Noname manuscript No.

(will be inserted by the editor)

Data science and AI in FinTech: An overview

Longbing Cao · Qiang
Yang · Philip S. Yu

Received: date / Accepted: date

Abstract Financial technology (FinTech) has been playing an increasingly critical role in driving modern
economies, society, technology, and many other areas. Smart FinTech is the new-generation FinTech, largely
inspired and empowered by data science and new-generation AI and (DSAI) techniques. Smart FinTech synthe-
sizes broad DSAI and transforms finance and economies to drive intelligent, automated, whole-of-business and
personalized economic and financial businesses, services and systems. The research on data science and AI in
FinTech involves many latest progress made in smart FinTech for BankingTech, TradeTech, LendTech, InsurTech,
WealthTech, PayTech, RiskTech, cryptocurrencies, and blockchain, and the DSAI techniques including complex
system methods, quantitative methods, intelligent interactions, recognition and responses, data analytics, deep
learning, federated learning, privacy-preserving processing, augmentation, optimization, and system intelligence
enhancement. Here, we present a highly dense research overview of smart financial businesses and their challenges,
the smart FinTech ecosystem, the DSAI techniques to enable smart FinTech, and some research directions of smart
FinTech futures to the DSAI communities.
Keywords Financial technology · FinTech · Financial service · Smart FinTech · Finance · Economics ·
Blockchain · Data science · AI · Intelligent systems · Machine learning · Deep learning · Federated learning ·
Privacy-preserving · Modeling · Mathematics · Statistics · Ethics

1 Introduction tional intelligence, event and behavior analysis, social

media/network analysis; and (3) more recent advances
In recent years, finance has become increasingly interac- including deep learning [22], automated interactions,
tive with new-generation data science and artificial in- learning and responses, transfer learning [47], federated
telligence (DSAI) techniques [2, 15, 7, 27, 48, 49]. In par- learning [57, 56], human-centered computing, and brain
ticular, FinTech (or Fintech) [13, 44] is at the epicen- and cognitive computing [42]. Other fundamental areas
tre of synthesizing, innovating and transforming finan- such as statistical modeling and mathematical modeling
cial services, economy, technology, media, communica- also play a critical role in enabling FinTech. In contrast,
tion, and society broadly driven by DSAI techniques [8]. our reference to finance includes areas such as capital
Here, DSAI broadly refers to (1) classic AI areas in- market, trading, banking, insurance, leading/loan, in-
cluding logic, planning, knowledge representation, mod- vestment, wealth management, risk management, mar-
eling, autonomous systems, multiagent systems, com- keting, compliance and regulation, payment, contract,
plexity science, expert system (ES), decision support auditing, accounting, and digital currencies and their
system (DSS), optimization, simulation, pattern recog- supporting infrastructure (including blockchain [45]),
nition, image processing, and natural language process- operations, services, management, security, and ethics.
ing (NLP); (2) advanced DSAI areas including intelli- Economics and finance (EcoFin) are increasingly syn-
gent interactions and conversations, intelligent identifi- ergized with FinTech and DSAI.
cation and authentication, privacy-preserving process- DSAI is the keystone enabler of the new genera-
ing [19], advanced representation learning, advanced tion of EcoFin and FinTech [15, 28, 35, 39]. The new-
analytics and learning, knowledge discovery, computa- generation DSAI is reshaping or even redefining the
2

concepts, objectives, content and tasks of EcoFin and – economic-financial innovations: e.g., new mecha-
FinTech. DSAI essentially and comprehensively trans- nisms and products;
forms the way that modern economic and financial – economic-financial markets: including products and
(economic-financial) businesses operate, transact, inter- services;
act and collaborate with their participants (incl. con- – economic-financial participants: including individ-
sumers, markets, and regulators) and environments. ual and retail investors, institutions and regulators;
DSAI nurtures new economic-financial mechanisms, – economic-financial behaviors: e.g., investor activi-
models, products, services, and many tangible and in- ties and company announcements;
tangible opportunities. As a result, DSAI not only – economic-financial events: e.g., company mergers
strengthens the efficiency, cost-effectiveness, customer and financial crises;
experience, risk mitigation, regulation, and security – economic-financial services: e.g., banking, insur-
of existing economic-financial systems, it also inno- ance, lending, financing, and crowdfunding services;
vates unprecedented and more intelligent, efficient, con- – economic-financial mechanisms: e.g., market mech-
venient, personalized, explainable, secure and proac- anisms, business models, and derivative pricing;
tive economic-financial products and services. The syn- – economic-financial systems: e.g., a company’s fi-
thetic product of DSAI and EcoFin forms the new nance;
area of smart FinTech, as shown in Fig. 2 in [8], – economic-financial infrastructure: including funda-
which presents a four-dimensional, systematic and in- mental support systems and blockchain;
teractive landscape of the synthesis between DSAI and – economic-financial pricing: including the valuation
EcoFin in forming smart FinTech. The landscape con- of underlying and derivative assets in capital mar-
nects the main EcoFin businesses to the EcoFin data kets;
and repositories, the broad-based DSAI techniques, and – economic-financial trading: including trading, in-
the EcoFin business objectives. Accordingly, the fam- vestment, and execution;
ily of FinTech has expanded from BankingTech, Trade- – economic-financial payment: e.g., online, mobile and
Tech, LendTech, InsurTech, WealthTech and PayTech contactless payment;
to RiskTech, etc. – economic-financial valuation: e.g., the estimation of
In the rest of this paper, we discuss and summa- property values, credit and intangible assets;
rize the main business domains and their challenges, – economic-financial marketing: including campaign,
the ecosystems of smart FinTech which substantially and customer care;
expand the above FinTech family, and the DSAI tech- – economic-financial relationship management: e.g.,
niques driving smart EcoFin businesses and FinTech. A stakeholder relations, and business partnerships;
brief introduction to this selected topic on data science – economic-financial resource management: including
and AI in FinTech is then given, followed by discussion the management of human, material, informational
on future directions. and intangible assets;
– economic-financial operations: e.g., the processes
and services for supporting financial innovation, de-
2 FinTech businesses and challenges sign and production;
– economic-financial compliance and regulation: e.g.,
Below, we briefly summarize the main businesses and the enforcement of operational orders and business
challenges in smart FinTech1. Typical applications of rules by the authorities;
smart FinTech include Internet banking, mobile pay- – economic-financial crisis, risk and security: e.g., in-
ments, online shopping, peer-to-peer leading, online vestment risk, systemic risk, and cybersecurity; and
crowdfunding projects, cryptocurrency, cross-market – economic-financial ethics: including social and ethi-
portfolio management, and global supply chain man- cal issues and privacy.
agement. FinTech challenges. Major challenges associated
FinTech business areas. Building on DSAI tech- with the above FinTech business areas as well as their
niques, these areas involve almost all aspects of an associated EcoFin businesses, problems, data and ob-
economic-financial system and its environment and jectives can be categorized into:
broadly all EcoFin businesses [8, 9, 2, 39]. Here, we high- – innovation challenges: e.g., DSAI techniques for in-
light the following major business areas in smart Fin- venting novel, efficient, intelligent and sustainable
Tech: mechanisms, products, services and platforms;
1
Interested readers can refer to a comprehensive review on – business complexities: such as DSAI techniques for
AI in finance in [8, 9]. representing, learning and managing the intricate
 3

working mechanisms, structures, interactions, rela-

tions, hierarchy, scale, dynamics, anomaly, uncer-
tainty, emergence and exceptions associated with a
market, a product or a participant;
– organizational and operational complexities: such
as DSAI techniques for understanding and manag-
ing the diversity and personalization of individuals
and departmental teams, departmental and insti-
tutional coherence and consensus, and inconsistent
and volatile efficiency and performance;
– human and social complexities: such as DSAI tech-
niques for modeling and managing the diversity
and inconsistency of a participant’s cognitive, emo-
tional and technical capabilities and performance,
and for enabling effective communications, cooper-
ation and collaboration within a department and
between stakeholders;
– environmental complexities: such as DSAI tech-
niques for modeling and managing the interactions
with contextual and environmental factors and sys-
tems and their influence on a target business system
and problem;
– regional and global challenges: such as understand-
ing and managing the relations between an economy
entity and its financial systems with the related re-
Fig. 1 The Smart FinTech ecosystem.
gional and global counterparts and stakeholders and
their influence on target problems;
– data complexities: such as extracting, representing, ecosystems of smart FinTech can be viewed in terms
analyzing and managing data quality issues, mis- of their (1) comprehensive FinTech-driven businesses
information and complicated data characteristics, and areas, and (2) processes, functions and activities.
e.g., uncertainty and extreme dimensionality, spar- On one hand, FinTech-driven businesses and areas are
sity and skewness; overwhelming and evolving. Fig. 1 overviews the smart
– dynamic complexities: such as modeling, predicting FinTech ecosystems and categorizes them as follows:
and managing evolving but nonstationary behav- smart banking, smart insurance, smart lending, smart
iors, events and activities of individual and block trading, smart assets and wealth management, smart
markets, products, services and participants; and payments, smart credit and risk management, smart
– integrative complexities: e.g., systematically model- Internet finance, smart marketing and services, smart
ing and managing the various aspects of the above currencies and blockchain, smart compliance and reg-
complexities that are often tightly and loosely cou- ulation, smart ethics and security, and smart innova-
pled with each other in an underlying economic- tions. Each of these primary areas is further defined
financial system. by their respective businesses and supporting (DSAI)
techniques [9, 27, 39, 44]. Below, we briefly illustrate how
3 The smart FinTech ecosystem DSAI techniques can assist in these smart FinTech ar-
eas.
FinTech is evolving to be a large family, compris- Smart banking offers active, personalized, auto-
ing various areas including BankingTech, TradeTech, mated, trusted, robust, secure, risk-averse and fragility-
LendTech, InsurTech, WealthTech, PayTech, and Risk- averse banking businesses and services, including retail
Tech. The smart FinTech ecosystem is a multi- banking, open banking, mobile banking, and business
dimensional synergy between EcoFin business objec- banking. The DSAI research for enabling smart bank-
tives, business areas, data and resources, and enabling ing includes DSAI techniques for detecting, analyzing
techniques, as shown in Fig. 2 in [8] which presents a and managing a bank’s network risk and risk conta-
four-dimensional landscape of the smart FinTech and gion; modeling, detecting and managing banking risk
the synthesis between these aspects. In general, the and fraud; conducting credit analysis and improving
4

pricing; optimizing, validating and risk-managing credit and optimizing novel, secure, smart and personalized fi-
loans; creating, mining, securing, risk-managing and op- nancial mechanisms, models, products, and services for
timizing digital currencies including cryptocurrencies; equity and derivative markets; making and optimizing
creating, securing, evaluating, risk-managing and opti- algorithmic trading, arbitrage trading, high frequency
mizing mobile banking businesses and services; person- (cross-market) trading, and institutional trading across
alizing, automating, validating, securing, risk-managing equity, commodity and currency markets; modeling, au-
and optimizing Internet/online banking; enabling, per- tomating and optimizing digital asset pricing and credit
sonalizing, automating, securing, risk-managing and scoring; enabling smart e-commerce for personalized re-
optimizing open banking; evaluating, securing and risk- tail businesses; enabling smart, efficient, personalized
managing shadow banking; and developing smarter and and secure Internet finance; conducting portfolio an-
more automated, personalized, mobile and engaging alytics and enabling smart portfolio management; en-
banking services; etc. abling predictive trading and strategic trading; design-
Smart insurance enables insurance products, sys- ing, automating and optimizing trading strategies, al-
tems and services to ensure safe/secure, cost-effective, gorithms and platforms; enabling and optimizing so-
proactive, tailored, trustful, resilient, secure and risk- cially responsible investment; modeling, evaluating and
averse health, car, home/content/building, travel and optimizing trading complementarity and substitution-
other businesses. The DSAI research for enabling smart ary; and characterizing and improving trading incen-
insurance includes those for tailoring individual, busi- tives and campaign; etc.
ness and commercial insurance products and services;
enabling early, active and evolving insurance fraud de- Smart Internet finance has evolved from Internet
tection; making active, personalized and time-varying banking to third-party payment, peer to peer lending
recommendations of insurance products and services; (P2P lending), crowdfunding, and digital currencies,
evaluating, analyzing, detecting, managing and opti- which also involve their supporting designs, operations,
mizing insurance risk and compliance; evaluating, au- services, evaluation, compliance, and security (trust).
tomating, detecting and optimizing insurance security; More broad Internet financial services include Inter-
creating novel insurance models, products and services; net platforms, data and technology-driven financing,
evaluating, automating and improving insurance busi- fund, lending, insurance, payment and markets asso-
ness operations; and integrating, recommending and ciated with individuals, corporate organizations, spon-
optimizing multi-policy, multi-product and cross-selling sors, and regulators. Typical FinTech for smart Inter-
insurance products and services; etc. net finance includes inventing new Internet technologies
Smart lending supports lending, loan and mort- and services to enable the above products and services;
gage businesses and services that are risk-averse, per- enabling efficient, trustful, secure and convenient In-
sonalized, context-oriented, predictive, efficient, robust ternet banking services, authentication, active compli-
and secure for individuals, enterprises or projects. DSAI ance checks, outlier detection, and connections to other
research directions for smart leading cover businesses banking businesses such as mobile banking; enabling
and tasks such as for addressing relevant aspects and equity, product and reward-based crowdfunding mech-
problems through blockchain; automating crowdfund- anisms, projects and their developments and risk man-
ing, e.g., campaign design and strategy optimization; agement, e.g., default prediction, evaluating the success
creating and optimizing distributed ledger technologies; and risk of crowdfunding projects, automatic screening
enabling fundraising such as by creating personalized of illegal fundraising, evaluating and managing herd-
recommendations, efficient fundraising models, and in- ing effect, social networking and mutual influences, and
telligent services; analyzing, detecting, managing and advising government regulation on investors and corpo-
mitigating lending risk and security; enabling smart, ef- rate; optimizing P2P lending bidding mechanisms and
ficient and low-risk peer-to-peer lending; and creating, loan evaluation, predicting and managing P2P lending
validating and optimizing smart contracts; etc. herding effect and information asymmetry, and evalu-
Smart markets and trading involves very broad ating their influence and systemic risk on investment
EcoFin businesses and widely-explored DSAI tech- and the economy; optimizing pricing, price volatility
niques for intelligent capital, commodity, currency, and fluctuation, payment accuracy, risk on investment,
energy, commerce, property and other markets, and regulation, operation security and efficiency of digital
supplies predictive, active, dynamic, risk-averse, anti- currencies. Other areas where DSAI may contribute in-
fragile, and high-utility trading strategies, support and clude enabling, securing, and regulating so-called big
services. The DSAI-driven smart markets and trading data finance, information-driven financial institutions,
can support many tasks, e.g., for designing, automating and smart systems and services for Internet finance.
 5

Smart billing and payment offers efficient, se- automated, dynamic, and high-performing blockchain
cure, risk-averse, fast and convenient systems and ser- infrastructures, computing and services for encrypted
vices for online, mobile, WiFi, contactless (including digital currencies. DSAI techniques can play a critical
for credit cards and by QR-codes) and IoT-oriented role in enabling smart blockchains. Examples are pre-
billing and payment services. Smart billing and pay- dicting the price movement of cryptocurrencies; con-
ment becomes increasing important in the digital so- structing risk-averse trading strategies and portfolios of
ciety and economy. DSAI-enabled smart payment re- bitcoins and other cryptocurrencies; detecting risk and
search include areas such as authenticating, automat- assuring smart contracts; enabling the efficient mining
ing, securing and risk-managing billing validity and of bitcoins in distributed systems; supporting proactive
contactless payment; supporting efficient and secure and systemic blockchain governance and regulation;
IoT device-based contactless payments; enabling, se- and detecting risk, intrusion and fragility in blockchain
curing, risk-managing and optimizing e-payments; au- systems, behaviors and activities; etc.
tomating, securing, risk-managing and optimizing In- Smart marketing and services offers cost-
ternet and online payments; automating, securing, risk- effective, relevant, proactive, personalized, positive, se-
managing and optimizing mobile payments; modeling, quential and context-oriented marketing, advertising
detecting, analyzing and mitigating payment risk; and and recommendation activities and services. DSAI-
validating, detecting, analyzing and mitigating billing enabled smart marketing involves many opportunities,
and payment risk and security; etc. such as conducting consumer sentiment and public emo-
Smart credit/risk management offers efficient, tion analysis; understanding consumer/client opinion
personalized, active, evolving, secure and sustainable and intention; quantifying, characterizing, evaluating,
credit and risk valuation, products, services and man- predicting and improving consumer confidence and a
agement for business and customers. Typical DSAI- product/service’s market reputation and trust; eval-
enabled smart credit and risk management tasks in- uating and optimizing customer relationship manage-
clude automating, evaluating and optimizing anti- ment; validating, enhancing, synergizing and optimiz-
money laundering; monitoring, detecting, categorizing, ing econometrics; evaluating, risk-managing, personal-
factorizing, predicting and intervening client financial izing and optimizing stakeholder relationships; mak-
security, retail investor risk, financial systematic risk, ing, evaluating and optimizing financial recommen-
financial institutional risk, financial network risk, and dations; enabling, automating, evaluating, securing,
cross-market risk; detecting, quantifying and predict- risk-managing and optimizing supply chain finance;
ing financial risk factors and areas; quantifying, ana- and supporting user privacy-preserving and advertiser
lyzing, detecting, predicting and mitigating financial information-protected federated advertising; etc.
crisis and crisis contagion; categorizing, monitoring, de- Smart compliance and regulation supports au-
tecting, analyzing, predicting, evaluating and managing tomated or human-machine-cooperative, risk-sensitive,
financial events, market movement, change, exception proactive, systematic, dynamic and evidence-based reg-
and emergence and their consequences and impact; etc. ulation and compliance operations, governance and risk
Smart asset and wealth management offers intervention and control. The opportunities associated
businesses and services safe, secure, personalized, anti- with DSAI-empowered smart compliance and regula-
fragile, and automated management of money, credit, tion are comprehensive and significant for any Fin-
properties, securities and intangible assets. DSAI- Tech business and technologies, such as enhancing cor-
supported smart asset and wealth management involves porate finance transparency; evaluating and optimiz-
many emergent research areas, including DSAI for mak- ing corporate governance and regulation; making and
ing, evaluating and optimizing data monetization; valu- optimizing cross-market regulation; creating and opti-
ating, analyzing, evaluating, optimizing and managing mizing digital currency regulation; detecting, analyz-
digital assets, Internet wealth, public welfare, institu- ing, risk-managing and managing financial crime; ver-
tional welfare, and superannuation; enabling, automat- ifying, automating, securing and optimizing financial
ing, risk-managing and optimizing digital financial ad- digital authentication; enabling, validating, verifying,
vising; designing and optimizing novel, personalized, se- detecting, risk-managing and optimizing financial dig-
cure and healthy financing mechanisms; enabling, au- ital identification; quantifying, validating, monitoring,
tomating, evaluating and optimizing roboadvising; and detecting, analyzing and mitigating financial fragility,
supporting secure, personalized, low-risk and sustain- crisis and stability; recognizing, detecting, analyzing
able venture capital management; etc. and mitigating financial fraud; enabling smarter and
Smart currencies and blockchain enables ef- more efficient and personalized financial market regu-
ficient, secure, risk-tolerant, automated or semi- lation, design and policy implication; automating, eval-
6

uating and optimizing financial operations; quantifying, ing and valuation, trading and exchange, supply
evaluating and managing information asymmetry and and demand, financial services (e.g., billing and pay-
transparency; enabling, automating and optimizing in- ment), etc.
ternational regulation; validating and improving mar- – Operate: enable and support the whole-of-business
ket legitimacy; quantifying, verifying, detecting, analyz- execution by providing technology (including infras-
ing and improving market social justice; and quantify- tructure) and managing resources (money, asset,
ing, evaluating, automating and improving marketplace personnel and management, data, facilities, etc.),
trust and coordination; etc. communications (internal and external), stakehold-
FinTech ethics and security involves whole- ers, processes, and operations, etc.
of-businesses, whole-of-time, whole-of-stakeholders, – Promote: market and advocate businesses and op-
whole-of-physical-cyber-spaces, privacy-preserving, portunities, including supporting advertising, com-
proactive, predictive, systematic, and dynamic man- petition, recommendations, new developments (e.g.,
agement of FinTech ethics and security. DSAI tech- clients, products, channels, etc.) and managing
niques can help with extracting evidence and profiling stakeholder relationships, etc.
ethical and security violations; modeling, detecting, – Optimize: evaluate, improve and expand the quality
evaluating and managing financial system security, and performance (including efficacy and efficiency)
financial network security, and financial instrument of EcoFin businesses and operations and make opti-
security; characterizing, quantifying, analyzing, eval- mal planning, strategies, reinforcement, recommen-
uating, predicting and managing financial system dations and innovations, etc.
vulnerability; and providing distributed, federated, – Safeguard: validate, secure and assure the compli-
privacy-preserving financial services, etc. ance, regulation, privacy, security, trust and ethics
FinTech innovations drive smart FinTech devel- etc. of EcoFin businesses and FinTech.
opments and applications. Every area of smart FinTech
needs continuous innovation and research, where DSAI
technology is essential. Examples of DSAI-driven Fin-
Tech innovations are data and learning-enabled sys- 4 DSAI techniques enabling smart FinTech
tems and services for automated pricing, credit scor-
ing, loan valuation, trading strategy generation, cus- The DSAI techniques required to achieve the aforemen-
tomer chatbots, financial planning, security alerting, tioned smart FinTech objectives and EcoFin businesses
and compliance mitigation; proactive and personal- are very broad and diversified [6, 11, 15, 16, 20–22, 24, 26,
ized recommendations of crowdfunding projects, cross- 27, 31, 33, 40, 41, 46]. There are different ways to catego-
product loan, insurance and investment portfolios and rize the techniques enabling smart FinTech. For exam-
pricing; tailored risk-mitigated systems and services for ple, Fig. 2 [9] categorizes the techniques into (a) math-
asset and wealth portfolio prediction, optimization and ematical and statistical modeling, (b) complex system
risk management for high-value customers; detecting methods, (c) classic analysis and learning methods, (d)
and intervening unethical and unsecure trading, lend- computational intelligence methods, (e) modern DSAI
ing, credit and loan valuation, payment, marketing, methods, (f) deep financial modeling methods, and (g)
competition, and regulation; offering whole-of-business, hybrid methods. Table 1 details the popular techniques
privacy-preserving and federated FinTech and EcoFin and their applications for smart EcoFin businesses and
businesses and services to large-scale, distributed and FinTech.
connected communities or societies, etc. Here, we expand the above scope and briefly discuss
On the other hand, all of the above smart FinTech
various intelligent services required for smart EcoFin
businesses and areas involve some common fundamen-
and FinTech. The typical categories and applications
tal processes, functions and activities. Fig. 2 illustrates
of various DSAI techniques for smart FinTech include:
the main processes and their key functions and activ-
ities: design, produce, operate, promote, optimize and
– complex system methods: techniques to characterize
safeguard.
and design complex EcoFin systems;
– Design: plan and devise EcoFin mechanisms, mar- – quantitative methods: techniques to quantify
kets, products, participants, rules and governance, EcoFin designs and businesses;
and innovations, etc. – intelligent interactions and responses: to enable in-
– Produce: convert and implement the designs into telligent human-machine interactions and effective
production with goods, services, systems and appli- conversations with stakeholders and autonomous or
cations and functions and activities including pric- personalized responses;
 7

– analytics and learning techniques: to analyze, eval- time-series analysis, signal analysis, statistical includ-
uate, optimize and recommend smart FinTech busi- ing Bayesian methods, random methods, utility theo-
nesses, services, and data; ries, and econometric methods. They are used in quan-
– optimization and augmentation techniques: to eval- tifying EcoFin businesses, e.g., formulating pricing and
uate, enhance and expand the quality, performance valuation models; quantifying the financial variables
and opportunities of smart businesses, data and Fin- describing businesses; developing quality and perfor-
Tech; mance evaluation matrices; evaluating and balancing
– deep financial modeling: to involve deep, reinforced, supply and demand relationships; testing hypotheses
transfer and Bayesian learning techniques and other and mechanisms; and analyzing issues such as uncer-
advances in the representation, analysis, prediction tainty and chaos in the businesses; etc. In addition,
and management of financial businesses and ser- more advanced analytics and learning methods such
vices; as deep learning methods are also used in quantitative
– system intelligence enhancement: to enhance au- finance and financial forecasting [51]. Interested read-
tonomous and intelligent interfaces, technologies, in- ers can refer to [1, 9, 5, 14], etc. for more information
frastructures and systems for smarter EcoFin busi- on these methods and their applications in EcoFin and
nesses and FinTech; and smart FinTech.
– whole-of-business privacy-preserving federated Fin- Intelligent interactions, recognition and re-
Tech: to provide unified but distributed secure fi- sponse. EcoFin businesses are embedded into personal
nancial services by connecting the entire aspects of daily life and societal activities, where interactions (in-
humans, systems, businesses, IoT devices, and data. cluding dialogue and communications) between people,
between humans and machines, and between machines
All of these techniques may be applied in each of
become essential and overwhelming. Such interactions
the aforementioned six processes and functions of smart
may take form of verbal dialogue (speech-based), tex-
FinTech. Each technique may play a distinct role in ad-
tual communications (e.g., hand-written, emails, ques-
dressing the respective EcoFin business challenges and
tions/answering, chatbots, or retrieval), visual dialogue
data complexities, as illustrated below.
(video-based), imagery interactions (e.g., images or QR
Complex system methods. The design of EcoFin
codes), sensor-based (e.g., IoT devices-based) interac-
businesses and smart FinTech often relies on and in-
tions, or multi-channel, multi-domain or multi-modal
volves theories of complex systems and social sciences,
(e.g., audio-video plus chat) dialogue in EcoFin or by
including brain and cognitive computing, complexity
FinTech. Other businesses and tasks built on the above
science methods, game theories, behavior science, be-
interactions include automated identification of inter-
havioral economics and finance, network sciences, and
actions and communications and response generation
agent-based modeling and simulation, etc. As illus-
for autonomous, open-domain, context-aware, multi-
trated in [54, 9, 18, 32, 10, 3], such theories and meth-
turn, multi-level, multi-lingual and knowledge-based di-
ods can be used for EcoFin designs, developments, and
alogue, interactions and communications. Examples are
decision-support, e.g.,
[36, 38, 17, 4]:
– modeling, simulating, representing, characterizing, – automatically recognizing the identity of interacting
designing and evaluating EcoFin and FinTech mech- objects from oral, image, video, retrieval or text-
anisms, markets, products, participants, rules and based dialogue, communications or queries;
governance, etc.; – understanding interaction intent;
– simulating and determining their working mecha- – analyzing sentiment and emotion;
nisms, system characteristics and complexities; – tracing historical dialogue context and feedback;
– characterizing psychological, emotional and behav- and
ioral factors and influence; – generating and recommending follow-up engage-
– resolving issues and conflicts; and ments and actions for automated responses, etc.
– validating and reinforcing rules, organizational
Analytics and learning methods. Both clas-
structures, processes, interactions and performance;
sic and advanced analytics and learning methods play
etc.
increasingly critical roles in making EcoFin and Fin-
Quantitative methods. Quantitative methods Tech smarter, more tailored, efficient, adaptive or evolv-
including mathematical and statistical modeling are ing. Widely used classic methods consist of pattern
widely applied in EcoFin designs and businesses. Eco- analysis methods, kernel methods, tree models, factor
nomics and finance heavily rely on typical quantita- models, relation models, representation learning, lan-
tive methods, including various numerical methods, guage models, image processing methods, signal pro-
8

