SemEval 2023 Task 9 Multilingual Tweet Intimacy Analysis

2025-04-15 1 0 904.05KB 5 页 10玖币

侵权投诉

SemEval 2023 Task 9: Multilingual Tweet Intimacy Analysis

Jiaxin Pei ♣Vítor Silva†Maarten Bos †Yozen Liu†

Leonardo Neves†David Jurgens♣Francesco Barbieri†

♣School of Information, University of Michigan, Ann Arbor, MI, USA

†Snap Inc, Santa Monica, CA, USA

♣{pedropei, jurgens}@umich.edu

†{fbarbieri, vsilvasousa, maarten, yliu2, lneves}@snap.com

Abstract

We propose MINT, a new Multilingual

intimacy analysis dataset covering 13,372

tweets in 10 languages including English,

French, Spanish, Italian, Portuguese, Korean,

Dutch, Chinese, Hindi, and Arabic. We bench-

marked a list of popular multilingual pre-

trained language models. The dataset is re-

leased along with the SemEval 2023 Task 9:

Multilingual Tweet Intimacy Analysis.

1 Introduction

Intimacy has long been viewed as a primary di-

mension of human relationships and interpersonal

interactions (Maslow,1981;Sullivan,2013;Prager,

1995). Existing studies suggest that intimacy is an

essential component of language and can be mod-

eled with computational methods (Pei and Jurgens,

2020). Textual intimacy is an important social as-

pect of language, and automatically analyzing it

can help to reveal important social norms in vari-

ous contexts (Pei and Jurgens,2020). Recognizing

intimacy can also serve as an important benchmark

to test the ability of computational models to under-

stand social information (Hovy and Yang,2021).

Despite the importance of intimacy in language,

resources on textual intimacy analysis remain rare.

Pei and Jurgens (2020) annotated the ﬁrst textual in-

timacy dataset containing 2,397 English questions

collected mostly from social media posts and ﬁc-

tional dialogues. However, phrases following the

question structure are used for interrogative situa-

tions, and models trained over it may not generalize

well over texts in other forms of languages.

To further promote computational modeling of

textual intimacy, we annotated a new multilingual

textual intimacy dataset named MINT. MINT cov-

ers tweets in 6 languages as the training data, in-

cluding English, Spanish, French, Portuguese, Ital-

ian, and Chinese, covering major languages used in

The Americas, Europe, and Asia. A total of 12,000

tweets are annotated for the 6 languages. To test

the model generalizability under the zero-shot set-

tings, we also annotated small test sets for Dutch,

Korean, Hindi, and Arabic (500 tweets for each).

We benchmarked a series of large multilingual

pre-trained language models including XLM-T

(Barbieri et al.,2021), XLM-R (Conneau et al.,

2019), BERT (Devlin et al.,2018), DistillBERT

(Sanh et al.,2019) and MiniLM (Wang et al.,2020).

We found that distilled models generally perform

worse than other normal models, while the XLM-

R model trained over the twitter dataset (XLM-

T) performs the best on 7 languages. While the

pre-trained language models are able to achieve

promising performance, zero-shot prediction of un-

seen languages remains challenging especially for

Korean and Hindi.

2 Data

We choose Twitter as the source of our dataset as

Twitter is a public media platform that naturally

includes multilingual text data, and from our analy-

sis, a fair amount of intimate texts. In this section,

we introduce the data collection and annotation

process for MINT.

2.1 Sampling

We use tweets sampled from 2018 to 2022. We

use the lang_id key in the tweet object to select

English and Chinese tweets. For other languages,

we use fastText (Joulin et al.,2016b,a) for language

identiﬁcation

and assign language labels when

the model conﬁdence is larger than 0.8. All the

mentions of unveriﬁed users are replaced with a

special token “@user” during pre-processing to

remove noise from random and very infrequent

usernames.

We ﬁne-tune XLM-T, a multilingual RoBERTa

model adapted to the Twitter domain (Barbieri

1https://fasttext.cc/docs/en/

language-identification.html

arXiv:2210.01108v2 [cs.CL] 3 Feb 2023

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：

SemEval2023Task9:MultilingualTweetIntimacyAnalysisJiaxinPei|VítorSilvayMaartenBosyYozenLiuyLeonardoNevesyDavidJurgens|FrancescoBarbieriy|SchoolofInformation,UniversityofMichigan,AnnArbor,MI,USAySnapInc,SantaMonica,CA,USA|{pedropei,jurgens}@umich.eduy{fbarbieri,vsilvasousa,maarten,yliu2,lneves}@snap....

展开>> 收起<<

SemEval 2023 Task 9 Multilingual Tweet Intimacy Analysis.pdf

共5页,预览1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！

SemEval 2023 Task 9 Multilingual Tweet Intimacy Analysis

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: