al.,2020) pre-trained on

ATOMIC

to generate related entities and events

based on the input sentence. However, despite hav-

ing a total of 23 commonsense relations, not all

are suitable for inferring implicit intents in assis-

tant scenarios. We utilize AppDialogue data (Sun

et al.,2016) to determine which commonsense re-

lations can better trigger the task-speciﬁc intents.

Given a high-level intention description

and its

task-speciﬁc sentences

sij

, we calculate the trig-

ger score of each relation

as an indicator of its

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：

Zero-ShotPromptingforImplicitIntentPredictionandRecommendationwithCommonsenseReasoningHui-ChiKuoYun-NungChenNationalTaiwanUniversity,Taipei,Taiwanr09922a21@csie.ntu.edu.twy.v.chen@ieee.orgAbstractThecurrentgenerationofintelligentassistantsrequireexplicituserrequeststoperformtasksorservices,oftenlead...

展开>> 收起<<

Zero-Shot Prompting for Implicit Intent Prediction and Recommendation with Commonsense Reasoning Hui-Chi Kuo Yun-Nung Chen.pdf

共8页,预览2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！

Zero-Shot Prompting for Implicit Intent Prediction and Recommendation with Commonsense Reasoning Hui-Chi Kuo Yun-Nung Chen

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: