arXiv:2211.06431v1 [cs.CY] 13 Nov 2022

Page 1

FinTech for Social Good:

A Research Agenda from NLP Perspective

Chung-Chi Chen,1 Hiroya Takamura,1 Hsin-Hsi Chen2

1 AIST, Japan

2 Department of Computer Science and Information Engineering

National Taiwan University, Taiwan

c.c.chen@acm.org, takamura.hiroya@aist.go.jp,

hhchen@ntu.edu.tw

Abstract

Making our research results positively impact

on society and environment is one of the goals

our community has been pursuing recently. Al-

though financial technology (FinTech) is one

of the popular application fields, we notice that

there is no discussion on how NLP can help

in FinTech for the social good. When men-

tioning FinTech for social good, people are

talking about financial inclusion and green fi-

nance. However, the role of NLP in these di-

rections only gets limited discussions. To fill

this gap, this paper shares our idea of how we

can use NLP in FinTech for social good. We

hope readers can rethink the relationship be-

tween finance and NLP based on our sharing,

and further join us in improving the financial

literacy of individual investors and improving

the supports for impact investment.

1 Introduction

Recently, our community is eager for the discus-

sions on AI for social good1 and NLP for social

good.2 Although we have several surveys and opin-

ion pieces from different aspects (Hovy and Spruit,

2016; Leins et al., 2020; Jin et al., 2021), there is

hardly any discussion on how NLP can help in Fin-

Tech for social good development. As the FinTech

industry matures, we think that it is time to discuss

the potential risk and what we can do to address

these issues with the help of NLP.

When mentioning FinTech for social good, there

are many discussions on how FinTech development

can improve financial inclusion (Ozili, 2021). Fi-

nancial inclusion is a notion to express that every-

one can get basic financial services and have an

opportunity to access financial products. With the

development of mobile banking and wireless inter-

net, more and more individuals in developing coun-

tries can access financial services such as cashless

1https://ijcai-22.org/

calls-ai-for-good/

2https://2021.aclweb.org/calls/papers/

payment, deposits, and insurance (Analytics, 2018).

With the development of online identity authenti-

cation, people in developed countries can easily

access financial products such as equity and op-

tions by opening an account within a few minutes.

Both are important achievements in the FinTech

industry.

However, we notice some attendant risks follow-

ing the improvement of financial inclusion. Based

on the statistics of the Taiwan Stock Exchange

(TWSE), people in their twenties occupy over 80%

of accounts opened. On the other hand, based on

the statistics of the Taiwan Academy of Banking

and Finance3, a government-supervised institution,

approximately 40% of people in their twenties are

financially illiterate. Even those who graduated

from a university, only 5% of them are consid-

ered to have a good sense to finance. Although

joining the financial market could make people

become wealthy, it may also lead to bankrupting

results. Given 80/20 Rule, i.e., 80% of individual

investors will only get profit and loss balance out-

come or deficit, which is supported in financial mar-

ket (Xiao, 2015), can we do something to protect

individual investors? One of the recent examples4

also indicates that making trading process becomes

easy and game-like may lead to negative outcomes

when the user does not have enough knowledge to

understand the risk of investment and the results

of these actions. Following this line of thought,

we want to discuss how we can improve financial

literacy with FinTech and NLP (FinNLP).

Taking the environment and society into consid-

eration when making daily decisions is one of the

tendencies for sustainable development. It is the

same in financial decision-making. Instead of only

considering return and risk, some people in the

3https://www.tabf.org.tw/English/

4https://edition.cnn.

com/2021/02/11/investing/

robinhood-lawsuit-suicide-alex-kearns/

index.html

arXiv:2211.06431v1 [cs.CY] 13 Nov 2022

Page 2

financial domain advocate adding social and envi-

ronmental impact to the decision-making process.

To be the bridge between the financial field and

the FinNLP community, we also make discussions

on how FinNLP can support the development of

impact investment.

In the rest of this paper, we will first discuss cur-

rent FinNLP research and the role of these studies

in FinTech development in Section 2. In Section 3,

we propose two directions, investor education and

impact investment, for FinNLP for social good.