Fig. 2 AI and data science enabling smart FinTech for financial businesses, processes and activities.

cessing techniques, classic neural networks, and ensem- – detecting illegal or incompliant behaviors and events
ble methods, etc. More advanced analytics and learn- such as discovering pool manipulation from collec-
ing dominates today’s financial data analysis, financial tive investment accounts and investor behaviors [8];
engineering, and FinTech innovations. Such techniques – modeling the impact of external events in financial
comprise knowledge representation methods, short-long markets such as the impact of information security
and informal-formal text processing and translation, on stock market movement [53]; and
imagery analytics, visual analytics, social media and – modeling cross-market investment strategies, invest-
network analysis, advanced optimization methods, ad- ment herding behaviors, volatility spillover, and risk
vanced machine learning methods (e.g., on multimodal, management [58, 59].
multisource, high-dimensional, low-quality data, etc.),
cross-business (e.g., products, policies, markets or cus-
tomer groups) analytics, and whole-of-business (e.g., all Deep financial modeling. Deep financial model-
products, services and customers of a bank) representa- ing focuses on key issues that can be categorized into (1)
tion, analysis, recommendation and reinforced services. deep financial representation and prediction, (2) deep
Numerous applications of these analytics and learning cross-market, section and factor modeling, and (3) deep
techniques have been reported in the literature and are distributed financial modeling [9]. These tasks apply
possible, e.g., as discussed in [8, 9] and Table 1. Exam- deep learning models, including basic deep neural net-
ples are works and their recent developments, such as diversified
deep neural mechanisms, architectures and networks
(e.g., recurrent neural networks, graph neural networks,
– predicting market trends or price movements such neural language models like Transformer and BERT
as jointly learning multi-source data and hetero- variants, image nets, attention networks, memory net-
geneous information fused together with financial works, adversarial learning, and autoencoders, etc.).
events or social media sentiments [60]; In addition, new developments on deep reinforcement
– generating algorithmic trading strategies such as an- learning, deep Bayesian learning, deep transfer learn-
alyzing relevant price-sensitive financial events from ing and deep federated learning have also intensively
the news and announcements about the security or transformed the landscape of DSAI-driven finance and
predicting security price movement [41]; FinTech [46, 57, 50, 52]. Examples are
 9

– recurrent neural networks for multivariate time- computing, evolutionary computing and fuzzy set the-
series forecasting, temporal and sequential model- ories; active and adaptive learning and meta-learning
ing of financial variables, and price or index trend methods; and optimization methods for deep neural
forecasting [51, 26, 52]; networks.
– deep reinforcement learning such as deep Q- System intelligence enhancement. The smart-
networks for stock forecasting, optimal portfolio ness of FinTech not only relies on the above tech-
management, high-frequency trading strategy devel- niques but also requires extra system intelligence and
opment for algorithmic trading [37, 34, 30]; its enhancement for smart EcoFin businesses and ser-
– deep neural networks such as recurrent, regression vices. Here, we briefly highlight the following techniques
and graph neural networks and deep transfer learn- to enhance the system intelligence of FinTech: smart
ing models for representing macro to micro-level fi- blockchain, intelligent (re-)identification and authen-
nancial data or cross-sectional stock representation tication, automated analytics and learning, intelligent
for portfolio prediction and strategy development recommendation, tailored interactions and services, and
[43, 46, 55]; ethical and interpretable FinTech [45, 23, 25].
– deep learning of illegal, noncompliant, risky and
– Smart cryptos and blockchain: for secure, dis-
fraudulent behaviors such as insider trading and
tributed and automated infrastructures and sys-
market manipulation in capital markets and finan-
tems; encrypted, efficient, robust and distributed
cial accounting and reporting fraud in financial ser-
valuation, investment, exchange and portfolio man-
vices and statements [12]; and
agement of digital currencies (bitcoins and other
– federated learning models for privacy-preserving
cryptos); distributed and risk-tolerant trading
open-domain and whole-of-business financial appli-
strategies and contracting; adaptive risk and event
cations and services [19, 57].
management and regulation; and novel blockchain
Augmentation and optimization methods. businesses and services.
Augmentation and optimization techniques [29] may – Intelligent (re-)identification and authentication: of
be useful to tackle business and data complexities in identity, access, biometrics (eg., face, fingerprint,
EcoFin businesses; enhance FinTech design objectives, voiceprint, speaker or sign language), liveness (face
mechanisms, capacity and strategies; or improve Fin- or fingerprint) or from optical character recognition
Tech capability, quality and performance. EcoFin and (OCR), dialogue systems, automated speech recog-
FinTech may involve image, visual, audio, textual and nition systems, text-to-speech translation systems.
transactional data with various quality issues such in- – Automated analytics and learning: of automatically
consistencies, noises, heterogeneity and feature hidden- collected and processed EcoFin data and businesses,
ness. Data augmentation techniques such as adding updating EcoFin services or suggesting engagement
random noises, adversarial training, masking, gener- opportunities or actions (such as alerts, responses,
ative adversarial networks and neural transformation and recommendations), evidenced by validated an-
may improve data quality and better serve FinTech alytical results.
objectives, e.g., contrast, translation, saturation and – Intelligent recommendation: of EcoFin services and
color augmentation for visual data; word shuffling, ran- businesses (such as stocks, portfolios, financial
dom swapping, insertion or deletion, and replacing syn- plans, and promotions) tailored for specific clients
onyms for text; and cropping, changing speed and fre- by considering their needs, habits and intention of
quency and wavelet and Fourier transform for audio consumption, evolving circumstances, and feedback,
and time-series data. In addition, augmentation may etc.
enhance technical capability or analytical results, e.g., – Intelligent interactions and services: with cus-
by adding distributional noises, biases or drifts to data tomers by interacting with them through auto-
or objectives, which also improve FinTech capacity and mated, context-aware and personalized visual, ver-
generality. On the other hand, optimization is increas- bal, imagery, textual or multi-modal dialogues or
ingly essential for building robust and actionable Fin- communications and by considering their circum-
Tech, e.g., optimizing pricing and valuation, investment stances, interests, intentions, demand and evolving
value at risk, transaction costs, portfolio management, context, etc.
and cross-market investment and risk management [11, – Privacy-preserving microservice architectures:
21]. Typical methods include mathematical optimiza- breaks down complex large-scale enterprise-wide
tion methods; statistical optimization such as Bayesian infrastructures to local services-oriented privacy-
variational inference; information theory-driven opti- preserving federated components and microservices
mization; neural optimization methods such as neural for efficient computing and services.
10

