A Comprehensive Investigation of the Advancements for the

Computational Advertising Research

Zihang Zhang

School of Accounting, Zhongnan University of Economics and Law, Donghu High-tech Zone, Wuhan, China

Keywords: Computational Advertising, Digital Era, Data-Driven.

Abstract: In the contemporary digital landscape, Computational Advertising (CA) has ascended as a pivotal paradigm,

harnessing the power of data and algorithms to orchestrate highly targeted and personalized advertising

campaigns. This scholarly article offers a comprehensive analysis of the current state of CA, interrogating its

foundational theoretical frameworks, examining practical applications, and delineating the most recent

advancements in academic research. The study underscores the amalgamation of multidisciplinary

methodologies, with a particular emphasis on real-time bidding mechanisms, sophisticated personalized

advertising technologies, and the strategic employment of big data analytics and user profiling for optimized

ad delivery. The article critically engages with the ethical implications of CA's rapid proliferation,

highlighting concerns surrounding user privacy and data security. In light of these challenges, the research

proposes future directions that seek to harmonize technological innovation with the protection of consumer

rights, thereby fostering the sustainable evolution of the advertising industry. Through an academic lens, this

article scrutinizes the intricate interplay between technological advancements, ethical considerations, and the

evolving dynamics of consumer engagement, contributing to the ongoing discourse within the field and

shaping future research trajectories.

1 INTRODUCTION

With the advent of the digital era, Computational

Advertising (CA) has become an integral component

of the advertising industry. This data and algorithm-

driven approach to advertising has revolutionized the

way brands connect with their audience by enabling

highly targeted and personalized messaging.

However, the rapid evolution of Computational

Advertising has also given rise to a myriad of ethical

risks and challenges. These concerns not only

threaten consumer privacy and data security but also

pose significant questions regarding the sustainable

development of the advertising domain.

The digital era has ushered in a transformative

phase in the field of advertising, with computational

advertising emerging as a key driver of innovation. In

their seminal article (Helberger et al., 2020). Jisu Huh

and colleagues offer a definitive guide that outlines

the theoretical foundations and practical usage within

this evolving field. This paper methodically

introduces the idea of computational advertising,

showcasing it as a technology-focused strategy using

https://orcid.org/0009-0000-4113-4468

advanced algorithms and computing capabilities to

boost the effectiveness of advertising tactics.

Huh and Malthouse laid the groundwork with a

definition that succinctly captures the core of CA

(Huh et al., 2020), subsequently documenting their

development from the initial phase of mail-order

systems to the modern digital landscape marked by

the advent of big data and artificial intelligence. The

evolution of this historical context is crucial to fully

understand CA's present condition, a cornerstone in

advertising where it allows for immediate targeting

and customization across various forms of media.

The document also clarifies the future by

forecasting forthcoming challenges and opportunities

for CA. The draft creates a future-oriented and

empirically-based research agenda about safety for

consumers, competitive equity, and ethical

technology use, and this proves that pro-innovation

policies must be nurtured. (Huh et al., 2020).

The article lays a foundation for such studies by

assessing current situation in the field of CA research

and determining its effects on publications and

academic practices in general. This article is a

350

Zhang, Z.

A Comprehensive Investigation of the Advancements for the Computational Advertising Research.

DOI: 10.5220/0013230400004568

In Proceedings of the 1st International Conference on E-commerce and Artiﬁcial Intelligence (ECAI 2024), pages 350-354

ISBN: 978-989-758-726-9

thorough literature review, pointing out a few

promising possibilities and research tracks which

should be taken into consideration in future

investigations. The paper demonstrates the CA

concept so well by providing fresh ideas in terms of

cross-disciplinary research, on strategic advertising,

and on transforming the brand's relationship with its

audience. Marketing specifically in the digital realm

within the last decade is becoming an increasing topic

of interest and merits careful scrutiny to provide

current discussions within the academic community.

2 METHOD

As the online ad research reaches new heights, the

provocative techniques from various disciplines have

been fused to examine the complexity of the

computational advertising. Informed by the latest

findings, the current research trend is moving into a

more all-round and multidisciplinary direction. The

core computational advertising activity is the

discovery of effective ads suited for a certain person

in a defined surrounding. The journey requires the

unity of many domains in which a search, text

analysis, information retrieval, statistical modeling,

machine learning, categorization protocols,

optimization topics, and microeconomic perspectives

are involved.

In the past few years, computational advertising

research has seen dynamic development with some

main topics being real-time bidding (RTB) adopted

effectively, heterogeneous ad technologies

developed, and big data and user profiling for

accurate ad targeting deployed. The development

goes beyond simple algorithm improvement to

include consonance with ad logos, studying the role

of tech solutions to increase effectiveness of

coordinated marketing strategies as well as effective

personalization across all customer touchpoints.

2.1 Real-Time Bidding (RTB)

Real-Time Bidding stands as a significant

advancement in the realms of display and mobile ads

in recent times. RTB enables the accurate targeting of

users according to their habits and choices via an

immediate auction process, thus improving both the

efficiency and the conversion rates of advertisements.

This method, which relies on data and behavior for

advertising dissemination, offers a more tailored and

effective strategy than conventional keyword or

content matching techniques (Wang et al., 2016).

RTB has radically transformed digital marketing,

offering fresh avenues for research in automation,

integration, and optimization (Wang et al., 2015).

Recent algorithmic frameworks in RTB are chiefly

oriented towards these areas:

2.1.1 The Integration of Deep Learning and

Reinforcement Learning

In 2023, a study suggested a unique strategy that

integrates deep learning with reinforcement learning

methods to enhance the effectiveness and precision of

RTB. The technique utilizes deep neural networks for

forecasting auction specifics and market costs, and

reinforcement learning algorithms ascertain the ideal

bid price (Sharma, 2023).

2.1.2 Statistical Arbitrage Mining (SAM)

In 2015, research unveiled SAMer, a meta-bidder

employing statistical arbitrage to optimize anticipated

net gains. SAMer pursues the best bids by enhancing

feature optimization and utilizes traditional data-

driven learning techniques to evaluate prospective

income and expenses (Zhang et al., 2015).

2.1.3 Conversion Rate Prediction Methods

Combining Regression and Triplet

Learning

A 2018 study suggested a method (CRT) combining

regression loss and triplet ranking loss, targeting

precise ranking figures and accurate regression

calculations to refine buyer conversion rate forecasts

in RTB (Shan et al., 2018).

2.1.4 SKOTT Optimization Layer

In 2018, research unveiled SKOTT, an algorithm

aimed at enhancing specific key performance

indicators (KPIs). This is achieved by efficiently

setting up DSPs and encouraging them to compete

with one another. SKOTT, a sophisticated iterative

algorithm reliant on gradient descent, tackles

challenges like budget distribution, calculating

anticipated average bids, and averting underdelivery.

2.2 Machine Learning and Statistical

Models

Analytics models based on machine learning have

become an integral aspect of this computational

advertising domain. Examining past patterns of data,

models may be trained to predict user behavior under

certain advertisement conditions. Thus, selection of

ads is made better, and they are distributed through

better methods. The Deep Image CTR Model (DICM)

includes the characteristics of the image content that

A Comprehensive Investigation of the Advancements for the Computational Advertising Research

351

impacts the user behavior and the advertisement

creativity to engage users.(Ge et al., 2017).

To enhance ad distribution's efficiency and

effectiveness, the immediate assessment and fine-

tuning of advertisements are also crucial. The process

encompasses applying statistical feature analysis

techniques to gather statistical data and regression

analysis for the advertisement's specific placement

effectiveness, coupled with big data management

systems for managing crowds and computing

profiles, aiming to tailor advertisement placements to

meet real business requirements (Wang et al., 2021).

Regarding the training and enhancement methods

for forecasting user reactions to certain adverts in

computational advertising studies, these usually

encompass several stages.