Some challenges and possible solutions will be pro-

vided in Section 4. Finally, we conclude this paper

in Section 5. Our intention is to propose some dif-

ferent directions, which got few discussions before,

to inspire researchers to figure out some ways for

FinNLP for social good together.

2 Current FinNLP Research

FinTech is an emerging term from 2015, and the

development of FinNLP has three major directions:

improving the working processes in financial insti-

tutions, predicting market information, and provid-

ing automatic financial services to customers (Chen

et al., 2020). Improving the working processes can

enlarge the capability of financial institutions to

serve more people, and providing automatic finan-

cial services can improve the customers’ experi-

ence. Both are the key forces in achieving financial

inclusion.

From the aspect of improving working processes,

one of the widely discussed topics is identity fraud

detection because it is a crucial issue when peo-

ple use services remotely. One ideal solution is

to ask some new questions that could not be an-

swered based on user-provided data. Wang et al.

(2019) share the experimental results of this idea,

and generate questions based on extended personal

knowledge graphs to improve performance.

Forecasting market information is the most popu-

lar direction in FinNLP development. Stock move-

ment (Xu and Cohen, 2018; Tang et al., 2021),

volatility (Rekabsaz et al., 2017; Dereli and Sar-

aclar, 2019; Qin and Yang, 2019; Chen et al.,

2021b), sales (Lin et al., 2019), bubble (Sawhney

et al., 2022), and several different kinds of mar-

ket information are targets of market information

forecasting tasks. In addition to constructing end-

to-end models for prediction, there are some other

discussions on how to fool NLP-based market pre-

diction models (Xie et al., 2022) and the problems

of pre-trained language models in market informa-

tion forecasting applications (Chuang and Yang,

2022).

Instead of giving predictions directly, leveraging

NLP techniques to provide customers information

for them to make the final decision is a human-

centered approach. Some systems are proposed for

this purpose. Liou et al. (2021) propose a system

to help investors and journalists identify the related

companies based on a given news article. Hassan-

zadeh et al. (2022) present a comprehensive system

for news retrieval, event identification, causal anal-

ysis, and causal knowledge graph extraction. Users

can make decisions based on the outputs of these

systems.

3 FinNLP for Social Good

Thanks to the effort of researchers in the FinNLP

community, models can better understand finan-

cial documents than before, and some application

scenarios can be implemented for the customers

of financial institutions. However, how to make

social good based on our findings has been hardly

discussed so far. In this section, we propose two di-

rections: investor education and impact investment.

Investor education is a way to improve financial

literacy, which is one of the world-level projects

proposed by the Organisation for Economic Co-

operation and Development (OECD)5. Impact in-

vestment is a concept that adds one more dimension

—impact on the environment and society— into con-

sideration when making investment decisions. The

goal of these two directions is to construct a sus-

tainable financial environment. In this section, we

provide the link between these topics and FinNLP.

3.1 Investor Education

Currently, investor education is provided passively.

For example, the United States Securities and Ex-

change Commission (SEC) provides many teach-

ing materials for investors,6 and TWSE also makes

many efforts to prepare materials for investor edu-

cation7. We think that NLP provides an opportunity

to change from passive to active. Instead of asking

investors to take a look at the materials, we can

score their investment ideas and provide feedback

5https://www.oecd.org/finance/

financial-education/

6https://www.sec.gov/education/

investor-education

7https://investoredu.twse.com.tw/

Pages/TWSE.aspx

Page 3

on their ideas. Thus, for investor education, we pro-

pose two directions: scoring investment reasons,

and generating risk reminders.

Scoring opinions is not a new topic in our

community. For example, scoring persuasive es-

says (Ghosh et al., 2016) and evaluating the help-

fulness of product reviews (Ocampo Diaz and Ng,

2018) have been discussed for a long time. How-

ever, scoring investment opinions is still in the early

stage. Ying and Duboue (2019) provide a pilot

exploration on scoring investment reasons by an-

notating 2,622 rationales into four levels. Chen

et al. (2021a) evaluate amateurs’ opinions based

on whether they share some characteristics with

experts’ opinions. These studies also point out two

different goals of scoring investment opinions: (1)

scoring rationality and (2) using profitability as the

score.