– Ethical and explainable FinTech: offers EcoFin busi- – offer privacy-preserved, connected, federated and
nesses and services that are privacy-preserved, ac- differentiated financial systems and services;
countable, transparent, unbiased and interpretable, – conduct online, cloud-based, privacy-preserving fed-
e.g., in FinTech and business designs, products, pric- erated analytics and learning of all businesses, users
ing, quality, accessibility, market campaigns, recom- and data;
mendations, and risk management. – support customized, context-aware, multi-channel
and instant interactions, communications and cus-
Whole-of-business and privacy-preserving tomer services;
federated FinTech. With the networked and glob- – provide secure and personalized advertising, market
alized world, economic activities from design to tech- campaigns, and recommendations.
nical development, manufacturing, sales and trad-
ing, supply chains, cross-border and multi-national e- Federated learning offers a decentralized method to
commerce, customer services and financial services are train a machine learning model based on multi-party’s
increasingly connected and chained. Such EcoFin data in a secure and privacy-preserving manner. It al-
businesses often involve multiple, large-scale, dis- lows participating financial and non-financial organiza-
tributed and privacy-sensitive products, services, plat- tions to collaborate without worry that their data pri-
forms, applications, markets, organizations, applica- vacy or key model parameters are leaked out unexpect-
tion domains, customer communities, regions, and data edly. Often organizations’ data and models complement
sources. An example is a smart residential district each other either by enhancing the diversity of train-
or property chain, where intelligent buildings con- ing samples or feature spaces. These dimensions allow
nect housing services, strata services, utility services, machine learning models to be more robust and more
carparking services, community services, shopping, and accurate in financial applications. Many federated ap-
healthcare, etc. Such multi-aspect businesses require plications have been successfully implemented, in finan-
smart home/city-oriented FinTech to cater for whole- cial credit risk modeling, Internet marketing, call center
of-business, distributed and privacy-preserving busi- automation, insurance and asset management [57, 56].
ness objectives, intelligent infrastructures, functionali-
ties, operations, services, and stakeholder management.
5 About the edition on DSAI in FinTech
The hybrid systems also involve multi-source, hetero-
geneous, distributed, individual-to-organizational, and
There are several initiatives on AI and data science in
privacy-sensitive customers and their private activi-
finance and FinTech, such as the special track on AI
in FinTech organized in IJCAI’20202 and the panel on
ties, behaviors, interactions and preferences, which are
recorded by Internet services, CCTV systems, mo-
AI in FinTech: Challenges and future3. The special edi-
bile apps, and WiFi systems, etc. Whole-of-business
and privacy-preserving federated systems [19, 57] and tion on Data Science and AI in FinTech is a major
initiative of JDSA in promoting the research and appli-
FinTech are highly demanding with functions such
cations of data science and AI in finance, the economy,
as securely linking all businesses; systematically serv-
society, and daily life for individual, public, economic
ing the whole community; providing unified secure ac-
and social good. The dozens of papers available online
cess, identification, authentication; supporting privacy-
cover a wide spectrum of DSAI techniques and busi-
preserving and personalized interactions, conversations
ness domains and applications. Fig. 3 provides an ini-
and responses within the community and between users;
tial picture of this collection of research as indicated by
and providing multi-product/policy and cross-business the frequently occurring keywords identified in these
offers and personalized recommendations. In a more papers, particularly backed by the DSAA’2020 Journal
general context, such whole-of-business and privacy- Track on Data Science and AI in FinTech4 and the spe-
cial issue on the same topic5. Below, we briefly outline
preserving federated FinTech and intelligent EcoFin
businesses is able to
the scope and capacity of the early-stage collection of
– connect all users, systems, organizations and data papers on this important topic.
through heterogeneous, private-preserving and se- Research areas. This special edition already cov-
cure networks; ers a wide spectrum of research topics, tasks and meth-
– fuse multi/cross-aspect and privacy-preserved busi- ods. We summarize them in terms of the following top-
nesses and services; 2
https://static.ijcai.org/2020-accepted papers.html
– support transparent (e.g., from heterogeneous sign- 3
https://ijcai20.org/ai-in-fintech-challenges-and-future
on apps) and privacy-preserving access, identifica- 4
http://dsaa2020.dsaa.co/journal-track-papers/
tion and authentication; 5
https://www.springer.com/journal/41060/updates/19117592
 11

Fig. 3 The word cloud of the collection on DSAI in FinTech.

ics: time series transformation, data manipulation and time) and streaming data; handling noise and miss-
augmentation, financial forecasting and prediction, pat- ing values; feature extraction and engineering; di-
tern mining, clustering and classification, behavior and mension reduction; sampling and resampling meth-
event analysis, causality analysis, relation learning, net- ods; visualization of spatial (altitude), temporal and
work analysis, anomaly and fraud detection, emotion historical data.
and sentiment analysis, influence and impact model- – Financial forecasting and prediction: including mul-
ing, active, dynamic, online and adaptive learning, deep tivariate time-series analysis such as vector regres-
learning, performance evaluation, and other topics. sion; cross-sectional short-to-long-range forecast-
ing of multivariate financial variables by autore-
– Multivariate time series transformation: including gression; classic classifiers, query-driven statistical
the transformation, modeling and representation learning, and resampling with temporal and rele-
of discrete and continuous multivariate time series vance bias for time-series forecasting; biased resam-
(e.g., by copula transformations); handling multi- pling for imbalanced spatio-temporal forecasting;
collinearity; factor analysis; partial least squares multi-step-ahead nonlinear multivariate forecasting;
regression; structural equation modeling; dynamic jump-diffusion models for high-order estimation of
time warping and subsequence extraction. options prices; predicting socioeconomic status by
– Data manipulation and augmentation: includ- hypergraph-based factor graph modeling of mobile
ing handling data complexities such as high- call records.
dimensional, high-order, sparse, incomplete, hier- – Pattern mining: such as high utility itemset mining
archical, cross-sectional, long-range, asymmetric, of purchase records.
skewed, evolving, large-scale, imbalanced, live (real-
12

– Clustering and classification: of large-scale behav- credit scoring by recurrent neural networks, graph
ioral data (e.g., human actions and interactions) and neural networks and autoencoders; sequential mod-
blockchain bitcoin networks; cloud-based scalable eling and attention mechanisms of predicting data
clustering; diffusion-based region clustering; discov- quality issues.
ering time-series motifs (subsequences) by dynamic – Performance evaluation: including evaluating the
time warping; etc. overfitting, bias, usability, utility, representative-
– Behavior and event analysis: including user mod- ness, understandability, interpretability, explain-
eling of online and social behaviors and people’s ability, transparency, accountability and liability of
lifestyle; extraction and classification of financial results.
events from news articles; trajectory prediction by – Other topics: including transfer learning and incre-
semantic clustering and neural network regressors; mental learning of financial frauds; recommenda-
analyzing patterns of human behaviors and group tions of individual and group purchase patterns by
interactions; complex event processing by detecting mixture models of point processes; online change de-
event and service patterns; consumer purchase be- tection in stochastic processes; distributional stream
haviors. drifts and concept changes in business transactions;
– Causality analysis: such as linear and nonlinear and imbalanced learning.
Granger causality; causally anomalous multivariate Business domains and applications. Many busi-
time series; causal tree-based causal inference with ness problems, objectives, and applications have been
instrumental variables; etc. addressed in this edition. They cover issues including
– Relation learning: including multivariate depen- (1) various business domains such as investment and
dency (e.g., by copula methods), input-output trading, banking, retail industry and blockchain (digi-
causality (e.g., by Granger causality), query-answer tal currency, bitcoin, and cryptocurrency), pricing and
relations, connectivity (e.g., distance, degree, cen- valuation, and recommendation; (2) understanding the
trality and densification, etc.) in bitcoin user-based complexities of businesses, participants and their be-
graphs, and correlation skewness and multivariate haviors, such as population and community, customer
skewness. behaviors and lifestyle, mobility, dynamics and change,
– Network analysis: including social media-based net- financial events, information exchange, public senti-
works by social network and social media analyses; ment and emotion, sectional and global businesses, so-
cryptocurrency and blockchain networks by graph cioeconomic impact, and cause-effect; and (3) support-
theories; mobile phone calls-based networks by node ing compliance and regulation and other issues, such as
embeddings. market surveillance, risk, data quality issues, fairness
– Anomaly and fraud detection: by causal anomaly and ethics.
detection, clustering methods, classification, incre- – Investment and trading: including equity invest-
mental learning, transfer learning, network science ment, stock trading strategy backtesting, invest-
methods, etc. ment management, energy consumption, retail pay-
– Emotion and sentiment analysis: by social media ment, sales of appliances, credit scoring.
analysis, topic modeling, text stream classification, – Banking: such as credit card fraud detection, credit
visualization of sentiments, etc. scoring, and processing data quality issues.
– Influence and impact modeling: such as the quanti- – Retail businesses: including discovering high-utility
tative estimation of social and economic impact of products in supermarket transactions for revenue
data science on society, economy, industry and busi- ranking and product indexing and placement, and
ness; influence diffusion and propagation; socioeco- region-specific product purchase patterns.
nomic development and well-being; measuring sub- – Blockchain: such as modeling the relations between
jective and objective well-being; individual socioe- bitcoin owners and the concentration of user rich-
conomic status; etc. ness, and estimating cryptocurrency asset pricing.
– Active, dynamic, online and adaptive learning: in- – Community and citizen: such as minor heritage as-
cluding online preprocessing and optimization of sets; population indications (residents, commuters
news data and entity prediction; exploratory ac- and visitors) in mobile phone calls; public well-being
tive learning; stochastic semi-supervised learning; and social good related to health, job opportunities,
oversampling; multiple instance learning; incremen- socioeconomic status, socioeconomic development,
tal dynamic factor analysis. safety, environment and politics.
– Deep learning: including interpretable neural net- – Customer behaviors and lifestyle: such as individual
works to visualize financial texts; link prediction for and group purchase behaviors and preferences; user
 13

