ENHANCING PERSONALIZED
MARKETING STRATEGIES THROUGH
MULTIMODAL CUSTOMER
SEGMENTATION
RASHMITH R ALVA 4NM21IS123
RIDHI SHETTY 4NM21IS126
SAKSHI SHETTY 4NM21IS133
SHARANYA DINAKAR 4NM21IS151
GUIDE NAME-DR BOLA SUNIL KAMATH
Agenda
● Literature Survey
● Literature survey in detail
● Conclusion
● Motivation of the project
● Problem statement
Literature Survey
● Customer segmentation using K-means clustering and the
adaptive particle swarm optimization algorithm-2021
● Multimodal deep learning of word-of-mouth text and demographics to
predict customer rating: Handling consumer heterogeneity in
marketing-2024
● Customer Segmentation Using Artificial Neural Network-2023
Paper 1:
Title: Customer segmentation using K-means clustering and the adaptive
particle swarm optimization algorithm-2021
Author: Yue Li , Xiaoquan Chu , Dong Tian , Jianying Feng , Weisong Mu
Source:
https://www.sciencedirect.com/science/article/abs/pii/S1568494621008462
Year of Publication: 2021
Main Features:
1. Adaptive Learning Particle Swarm Optimization (ALPSO)
Algorithm:
To improve the PSO algorithm's optimization accuracy, a new
technique called ALPSO is presented. In doing so, the problem
of PSO entering local optima is addressed.
2. KM-ALPSO:
To lessen the reliance of the K-means clustering method on
initial cluster centers, a novel strategy called KM-ALPSO is put
forth. This approach seeks to increase clustering accuracy by
employing ALPSO to optimize the K-means cluster centers.
3. IKM-ALPSO:
The IKM-ALPSO algorithm is a KM-ALPSO variant that is
especially made to handle mixed data for clustering. The
problem of grouping datasets with both categorical and
numerical attributes is addressed by this.
4. Performance Evaluation:
To show that the suggested approaches are better than the
current models, comparative experiments are carried out. The
algorithms' efficacy is evaluated using a variety of benchmark
datasets, including UCI datasets.
Areas of Improvement:
1. Optimization Accuracy:
By incorporating adaptive learning mechanisms into the ALPSO
algorithm, the main goal is to increase the optimization accuracy of
PSO.
2. Cutting Down on Initial Center Dependency:
Another important area for improvement is addressing the
dependence of K-means clustering on initial cluster centers. To
improve the clustering outcomes, the KM-ALPSO approach seeks to
optimize these initial centers.
3. Handling Mixed Data:
The capacity to group datasets with a combination of categorical and
numerical attributes is a crucial component. Because it is designed to
tackle this particular problem, the IKM-ALPSO method is more
adaptable in practical settings.
4. Customer Segmentation:
Using these enhanced clustering algorithms for customer
segmentation jobs is the ultimate goal. Businesses can improve their
competitiveness in the market by better understanding their
consumers' demands and customizing products and services by
precisely segmenting their client base based on their behaviors and
traits.
Paper 2:
Title: Multimodal deep learning of word-of-mouth text and demographics to predict
customer rating: Handling consumer heterogeneity in marketing-2024
Author:Junichiro Niimi
Source: https://arxiv.org/pdf/
Year of Publication: 2024
Main Features:
1. Consumer Heterogeneity Consideration:
In addressing the problem of consumer heterogeneity in marketing,
the paper highlights how crucial it is to comprehend the internal or
psychological variations among consumers that behavioral logs are
unable to record.
2. Use of Online Product Reviews:
It draws attention to the importance of online product reviews as a
useful data source for identifying subtle variations in customer
behavior.
3. Multimodal Learning Approach:
To build a product evaluation model that takes customer
heterogeneity into account, the study combines textual data from
online product evaluations with conventional cross-sectional data
(such as consumer profile information).
4. Integration of Machine Learning Techniques:
To handle numerous datasets at once and provide joint
representations from multiple modalities, the article makes use of
recent breakthroughs in machine learning techniques, particularly
large-scale language models like BERT.
Areas of Improvement:
1. Model Generalization and Performance:
Although the paper shows increases in prediction accuracy through
multimodal learning, more research into variables affecting
multimodal model performance—such as the problem to be solved,
relationships between modalities, and data quality—could improve
understanding and direct the development of new models.