Since there are too many uncertainty factors in

the financial market, it is hard to say whether the

reasonable analysis will lead to profitable outcomes.

Even professionals’ analysis may sometimes be

inaccurate (Zong et al., 2020). Therefore, the goal

of scoring investment reasons is different from that

of making predictions. We aim to inform investors

whether their reasons can support their decision

based on financial theories. Additionally, we can

let investors know to what extent it will lead to

a profitable outcome based on the learned results

with historical data.

However, only providing the score is not enough

for investors to learn from models’ output. Provid-

ing feedback is more useful than just letting users

know the score. For example, grammatical error

correction (Xie et al., 2018) is one of the common

next steps after scoring essays. Providing feedback

to the grammatical error (Nagata, 2019) is a more

challenging task. In investor education, we think

that providing risk reminders to the given invest-

ment reasons is more important than just providing

scores. For example, when users input their reasons

as the follows: 8

Key takeaways from Quanta’s 1Q22 ana-

lyst call included: 1) 1Q22 GMs dipped

to an eight-quarter low due to inferior

product mix and inefficient production;

2) weak 2Q22 earnings outlook with

both PC demand weakness and supply

8Both opinion and risk reminders parts are selected from

J.P. Morgan Asia Pacific Equity Research’s report on 13 May

2022.

constraints; 3) 2Q22 NB shipment guid-

ance at 20% qoq declines missed mar-

ket expectations, implying likely m/s

loss towards Hon Hai, in our view; 4)

Server order book remains intact with

double-digit growth target this year; 5)

full-year capex at NT$10bn with over-

seas expansion (Thailand, North Amer-

ica); and 6) the board approved a NT$6.6

DPS, implying 8% yields at current stock

prices. Quanta stock has moved largely

in line with Taiex YTD, while we expect

the earnings downgrade (first time since

2019) to weigh on the stock price in the

next six months. We cut 2022/23 earn-

ings by 15%/7% and lower our Dec-22

PT to NT$77. Stay Neutral.

In addition to providing scores, we think that it

will be more meaningful if models can generate the

risk reminders to this neutral opinion as follows:

Key downside risks include margin ero-

sion in servers and a potential slowdown

of PC demand post COVID-19.

Key upside risks include 1) a longer-

than-expected server upcycle; and 2)

stronger- than-expected MacBook mar-

ket share gains offset consumer PC weak-

ness.

This kind of risk reminder is different from the

outputs of end-to-end market movement prediction

models. It provides some scenarios that may make

the neutral claim become inaccurate. We think that

it is an ideal way to provide feedback on financial

opinions. In this way, investors can take the gener-

ated risk into consideration when making decisions.

It will also help investors consider their investment

reasons more comprehensively.

3.2 Impact Investment

Beyond considering return and risk in the invest-

ment, the impact on the environment and society

has been widely discussed and suggested to add to

the financial decision-making process (Emerson,

2012; Chiappini, 2017). More and more concrete

ideas for assessment and auditing are proposed.

Recently, one of the most popular guidelines is

SASB Standards, proposed by Sustainability Ac-

counting Standards Board.9 This standard sorts

9https://www.sasb.org/

Page 4

out the non-monetary issues related to environmen-

tal, social, and governance (ESG) that may influ-

ence the financial performance of companies. It

is intuitive that NLP can help in capturing ESG-

related terms (Kang et al., 2021) or tracking se-

mantic change in companies’ reports (Purver et al.,

2022). However, can FinNLP help more on impact

investment? We propose two directions for this

question: non-monetary opportunity/risk identifica-

tion and scenario planning.

As we shown in Section 2, many studies pay

attention to monetary return or volatility prediction.