resting, sleep, wakeup and exercise behaviors and time and in any form through channels such as QR
quality; online and social behavior patterns; and ac- codes, WiFi networks, mobile applications, social me-
cidental clicks of mobile ads and online advertising dia networks, short messaging platforms, intelligent dig-
click-through. ital assistants, and the Internet. They accumulate fast-
– Mobility: such as on aircraft flight plans and trajec- growing tangible and intangible assets and services,
tories, and mobile phone call-based human mobility. numerous products, applications and services, and in-
– Dynamics and change: such as user behavior creasingly bigger financial data. AI and data science
changes, and live stream drifts in multivariate time are playing an increasingly critical role in making fi-
series. nance smarter and fostering the ever-growing smartness
– Financial events: such as social and public events, of FinTech and the intelligence of autonomous finan-
and web service events and patterns; and story- cial systems and personalized financial services. To this
telling by generating stories by extracting and pre- end, DSAI advances include intelligent identification
dicting entities and events in financial news articles. and authentication, autonomous interactions and com-
– Information exchange: such as citizens as informa- munications, cloud analytics, deep learning, federated
tion providers and sentiment flow in social media. learning, cross-market analytics, deep financial model-
– Public emotion: such as social sentiment, negative ing, and automated interactions and responses are also
sentiments (e.g., depression, lack of interest, guilt, evolving to address the emerging challenges and op-
suicide, etc.) of at-risk populations, and market portunities in smarter financial businesses and bigger
mood; and in online communities, and public com- financial data.
munities. Opportunities to DSAI communities. Finance
– Cause-effect: such as evaluating heterogeneous is one of the mostly vivacious, data-rich, technology-
causal effects (e.g., credit for small firms) by causal savvy and intelligent areas. The advances of DSAI will
inference. further drive smart futures [9] of finance and FinTech,
– Market surveillance: such as monitoring order be- in areas including but not limited to:
haviors in stock market order book and multivariate
time series by causal outlier detection. – offering intelligent (e.g., proactive, optimal, au-
– Risk: such as fraudulent behaviors in banking etc. tonomous, evidence-based, and differentiated) re-
businesses, and corruption of contracting markets. view, planning, generation and recommendations of
– Data quality issues: including identifying and pro- financial objectives, areas, strategies, and actions;
cessing diversified quality issues as discussed in the – enabling anytime, anywhere and any-form access,
above on data manipulation and augmentation in identification, authentication, consumption, pay-
financial businesses such as for banking regulation, ment, communications, and responses to multiple
and improving fairness and decision-support. heterogeneous networks and applications;
– Ethics and fairness: including privacy issues, selec- – enabling lifelong (evolving), all-purpose and whole-
tion bias in retail businesses, algorithmic bias in rec- of-business (multi-policy) businesses and services to
ommendations, temporal and relevance bias (imbal- individual and group customers;
ance), and fairness impact of biased missing values – suggesting novel opportunities such as new prod-
and selection bias. ucts, new functions, new markets, new customers,
Obviously, the above summary of the research areas and new developments;
and business applications in this edition will be sub- – providing whole-of-business, federated and privacy-
stantially expanded over time. preserving business support, data-driven learning,
evidence extraction, recommendations and services;
– supporting global, all-purpose, cross-lingual, cross-
6 Discussion and opportunities modal, cross-business, and cross-apps privacy-
preserving, transparent, tailored and active busi-
Here, we discuss the trend of the era of smart FinTech ness fusion infrastructures, financial services, busi-
and some opportunities for smart FinTech futures by ness recommendations and services;
the DSAI communities. – understanding user intention, emotion, preference,
The era of smart FinTech. Smart FinTech is in- feedback and their evolution to offer positive, active
creasingly dominating and evolving into a paramount and tailored customer experience and services; and
pillar of modern and future economies, society, and de- – automating integrative whole-of-process businesses
velopments. FinTech connects every person, organiza- and services, including autonomous product catego-
tion, product, service and activity anywhere, at any rization, catalogue refilling, market positioning and
14

promotion, billing, logistics dispatching, payment, – how can DSAI empower more transparent and ro-
product evaluation, and customer services, etc. bust digital currencies and blockchain systems and
In addition, there are still many questions for the services?
DSAI communities to answer, e.g., – how has deep learning significantly transformed
– how can AI and data science translate the existing finance? and what are the essential finance and
finance and FinTech? FinTech-oriented AI, deep learning and data science
– what are the new smart futures of DSAI-driven fi- techniques?
nance and FinTech? – how can brain and cognitive computing and human-
– what financial business and data complexities are centered computing transform finance and FinTech?
not yet addressed by the existing analytics and – how can DSAI complement quantum computing for
learning family? world-wide, secure and efficient financial infrastruc-
– how can DSAI algorithms autonomously access and tures and services?
analyze the whole-of-business data, generate whole- – to what extent can autonomous, personalized and
of-enterprise (or an individual object) data ge- precision finance possibly transform personal fi-
nomics, and make actionable recommendations for nance, corporate finance, public finance and Inter-
the management? net finance?
– how can DSAI enable efficient, privacy-preserving
personal finance and Internet-based open banking
and finance in a global or large-scale open domain?
 15

Table 1: DSAI Techniques and Their Representative Applications for Smart FinTech.

DSAI areas Techniques Specific DSAI methods Smart FinTech applications

Numerical Linear and nonlinear equations, Valuation, pricing, portfolio simulation, port-
Mathematical methods least squares problem, finite differ- folio optimization, capital budgeting, hedging,
modeling ence methods, dependence model- price movement prediction, risk modeling, and
ing, Monte-Carlo simulation trend forecasting.
Time- State space modeling, time-series Price prediction, trend forecasting, market
series analysis, spectral analysis, long- movement prediction, IPO prediction, equity-
and memory time-series analysis, non- derivative correlation analysis, change detec-
signal stationary analysis tion, financial crisis analysis, trading strategy
analysis discovery, and cross-market analysis.
Statistical Factor models, stochastic volatil- Market trend forecasting, pricing, valuation,
learning ity models, copula methods, price estimation, VaR forecasting, financial
methods nonparametric methods, Bayesian variable dependency modeling, portfolio per-
networks formance estimate, and cross-market analysis.
Random Random sampling, random walk Abnormal behavior analysis, outlier detec-
methods models, random forest, stochastic tion, market event analysis, market movement
theory, fuzzy set theory, quantum modeling, influence transition analysis, asso-
mechanics ciated account analysis, crowdsourcing mod-
eling, and marketing modeling.
Complexity Systems theory, complex adap- System complexity modeling, market simula-
Complex
science tive systems, chaos theory, ran- tion, market mechanism design, globalization
system
dom fractal theory analysis, crisis contagion, and market infor-
methods
mation flow.
Game Zero-sum game, differential game, Interaction modeling, policy simulation and
theory combinatorial game, evolutionary optimization, regional conflict modeling,
game, Bayesian game mechanism testing, coalition formation, and
cryptocurrency mechanism testing.
Network Linkage analysis, graph methods, Modeling entity movement, community for-
science power law, small worlds, contagion mation, interactions and linkage, influence
theory and contagion propagation; pool manipula-
tion analysis, and analyzing investor relations.
Agent- Multiagent systems, belief-desire- Testing economic hypotheses, simulating
based intention model, reactive model, policies, supply chain relation modeling
model- swarm intelligence, reinforcement and optimization, cooperation analysis, self-
ing learning organization modeling, portfolio optimiza-
tion, and reinforcement learning.
Pattern Frequent itemset mining, sequence Trading behavior analysis, abnormal trading
Classic mining analysis, combined pattern min- analysis, outlier detection, investor relation
analytics methods ing, high-utility pattern mining, analysis, customer profiling, high-utility trad-
and tree pattern, network pattern, ing pattern analysis, and cross-market trading
learning knot pattern, interactive pattern behavior analysis.
Kernel Vector space kernel, tree kernel, Price and market movement prediction, cross-
learning support vector machine, spectral market time series analysis, financial crisis
methods kernel, Fisher kernel, nonlinear analysis, crowdfunding estimate, market de-
kernel, multi-kernel methods pendency modeling, and customer profiling
and classification.
Continued on next page
16