2. Pre-trained Model Optimization:
While comparing the effectiveness of several pre-trained BERT
models, the study does not go into great detail about fine-tuning these
models for the particular task at hand. Prediction accuracy may be
increased by more research into model designs and fine-tuning
strategies.
3. Endogeneity Consideration:
In order to fully comprehend the relationships between consumer
heterogeneity, online product reviews, and consumer behavior, the
paper highlights the necessity of carefully examining potential
endogeneity between variables in the analysis. This suggests the need
for more thorough causal modeling approaches.
Paper 3:
Title: Customer Segmentation Using Artificial Neural Network
Author: Vikas Mendhe, Avinash Sonule
Source:https://www.researchgate.net/publication/
377749687_Customer_Segmentation_Using_Artificial_Neural_Network
Year of Publication: 2024
Main Features:
1. Customer Segmentation:
Based on a range of characteristics and variables, including age,
products, deposits, and demographics, the study divides its clientele
into discrete groups using customer segmentation methodologies.
2. Machine Learning Model:
To train and evaluate customer data for segmentation, it uses an
artificial neural network called a Multi-layer Perceptron (MLP).
3. Data study:
In order to customize features for training the MLP model and
achieve a good generalization of the customer segmentation strategy,
the system performs a thorough study of data on customer-related
data tables.
4. Automated Decision Mechanism:
Following training, the model can evaluate customer data
automatically and decide whether to promote or keep a client in the
same position, potentially increasing customer happiness.
Areas of Improvement:
1. Enhanced Feature Engineering:
Further refinement of features used for training the MLP model could
potentially improve its accuracy and generalization.
2. Hyperparameter tuning:
While grid search is mentioned in the study as a method of
hyperparameter tuning, performance may be improved by
investigating more complex optimization strategies or fine-tuning
hyperparameters.
3. Evaluation Metrics:
Although the paper cites a high accuracy score of 91.7%, extra
metrics like F1-score, precision, and recall could offer a more
thorough picture of the model's performance.
4. Real-time Integration:
By including the suggested system into a real-time data framework, it
will be possible to continuously analyze customer-related data, which
will facilitate quicker and more flexible decision-making.
5. Extension to Other Industries:
Although the article concentrates on the banking industry, examining
the suggested approach's suitability for other industries could
increase its influence and usefulness.
Paper 4:
Title: A Mathematical Model for Customer Segmentation Leveraging Deep
Learning, Explainable AI, and RFM Analysis in Targeted Marketing
Author: Fatma M. Talaat, Abdussalam Aljadani, Bshair Alharthi,
Mohammed A. Farsi, Mahmoud Badawy, Mostafa Elhousseini
Source: https://www.mdpi.com/2227-7390/11/18/3930
Year of Publication: 2023
Main Features:
1. Mathematical DeepLimeSeg: Introduced a mathematical model for the
DeepLimeSeg algorithm, blending deep learning and explainable AI for
customer segmentation.
2. Data Integration: Utilized demographics, behavioral patterns,
and purchase history within DeepLimeSeg’s mathematical framework.
3. Lime-based Explainability: Incorporated a mathematically detailed
Lime module for clear segmentation explanations, enhancing marketing
strategies.
4. Validation and Comparison: Validated DeepLimeSeg using real-world
datasets and compared it against the mathematically driven RFM
analysis.
5. Strategic Insights: Offered businesses a mathematically grounded
approach for informed marketing and product decisions.
6. Algorithmic Limitations: Discussed assumptions, overfitting,
and underfitting, highlighting areas of improvement.
Areas of Improvement:
1. Detailed Methodology Description: The paper briefly mentions the use
of RFM analysis and the DeepLimeSeg algorithm for customer segmentation
but lacks detailed explanations of these methodologies. Providing a more
thorough description, including step-by-step implementation and specific
parameters used, would improve understanding.
2. Explanation of Evaluation Metrics: While the paper mentions the use
of evaluation metrics such as Mean Squared Error (MSE), Mean Absolute
Error (MAE), and R-squared, it could provide a more comprehensive
explanation of these metrics and their relevance to customer segmentation
tasks. This would help readers interpret the model performance more
effectively.
3. Limitations and Future Directions: While the paper acknowledges
some limitations, such as dataset specificity and computational resources, it
could expand on this discussion. Providing more thorough insights into the
challenges faced and potential future research directions would add depth
to the paper.