It can be based on news articles, social media, pro-

fessional reports, meeting between investors and

company managers, and so on. However, limited

studies try to identify the opportunity and risk from

non-monetary aspects, i.e., ESG aspect. Since the

impact period of ESG-related events may be longer

than that of traditional monetary events, such as

the increase of earning, decrease of sales, etc, we

think that one of the important directions is to iden-

tify whether the given event is the opportunity or

risk to ESG and how it will influence companies’

sustainable operation. As shown in Li et al. (2022),

simple statistical methods perform better than neu-

ral network models in multivariate long sequence

time-series forecasting. That raises an open is-

sue: whether their claim holds true when we at-

tempt to evaluate long-term non-monetary impacts?

Since many market information forecasting studies

mainly pay attention to short-term prediction (e.g.,

next day, 3-day, and 1 week), we want to encourage

more studies to pay attention to long-term value

assessment.

In addition to identifying the opportunity and

risk, we think that scenario planning (analysis) is

one of the key directions that FinNLP can help in

impact investment. The goal of scenario planning

is to understand the possible impact of the event

better rather than to predict the future. For example,

in addition to identifying whether it is a risk to the

agricultural industry when a piece of news related

to “climate anomalies” is given, we expect models

to generate some plausible scenarios such as “it

would lead to lack of water”, and this scenario can

be extended to “it may influence the agricultural

industry”.

Besides the most possible scenario, the worst-

case scenario should also be generated for discus-

sion. That is, when performing scenario planning,

experts first generate many scenarios regardless of

the probability of occurrence. And then, experts

will select a few key scenarios for discussions and

try to figure out some solutions to these key scenar-

ios. Because scenario planning plays an important

role in both monetary and non-monetary based fi-

nancial decision, we think that it is a good direction

to explore for providing better help to both individ-

ual and industrial investors.

4 Challenges and Opportunities

Because we just start in the proposed directions,

there are many open research questions and oppor-

tunities. In this section, we highlight three specific

topics.

4.1 Evaluation

One of the conveniences of market information

forecasting tasks is that we can use market infor-

mation as ground truth. It is open access and can

be quickly obtained. However, when exploring the

proposed risk reminder and scenario planning tasks,

we should not just focus on the final outcome in

the market. For example, the key downside/upside

risks shown in Section 3.1 may not happen finally,

but these reminders are important for leading in-

vestors to have careful consideration before mak-

ing decisions. In other words, previous studies are

precision-focused studies, but the proposed direc-

tions aim to highlight the importance of covering

as more as crucial factors as possible. Human eval-

uation is an intuitive direction. However, how to

automatically evaluate the generated scenarios is

still an open issue.

4.2 Implementation

It is hard to imagine that an analyst says, “I have

read a lot of news from 2008 to 2010, analysis

reports from 2003 to 2012, cooperate filing from

1994 to 2019, and earnings call transcripts from

2004 to 2019. Based on these experiences and

yesterday’s Made by Google product launch, I

think Pixel Watch would be the hottest product

in 2022.” However, we are now using pretrained

FinBERTs (Araci, 2019; Huang et al., 2022) in

this way. There is no doubt that some common

senses and word meanings can be learned via pre-

trained processes. However, the financial market

changes all the time, and there is a bulk of infor-

mation daily. How to add the latest knowledge

and news into models becomes a challenge when

we want to score investment reasons and generate

Page 5

risk reminders. One of the possible ways is using

the knowledge in textual book and formal material

to construct knowledge graph (KG) as the base of

background knowledge. And then, we can further

link this KG with the temporal-aware KG (Goel

et al., 2020), which tracks the change of latest in-

formation.

4.3 Utilization

Currently, the FinTech industry provides more con-

venient interfaces to investors, and also makes fi-

nancial information become transparent. Many

trading bots have been constructed to displace indi-

vidual investors when making trading decisions. In

this paper, we propose a human-centered AI con-

cept with FinNLP, which aims to empower human

abilities instead of replacing human. We think that

the proposed directions for investor education are

the next step for improving the investment envi-

ronment of individual investors. There are many

open research questions when exploring the pro-

posed directions. To what extent state-of-the-art

market information prediction models can help in

scoring investment reasons? What is the difference

between general storytelling tasks and financial

scenario generation? We also think that investor ed-

ucation and current FinNLP studies are reciprocal.