2.2.1 Data Collection and Preprocessing

Initially, an extensive collection of data on user

behavior and advertising characteristics is required.

Such information could encompass the click patterns,

buying patterns, and browsing patterns of users.

Moreover, techniques for recognizing sentiments are

applicable in analyzing how the ad content

emotionally responds, like the attributes

Convolutional Neural Networks (CNNs) identify to

capture the emotional essence in advertisements

(Gharibshah et al., 2021).

2.2.2 Feature Engineering

After data preprocessing, the next stage is the feature

engineering. This step involves extracting useful

information from the raw data in order to construct

features that can represent users' responses to

advertisements. For example, deep learning models

such as Convolutional Neural Networks (CNN) and

Long Short-Term Memory Networks (LSTM) can be

used to automatically extract and learn these features.

2.2.3 Model Selection and Training

Choosing a suitable machine learning model is the

next imperative step. Popularly employed methods

for supervised learning include logistic regression,

support vector machines, decision trees, random

forests, and convolutional deep learning. For

unsupervised learning, long short-term memory

networks can also be mentioned. Such models should

be trained with historical data to be able to make

predictions about the possible responses of a user to

certain advertisements.

2.2.4 Model Optimization and Validation

After finalizing the training phase, it's necessary to

evaluate and validate the models using fresh datasets

to determine their forecast accuracy. This process

usually includes methods like cross-validation to

guarantee the model's capacity for generalization.

Furthermore, techniques like A/B testing serve as

tools for assessing various advertising tactics in real-

time scenarios.

2.2.5 Real-time Applications and Feedback

Loops

Practically, it's imperative for machine learning

models to handle live data and persistently modify

their forecasts in response to novel inputs. Such a

model must be capable of being flexible and

adaptable to sustain its effectiveness amidst evolving

market dynamics (Rajan, 2018).

3 DISCUSSIONS

3.1 Limitations and Challenges in the

Field

Although significant progress has been made, there

are still some limitations and challenges in this field.

3.1.1 Explainability

In computational advertising, the algorithms are often

so intricate that they are not comprehensible. This is

the root cause of opacity in the decision-making

process supported by those algorithms. To this end,

this lack of transparency not only destabilizes the

trust of advertisers but also makes it difficult for

consumers to perceive the reason why they are

presented with specific ads.

3.1.2 Applicability

Research and applications of computational

advertising are often highly dependent on the quality

and availability of data. However, issues such as

privacy leakage and data bias may arise during data

collection and processing, which can limit the

applicability and effectiveness of computational

advertising techniques (Gao et al., 2023).

3.1.3 Privacy

The collection and use of user data is inevitable in

computational advertising. However, it also raises

significant concerns about user privacy protection.

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

352

How to protect user privacy while improving

advertising effectiveness has become an urgent issue.

(Helberger et al., 2020).

3.2 Future Prospects

Some several possible solutions can be considered in

this case to solve the limitations and challenges

mentioned above.

3.2.1 Adoption of a User-centered Design

Framework

According to the study carried out by (Hosain et al.,

2023), the development of transparent and

interpretable AI systems requires interdisciplinary

collaboration, including computer science, artificial

intelligence, ethics, law, and social sciences. The

design should be user-centered to ensure that the

system is not only technically feasible but also

socially and ethically acceptable.

3.2.2 Adopt a User-centered Design

Framework

Scott M proposed the SHapley Additive exPlanations

(SHAP) framework (Lundberg et al., 2017), a unified

approach to interpreting the predictions of complex

models. By assigning importance values to each

feature, SHAP helps users understand the model's

decision-making process, which improves the

transparency and interpretability of the model.

3.2.3 Maintenance of Localized User

Profiles for Devices

Traditional recommender systems rely on server-side

large-scale vector computation, which is not only

inefficient but also may compromise user privacy. A

new approach is to store user profiles entirely on the

user's device and obtain appropriate

recommendations from web portals in an encrypted

way, which can effectively protect user privacy

(Tulabandhula et al., 2017).