Table 1 – Continued from previous page

DSAI areas Techniques Specific DSAI methods Smart FinTech applications
Event Sequence analysis, Markov chain Financial event analysis, investor behavior
and be- process, high-impact behavior, analysis, price co-movement prediction, ab-
havior high-utility behavior, nonoccur- normal behavior analysis, market exception
analysis ring behavior analysis and change analysis, market event detection,
and group behavior analysis.
Document Language models, case-based rea- Financial event analysis, investor sentiment
analysis soning, statistical language model, analysis, company valuation, financial review
and NLP Bayesian model, latent Dirichlet and auditing analysis, misinformation and ru-
allocation, Transformer, BERT mor analysis, automated question-answering,
and keyword searching.
Model- Probabilistic graphical model, Hypothesis testing, customer clustering, price
based Bayesian networks, expectation- prediction, trend forecasting, index model-
methods maximization model, clustering, ing, event analysis, fraud detection, movement
classification, deep neural models forecasting, valuation, and risk scoring and
prediction.
Social Topic modeling, sentiment analy- Social influence analysis, investment linkage
media sis, emotional analysis, influence analysis, associated account analysis, senti-
analysis analysis, linkage analysis, interac- ment analysis, influence modeling, market and
tion learning price movement, detecting manipulation and
insider trading, and understanding company
branding and development.
Mathematical Nonlinear, stochastic and dynamic Optimizing policies, portfolios, trading strate-
Optimization opti- programming, information theory, gies, VaR, and market performance, cost-
and mization Bayesian optimization benefit optimization.
augmentation Neural Wavelet neural network, genetic Macroeconomic and microeconomic factor
com- neural network, recurrent neural correlation analysis, valuation and pricing
puting network, deep neural network modeling, relation analysis, sequence mod-
methods eling, portfolio optimization, and trend and
movement prediction.
Evolutionary Ant algorithm, genetic program- New product simulation, financial objective
comput- ming, self-organizing map, artifi- optimization, portfolio optimization, market-
ing cial immune system, swarm intel- ing strategy optimization, price and policy
methods ligence, neural-genetic algorithm testing, market risk analysis, and market per-
formance optimization.
Fuzzy Fuzzy set theory, fuzzy logic, fuzzy Modeling market momentum, financial sol-
set neural network, genetic fuzzy logic vency analysis, risk and capital cost modeling,
methods and market uncertainty modeling.
Data Audio augmentation, image and Enhancing low-quality financial data, trans-
augmen- visual augmentation, textual aug- forming data capacity, discovering latent fea-
tation mentation tures in data
Representation Probabilistic model, graph net- Representation of stocks, assets, capital mar-
Advanced
learning work, network embedding, tree kets, portfolios, financial events, behaviors,
analytics
model, neural embedding and financial reports.
and
Text Conceptualization, similarity Text-based trend forecasting of price,
learning
process- learning of terms, tags and market, sentiment and reputation, ques-
ing and phrases, dependency parsing, tion/answering modeling and recommenda-
transla- word embedding, deep neural tion, bilingual or multilingual translation.
tion models, language translation
models, neural translation net-
works
Continued on next page
 17

Table 1 – Continued from previous page

DSAI areas Techniques Specific DSAI methods Smart FinTech applications
Reinforcement Bellman Equation, actor-critic Simulating and optimizing supply/demand of
learning model, Markov dynamic progress, new assets and services, optimizing portfolios
deep Q-network, adversarial and trading strategies, option valuation opti-
reinforcement learning mization.
Deep Convolutional neural network, at- Market modeling, behavior modeling, trading
learning tention network, generative adver- modeling, risk analysis, price and movement
methods sarial network, autoencoder, deep prediction, financial event modeling, cross-
Bayesian network market analysis, cross-sectional modeling
Federated Multi-task learning, cross-domain Offering privacy-preserving and user
and transfer learning, distributed ma- information-protected financial user in-
transfer chine learning, federated learning, teractions, dialogue, response, advertising,
learning privacy-preserving processing, etc. recommendation
Intelligent Intelligent Imagery dialogue, visual dialogue, Automated recognition, dialogue and follow-
conver- identi- verbal/speech recognition, textual up response with financial customers and
sation fication communications, multi-modal stakeholders; recognizing their identities, live-
and and dialogue, IoT devices-based ness, biometric prints, voices, images, and vi-
response interac- recognition, retrieval and ques- sual actions for customer services and commu-
tions tion/answering, etc. nications, etc.
Intent Attraction modeling, intent mod- Understanding customer intention, evolving
and eling, emotion learning, sentiment demand and negative feedback on financial
emotion analysis, feedback analysis, etc. products and services, tailoring active inter-
learning actions with providers, and recommending fu-
ture services, etc.
Smart Secure and efficient encryption, Secure, distributed, efficient and risk-free in-
System
blockchain distributed ledger technology, frastructure, cryptocurrencies and financial
intelli-
smart contract protocols, user services (e.g., bitcoin portfolio management),
gence
behavior analysis, anti-intrusion etc.
enhancement
and fragility, etc.
Intelligent Identity and object recognition, Offering and consuming financial businesses
authen- object segmentation, signal pro- and services (e.g., online and short-messaging-
tication cessing, visual analytics, and tex- based commerce) by mobile applications, in
tual translation, personal (re- the IoT environment or WiFi networks, etc.
)identification, etc.
Intelligent Live recommendation, cold- Customizing and suggesting the right prod-
recom- start recommendation, tailed ucts and services to the right people at the
menda- and sparse recommendation, right time, etc.
tion personalized recommendation,
cross-domain recommendation,
next-basket/item recommenda-
tion, etc.
Automated AutoML, active and adaptive Automatically fusing, processing and analyz-
learning learning, batch processing, au- ing heterogeneous financial information; auto-
tomated data integration and mated performance optimization by optimal
processing, architecture search, model search and updating, etc.
context-aware learning, etc.
Ethical Feature coupling learning, rep- Offering unbiased, accountable and user-
FinTech resentation disentanglement, ex- explainable products, prices, services, and rec-
plainable learning, visualization, ommendations, etc.
unbiased learning, fairness learn-
ing, etc.
Continued on next page
18

Table 1 – Continued from previous page

DSAI areas Techniques Specific DSAI methods Smart FinTech applications
Parallel Evolutionary neural models, deep Price and market movement fore-
Hybrid ensemble Bayesian model, copula graph casting, multi-aspect risk analysis,
methods neural network, combining com- macro/microeconomic factor analysis, fi-
plexity science and game theory nancial event detection, customer profiling,
blockchain and cryptocurrency modeling.
Sequential Time-series analysis plus classifi- Financial review-based fraud detection,
and hier- cation, macro/microeconomic de- macroeconomic influence on market move-
archical pendency modeling, deep sequen- ment, social media impact on price movement,
hy- tial modeling-based event detec- epidemic evolution, and impact on index.
bridiza- tion
tion
Cross- Deep multi-time series analysis, Psychological factor and irrational market be-
disciplinary copula deep models, autoregres- havior analysis, behavioral finance, misinfor-
hy- sion deep model, behavioral eco- mation and mispricing on market inefficiency,
bridiza- nomics and finance financial ethics modeling, interpretable finan-
tion cial modeling.
Behavioral Prospect theory, nudge theory, Modeling investment rationality, irrational
eco- natural experiment, experimental behaviors, mispricing, market efficiency, men-
nomics economics, behavior informatics, tal activities, decision-making process, inten-
and intention learning, next-best ac- tion, emotion, and cognitive activities.
finance tion modeling
 19