For example, scoring rationales can help us filter

out low-quality opinions for opinion-based market

information prediction models. It would be helpful

in improving the performance of existing tasks.

5 Conclusion

In this paper, we point out the risk behind the recent

improvement in financial inclusion, and share a re-

search agenda from FinNLP for social good aspect.

We hope this paper can lead readers to rethink the

role of NLP in the financial field. Different from re-

cent FinNLP studies, which focus on (1) extracting

cooperation reports, analyst reports, news articles,

and so on, and (2) using the extracted information

for constructing economics/financial indicators or

end-to-end market information prediction models,

we propose two directions that could help improve

financial literacy and augment investors’ ability in

making impact-oriented financial decisions. We are

not saying the proposed directions are completely

new, but we want to guide our community to con-

sider to what extent recent findings can be applied

to these directions and what we can do to achieve

these goals. Our intent is that this paper can set

an example for others to follow, and future works

can further propose more ideas and solutions for

FinNLP for social good.

References

Deep Knowledge Analytics. 2018. Financial inclusion

developing world landscape overview. Recuperado

de http://analytics. dkv. global/data/pdf/Financial-

Inclusion-Developing-World. pdf.

Dogu Araci. 2019. Finbert: Financial sentiment analy-

sis with pre-trained language models. arXiv preprint

arXiv:1908.10063.

Chung-Chi Chen, Hen-Hsen Huang, and Hsin-

Hsi Chen. 2020.

NLP in FinTech Applica-

tions: Past, Present and Future. arXiv preprint

arXiv:2005.01320.

Chung-Chi Chen, Hen-Hsen Huang, and Hsin-Hsi

Chen. 2021a. Evaluating the rationales of amateur

investors. In Proceedings of the Web Conference

2021, pages 3987–3998.

Chung-Chi Chen, Hen-Hsen Huang, Yu-Lieh Huang,

and Hsin-Hsi Chen. 2021b. Distilling numeral in-

formation for volatility forecasting. In Proceedings

of the 30th ACM International Conference on Infor-

mation Knowledge Management, page 2920–2924,

New York, NY, USA. Association for Computing

Machinery.

Helen Chiappini. 2017. Social impact funds: Defini-

tion, assessment and performance. Springer.

Chengyu Chuang and Yi Yang. 2022. Buy tesla, sell

ford: Assessing implicit stock market preference in

pre-trained language models. In Proceedings of the

60th Annual Meeting of the Association for Compu-

tational Linguistics (Volume 2: Short Papers), pages

100–105, Dublin, Ireland. Association for Computa-

tional Linguistics.

Nesat Dereli and Murat Saraclar. 2019. Convolutional

neural networks for financial text regression. In Pro-

ceedings of the 57th Annual Meeting of the Asso-

ciation for Computational Linguistics: Student Re-

search Workshop, pages 331–337, Florence, Italy.

Association for Computational Linguistics.

Jed Emerson. 2012. Risk, return and impact: Under-

standing diversification and performance within an

impact investing portfolio. Impact Assets, 2:1–15.

Debanjan Ghosh, Aquila Khanam, Yubo Han, and

Smaranda Muresan. 2016. Coarse-grained argumen-

tation features for scoring persuasive essays. In Pro-

ceedings of the 54th Annual Meeting of the Associa-

tion for Computational Linguistics (Volume 2: Short

Papers), pages 549–554, Berlin, Germany. Associa-

tion for Computational Linguistics.

Page 6

Rishab Goel, Seyed Mehran Kazemi, Marcus Brubaker,

and Pascal Poupart. 2020. Diachronic embedding

for temporal knowledge graph completion. In Pro-

ceedings of the AAAI Conference on Artificial Intel-

ligence, volume 34, pages 3988–3995.