3.2.4 Adoption of K-anonymization

Techniques

K-Anonymity is a data protection model that ensures

that each individual's information cannot be

individually distinguished from the anonymized

dataset (Sweeney, 2002). This approach effectively

minimizes the risk of personal information leakage

while allowing advertising systems to continue to use

this data for effective user targeting.

4 CONCLUSIONS

Computational advertising has turned into the main

player in the digital generation, which governs by

data and algorithms to target more precisely the

needed audience. But it also runs into ethical matters

such as privacy and security of data. Research has

crossed boundaries with - especially - use of

interdisciplinary methods like real-time bidding and

machine learning, which increased advertisement

efficacy and performance. Yet, there are still various

problems, namely algorithm opacity and data privacy.

In the future, we will specifically cultivate

explainable algorithms, data privacy protection, and

intelligent advertisements. It hopes to come up with a

unique development pattern that balances

technological innovation with consumer privacy

rights.

REFERENCES

Gao, B., Wang, Y., Xie, H., Hu, Y., & Hu, Y. 2023.

Artificial Intelligence in Advertising: Advancements,

Challenges, and Ethical Considerations in Targeting,

Personalization, Content Creation, and Ad

Optimization. Sage Open, 13(4).

Ge, T., & Wu, X. 2021. Accurate Delivery of Online

Advertising and the Evaluation of Advertising Effect

Based on Big Data Technology. Mobile Information

Systems, 2021, 1-10.

Gharibshah, Z., & Zhu, X. 2021. User response prediction

in online advertising. arXiv.

Helberger, N., Huh, J., Milne, G., Strycharz, J., &

Sundaram, H. 2020. Macro and Exogenous Factors in

Computational Advertising: Key Issues and New

Research Directions. Journal of Advertising, 49(4),

377–393.

Hosain, Md. T., Anik, M. H., Rafi, S., Tabassum, R., INSİA,

K., & SIDDIKY, Md. 2023. Path To Gain Functional

Transparency In Artificial Intelligence With

Meaningful Explainability. Journal of Metaverse, 3,

166-180.

Huh, J., & Malthouse, E. C. 2020. Advancing

Computational Advertising: Conceptualization of the

Field and Future Directions. Journal of Advertising,

49(4), 367–376.

Latanya Sweeney. 2002. K-anonymity: a model for

protecting privacy. Int. J. Uncertain. Fuzziness Knowl.-

Based Syst., 10(5), 557–570.

Lundberg, S., & Lee, S.-I. 2017. A Unified Approach to

Interpreting Model Predictions. ArXiv.org.

Sharma, P. 2023. Improving Real-Time Bidding in Online

Advertising using Markov Decision Processes and

Machine Learning Techniques. International Journal of

Advanced Engineering and Nano Technology, 10(7),

1–8.

A Comprehensive Investigation of the Advancements for the Computational Advertising Research

353

Suju Rajan. 2018. Computational Advertising at Scale. In

Proceedings of the 24th ACM SIGKDD International

Conference on Knowledge Discovery & Data Mining

(KDD '18), 2875.

Tulabandhula, T., Vaya, S., & Dhar, A. 2017. Privacy-

preserving Targeted Advertising. Journal of Business

Analytics, 3.

Wang, J., & Yuan, S. 2015. Real-Time Bidding: A New

Frontier of Computational Advertising Research.

Proceedings of the Eighth ACM International

Conference on Web Search and Data Mining.

Wang, J., Zhang, W., & Yuan, S. 2016. Display Advertising

with Real-Time Bidding (RTB) and Behavioural

Targeting. arXiv.

Zhang, W., & Wang, J. 2015. Statistical Arbitrage Mining

for Display Advertising. In Proceedings of the 21th

ACM SIGKDD International Conference on

Knowledge Discovery and Data Mining (KDD '15),

1465–1474.

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

354