References 26. J. B. Heaton, N. G. Polson, and J. H. Witte. Deep learn-

ing for finance: Deep portfolios. Applied Stochastic Mod-
1. T. G. Andersen, R. A. Davis, J.-P. Kreiß, and T. V. els in Business and Industry, 33:3–12, 2017.
Mikosch. Handbook of financial time series. Springer, 27. Y. Hilpisch. Artificial Intelligence in Finance. OReilly,
2009. 2020.
2. H. Arslanian and F. Fischer. The Future of Finance: 28. International Telecommunication Union. Assessing the
The Impact of FinTech, AI, and Crypto on Financial economic impact of artificial intelligence, 2018. Issue Pa-
Services. Palgrave Macmillan, 2019. per No. 1, September 2018.
3. M. Baddeley. Behavioural Economics and Finance. 29. B. K. Iwana and S. Uchida. An empirical survey of data
Routledge, 2013. augmentation for time series classification with neural
4. A. Brahme and U. Bhadade. Effect of various vi- networks. arXiv preprint arXiv:2007.15951, 2020.
sual speech units on language identification using 30. G. Jeong and H. Y. Kim. Improving financial trading
visual speech recognition. Int. J. Image Graph., decisions using deep q-learning: Predicting the number
20(4):2050029:1–2050029:27, 2020. of shares, action strategies, and transfer learning. Expert
5. L. Broemeling. Bayesian Analysis of Time Series. Chap- Syst. Appl., 117:125–138, 2019.
man and Hall/CRC, 2019. 31. C. Kearney and S. Liu. Textual sentiment in finance: A
6. B. G. Buchanan. Artificial intelligence in finance. The survey of methods and models. International Review of
Allen Turing Institute, 2019. Financial Analysis, 33:171–185, 2013.
7. L. Cao. Data science: challenges and directions. Com- 32. K. Khashanah and T. Alsulaiman. Network theory and
mun. ACM, 60(8):59–68, 2017. behavioral finance in a heterogeneous market environ-
8. L. Cao. AI in finance: A review. pages 1–35, 2020. ment. Complexity, 21(S2):530–554, 2016.
https://ssrn.com/abstract=3647625. 33. B. Kovalerchuk and E. E. Vityaev. Data Mining in
9. L. Cao. AI in finance: Challenges, tech- Finance: Advances in relational and hybrid methods.
niques and opportunities. pages 1–40, 2021. Kluwer Academic Publishers, 2000.
https://ssrn.com/abstract=3869625. 34. K. Lei, B. Zhang, Y. Li, M. Yang, and Y. Shen. Time-
10. K. Chatterjee and W. F. S. (Eds.). Game Theory and driven feature-aware jointly deep reinforcement learning
Business Applications. Springer, 2001. for financial signal representation and algorithmic trad-
11. G. Cornuéjols, J. Peña, and R. Tütüncü. Optimization ing. Expert Syst. Appl., 140, 2020.
Methods in Finance (2nd ed.). Cambridge University 35. B. Li and S. C. H. Hoi. Online portfolio selection: A
Press, 2018. survey. ACM Comput. Surv., 46(3):1–36, 2014.
12. P. Craja, A. Kim, and S. Lessmann. Deep learning for 36. H. Li, C. Huang, and L. Gu. Image pattern recognition in
detecting financial statement fraud. Decis. Support Syst., identification of financial bills risk management. Neural
139:113421, 2020. Comput. Appl., 33(3):867–876, 2021.
13. V. Dhar and R. M. Stein. Fintech platforms and strategy. 37. Y. Li, P. Ni, and V. Chang. Application of deep rein-
Commun. ACM, 60(10):32–35, 2017. forcement learning in stock trading strategies and stock
14. C. Doloc. Applications of Computational Intelligence in forecasting. Computing, 102(6):1305–1322, 2020.
Data-Driven Trading. John Wiley & Sons, 2020. 38. Z. Li, J. Kiseleva, and M. de Rijke. Improving response
15. C. L. Dunis, P. W. Middleton, A. Karathanasopolous, quality with backward reasoning in open-domain dia-
and K. Theofilatos. Artificial Intelligence in Financial logue systems. In F. Diaz, C. Shah, T. Suel, P. Castells,
Markets. Springer, 2019. R. Jones, and T. Sakai, editors, SIGIR’2021, pages 1940–
16. N. Ehrentreich. Agent-Based Modeling. Springer, 2008. 1944, 2021.
17. M. Firdaus, N. Thakur, and A. Ekbal. Multidm- 39. T. Lynn, J. G. Mooney, P. Rosati, and M. Cummins. Dis-
gcn: Aspect-guided response generation in multi-domain rupting finance: Fintech and strategy in the 21st century.
multi-modal dialogue system using graph convolution Palgrave Pivot, 2019.
network. In EMNLP’2020, pages 2318–2328, 2020. 40. T. L. Meng and M. Khushi. Reinforcement learning in
18. T. Fischer. News reaction in financial markets within financial markets. Data, 4(3):110:1–17, 2019.
a behavioral finance model with heterogeneous agents. 41. G. Mitra and L. Mitra. The Handbook of News Analytics
Algorithmic Finance, 1(2):123–139, 2011. in Finance. Wiley, 2012.
19. B. C. M. Fung, K. Wang, R. Chen, and P. S. Yu. Privacy- 42. D. S. Modha, R. Ananthanarayanan, S. K. Esser, A. Ndi-
preserving data publishing: A survey of recent develop- rango, A. J. Sherbondy, and R. Singh. Cognitive com-
ments. ACM Comput. Surv., 42(4):14:1–14:53, 2010. puting. Communications of the ACM, 54(8):62–71, 2011.
20. J. E. Gentle, W. K. Hardle, and Y. Mori. Handbook of 43. K. Nakagawa, M. Abe, and J. Komiyama. RIC-NN: A
Computational Finance. Springer, 2012. robust transferable deep learning framework for cross-
21. M. Gilli, D. Maringer, and E. Schumann. Numerical sectional investment strategy. In DSAA’2020, pages 370–
Methods and Optimization in Finance. Academic Press, 379. IEEE, 2020.
2019. 44. B. Nicoletti and W. Nicoletti. Future of FinTech.
22. I. Goodfellow, Y. Bengio, and A. Courville. Deep Learn- Springer, 2017.
ing. The MIT Press, 2016. 45. OECD. OECD Blockchain Primer. OECD,
23. I. Hadi. Intelligent authentication for identity and access 2018. https://www.oecd.org/finance/OECD-Blockchain-
management: a review paper. Iraqi Journal for Comput- Primer.pdf.
ers and Informatics, 45(1):6–10, 2019. 46. A. M. Ö zbayoglu, M. U. Gudelek, and O. B. Sezer. Deep
24. L. Hamill and N. Gilbert. Agent-Based Modelling in Eco- learning for financial applications : A survey. CoRR,
nomics. John Wiley & Sons, 2016. abs/2002.05786, 2020.
25. X. He, K. Zhao, and X. Chu. Automl: A survey of the 47. W. Pan and Q. Yang. Transfer learning in heterogeneous
state-of-the-art. Knowledge-Based Systems, 212:106622, collaborative filtering domains. Artif. Intell., 197:39–55,
2021. 2013.
 20

48. Y. Qi and J. Xiao. Fintech: AI powers financial services 55. Y. Wei and V. Chaudhary. The directionality function
to improve people’s lives. Commun. ACM, 61(11):65–69, defect of performance evaluation method in regression
2018. neural network for stock price prediction. In DSAA’2020,
49. L. Ryll, M. E. Barton, B. Z. Zhang, R. J. McWaters, pages 769–770, 2020.
E. Schizas, R. Hao, K. Bear, M. Preziuso, E. Seger, 56. Q. Yang, Y. Liu, T. Chen, and Y. Tong. Federated ma-
R. Wardrop, P. R. Rau, P. Debata, P. Rowan, N. Adams, chine learning: Concept and applications. ACM Trans.
M. Gray, and N. Yerolemou. Transforming paradigms: A Intell. Syst. Technol., 10(2):12:1–12:19, 2019.
global ai in financial services survey, 2020. 57. Q. Yang, Y. Liu, Y. Cheng, Y. Kang, T. Chen, and H. Yu.
50. M. Sewak. Deep Reinforcement Learning - Frontiers of Federated Learning. Synthesis Lectures on Artificial In-
Artificial Intelligence. Springer, 2019. telligence and Machine Learning. Morgan & Claypool
51. O. B. Sezer, M. U. Gudelek, and A. M. Ö zbayoglu. Finan- Publishers, 2019.
cial time series forecasting with deep learning : A system- 58. J. Zhang and Q. He. Dynamic cross-market volatil-
atic literature review: 2005-2019. CoRR, abs/1911.13288, ity spillover based on MSV model: Evidence from bit-
2019. coin, gold, crude oil, and stock markets. Complex.,
52. J. Sirignano and R. Cont. Universal features of price 2021:9912418:1–9912418:8, 2021.
formation in financial markets: perspectives from deep 59. J. Zhang and Y. Zhuang. Cross-market infection research
learning. Quantitative Finance, 19:1449–1459, 2019. on stock herding behavior based on DGC-MSV mod-
53. G. Spanos and L. Angelis. The impact of information se- els and bayesian network. Complex., 2021:6645151:1–
curity events to the stock market: A systematic literature 6645151:8, 2021.
review. Comput. Secur., 58:216–229, 2016. 60. X. Zhang, Y. Li, S. Wang, B. Fang, and P. S. Yu. En-
54. M. M. Waldrop. Complexity: The Emerging Science at hancing stock market prediction with extended coupled
the Edge of Order and Chaos. Simon & Schuster Paper- hidden markov model over multi-sourced data. Knowl.
backs, 1992. Inf. Syst., 61(2):1071–1090, 2019.