Oktie Hassanzadeh, Parul Awasthy, Ken Barker, Onkar

Bhardwaj, Debarun Bhattacharjya, Mark Feblowitz,

Lee Martie, Jian Ni, Kavitha Srinivas, and Lucy Yip.

2022. Knowledge-based news event analysis & fore-

casting toolkit. In International Joint Conference on

Artificial Intelligence.

Dirk Hovy and Shannon L Spruit. 2016. The social

impact of natural language processing. In Proceed-

ings of the 54th Annual Meeting of the Association

for Computational Linguistics (Volume 2: Short Pa-

pers), pages 591–598.

Allen H Huang, Hui Wang, and Yi Yang. 2022. Fin-

bert: A large language model for extracting informa-

tion from financial text. Contemporary Accounting

Research.

Zhijing Jin, Geeticka Chauhan, Brian Tse, Mrinmaya

Sachan, and Rada Mihalcea. 2021. How good is

nlp? a sober look at nlp tasks through the lens of so-

cial impact. In Findings of the Association for Com-

putational Linguistics: ACL-IJCNLP 2021, pages

3099–3113.

Juyeon Kang, Ismail El Maarouf, Sandra Bellato, and

Mei Gan. 2021. FinSim-3: The 3rd shared task on

learning semantic similarities for the financial do-

main. In Proceedings of the Third Workshop on Fi-

nancial Technology and Natural Language Process-

ing, pages 31–35, Online. -.

Kobi Leins, Jey Han Lau, and Timothy Baldwin. 2020.

Give me convenience and give her death: Who

should decide what uses of nlp are appropriate, and

on what basis? In Proceedings of the 58th Annual

Meeting of the Association for Computational Lin-

guistics, pages 2908–2913.

Hao Li, Jie Shao, Kewen Liao, and Mingjian Tang.

2022. Do simpler statistical methods perform bet-

ter in multivariate long sequence time-series fore-

casting? In Proceedings of the 31st ACM Interna-

tional Conference on Information Knowledge Man-

agement, CIKM ’22, page 4168–4172, New York,

NY, USA. Association for Computing Machinery.

Zhaojiang Lin, Andrea Madotto, Genta Indra Winata,

Zihan Liu, Yan Xu, Cong Gao, and Pascale Fung.

2019. Learning to learn sales prediction with so-

cial media sentiment. In Proceedings of the First

Workshop on Financial Technology and Natural Lan-

guage Processing, pages 47–53, Macao, China.

Yi-Ting Liou, Chung-Chi Chen, Tsun-Hsien Tang,

Hen-Hsen Huang, and Hsin-Hsi Chen. 2021. Fin-

sense: an assistant system for financial journalists

and investors. In Proceedings of the 14th ACM Inter-

national Conference on Web Search and Data Min-

ing, pages 882–885.

Ryo Nagata. 2019. Toward a task of feedback comment

generation for writing learning. In Proceedings of

the 2019 Conference on Empirical Methods in Nat-

ural Language Processing and the 9th International

Joint Conference on Natural Language Processing

(EMNLP-IJCNLP), pages 3206–3215, Hong Kong,

China. Association for Computational Linguistics.

Gerardo Ocampo Diaz and Vincent Ng. 2018. Mod-

eling and prediction of online product review help-

fulness: A survey. In Proceedings of the 56th An-

nual Meeting of the Association for Computational

Linguistics (Volume 1: Long Papers), pages 698–

708, Melbourne, Australia. Association for Compu-

tational Linguistics.

Peterson K Ozili. 2021. Financial inclusion research

around the world: A review. In Forum for social eco-

nomics, volume 50, pages 457–479. Taylor & Fran-

cis.

Matthew Purver, Matej Martinc, Riste Ichev, Igor

Loncarski, Katarina Sitar Šuštar, Aljoša Valentincic,

and Senja Pollak. 2022. Tracking changes in ESG

representation: Initial investigations in UK annual

reports. In Proceedings of the First Computing So-

cial Responsibility Workshop within the 13th Lan-

guage Resources and Evaluation Conference, pages

9–14, Marseille, France. European Language Re-

sources Association.

Yu Qin and Yi Yang. 2019. What you say and how you

say it matters: Predicting stock volatility using ver-

bal and vocal cues. In Proceedings of the 57th An-

nual Meeting of the Association for Computational

Linguistics, pages 390–401, Florence, Italy. Associ-

ation for Computational Linguistics.

Navid Rekabsaz, Mihai Lupu, Artem Baklanov,

Alexander Dür, Linda Andersson, and Allan Han-

bury. 2017. Volatility prediction using financial dis-

closures sentiments with word embedding-based IR

models. In Proceedings of the 55th Annual Meet-

ing of the Association for Computational Linguistics

(Volume 1: Long Papers), pages 1712–1721, Van-

couver, Canada. Association for Computational Lin-

guistics.

Ramit Sawhney, Shivam Agarwal, Vivek Mittal, Paolo

Rosso, Vikram Nanda, and Sudheer Chava. 2022.

Cryptocurrency bubble detection: A new stock mar-

ket dataset, financial task & hyperbolic models. In

Proceedings of the 2022 Conference of the North

American Chapter of the Association for Computa-

tional Linguistics: Human Language Technologies,

pages 5531–5545, Seattle, United States. Associa-

tion for Computational Linguistics.

Tsun-Hsien Tang, Chung-Chi Chen, Hen-Hsen Huang,

and Hsin-Hsi Chen. 2021. Retrieving implicit infor-

mation for stock movement prediction. In Proceed-

ings of the 44th International ACM SIGIR Confer-

ence on Research and Development in Information

Retrieval, pages 2010–2014.

Page 7

Weikang Wang, Jiajun Zhang, Qian Li, Chengqing

Zong, and Zhifei Li. 2019. Are you for real? detect-

ing identity fraud via dialogue interactions. In Pro-

ceedings of the 2019 Conference on Empirical Meth-

ods in Natural Language Processing and the 9th In-

ternational Joint Conference on Natural Language

Processing (EMNLP-IJCNLP), pages 1762–1771.

Wei Xiao. 2015. Does practice make perfect? evi-

dence from individual investors’ experiences and in-

vestment returns. Journal of Interdisciplinary Math-

ematics, 18(6):811–825.

Yong Xie, Dakuo Wang, Pin-Yu Chen, Jinjun Xiong,

Sijia Liu, and Oluwasanmi Koyejo. 2022. A word

is worth a thousand dollars: Adversarial attack on

tweets fools stock prediction. In Proceedings of the

2022 Conference of the North American Chapter of

the Association for Computational Linguistics: Hu-

man Language Technologies, pages 587–599, Seat-

tle, United States. Association for Computational

Linguistics.

Ziang Xie, Guillaume Genthial, Stanley Xie, Andrew

Ng, and Dan Jurafsky. 2018. Noising and denoising

natural language: Diverse backtranslation for gram-

mar correction. In Proceedings of the 2018 Confer-

ence of the North American Chapter of the Associ-

ation for Computational Linguistics: Human Lan-

guage Technologies, Volume 1 (Long Papers), pages

619–628, New Orleans, Louisiana. Association for

Computational Linguistics.

Yumo Xu and Shay B. Cohen. 2018. Stock move-

ment prediction from tweets and historical prices. In

Proceedings of the 56th Annual Meeting of the As-

sociation for Computational Linguistics (Volume 1:

Long Papers), pages 1970–1979, Melbourne, Aus-

tralia. Association for Computational Linguistics.

Annie Ying and Pablo Duboue. 2019. Rationale clas-

sification for educational trading platforms. In Pro-

ceedings of the First Workshop on Financial Technol-

ogy and Natural Language Processing, pages 14–20,

Macao, China.

Shi Zong, Alan Ritter, and Eduard Hovy. 2020. Mea-

suring forecasting skill from text. In Proceedings

of the 58th Annual Meeting of the Association for

Computational Linguistics, pages 5317–5331, On-

line. Association for Computational Linguistics.