4. Discussion of Results: The paper briefly discusses the results of the
evaluation, including MSE and R-squared values, but lacks a detailed
analysis. Providing a deeper discussion, including comparisons between RFM
analysis and DeepLimeSeg, and insights gained from the segmentation
results would enhance the paper's impact.
Paper 5:
Title: Estimating Customer Segmentation based on Customer Lifetime Value
Using Two-Stage Clustering Method
Author: Prandya Paramita Pramono,Isti Surjandari, Enrico Laoh
Source: https://ieeexplore.ieee.org/document/8887704
Year of Publication: 2019
Main Features:
1. The paper highlights the importance of Customer Relationship
Management (CRM) and the significance of Customer Lifetime Value (CLV)
in developing targeted marketing strategies.
2. The methodology involves data preparation, clustering using Ward's
method and K-Means clustering, calculation of CLV, and ranking of
customer segments based on CLV scores. Fuzzy Analytical Hierarchy
Process (FAHP) is utilized for weighting variables.
3. The paper employs hierarchical clustering (Ward's method) to
determine the initial number of clusters and K-Means clustering for final
segmentation. The Davies Bouldin Index (DBI) is used for evaluating
clustering results, and ANOVA analysis is conducted to assess the
significance of variables. Fuzzy AHP is employed for determining variable
weights.
4. The paper interprets the clusters based on their characteristics derived
from the CLV analysis. It categorizes segments into core customers, new
customers, and uncertain lost customers, providing insights into their
behavior and potential value to the company.
7. The paper concludes by summarizing the findings and emphasizing
the importance of focusing on high-value customer segments. It also
highlights potential future research directions, such as extending the
analysis to identify frequently purchased products for each segment.
8. The paper acknowledges the funding support from Universitas
Indonesia and the contributions of experts involved in the FAHP process.
Areas of Improvement:
1. While the paper describes the methodology used, further discussion on
why specific techniques were chosen over alternatives could enhance
understanding and justify methodological decisions.
2. Providing insights into data quality issues, challenges faced during pre-
processing, and strategies employed to address them would add depth to
the methodology section.
3. Addressing ethical considerations related to customer data privacy,
consent, and confidentiality would strengthen the ethical framework of the
study.
4. Discussing the generalizability of the findings beyond the beauty industry
sector in Indonesia would enhance the applicability of the research to a
broader context.
Paper 6:
Title: A Study on Heuristic and Non-Heuristic Clustering Techniques for
Customer Segmentation
Author: Pooja Pillai, Prandya Kulkarani
Source: https://ieeexplore.ieee.org/document/10307705
Year of Publication: 2023
Main Features:
1. The paper introduces the concept of using heuristic approaches,
particularly the genetic algorithm, to address the limitations of traditional
clustering methods.
2. It outlines a comprehensive methodology for customer segmentation
that integrates both heuristic and non-heuristic clustering techniques.
3. Details are provided on how the genetic algorithm is employed to
extract heuristic information for identifying optimal cluster centers,
addressing issues such as sensitivity to initial centers and local optima.
4. It concludes by summarizing the key findings and suggesting future
research directions, particularly in exploring the use of heuristic clustering
techniques in consumer segmentation and conducting comparative
analyses between different clustering methods.
Areas of Improvement:
1. Providing empirical evidence or case studies demonstrating the
effectiveness of the proposed methodology on real-world datasets would
enhance the paper's credibility.
2. Including quantitative metrics for comparing the performance of different
clustering algorithms would help in assessing their effectiveness objectively.
3. Discussing potential limitations or drawbacks of the proposed
methodology and addressing how these limitations could be mitigated would
provide a more balanced view of the approach.
Paper 7:
Title:Customer Segmentation for Digital Marketing Based on Shopping Patterns
Author:Juhaini Alie, Rendra Gustriansyah
Source:www.ieee.in
Year of publication:2023
Main Features:
Objective: Develop a customer segmentation model for digital marketing using RFM and PAM
methods.
Methodology: Utilizes historical customer purchase data to analyze recency, frequency, and
monetary factors.
Segmentation: Identifies five customer segments: main, potential, general, minimum, and
prospective customers.
Validation: Internal validation results indicate good quality segmentation.
Implications: Enables marketers to optimize services, adjust strategies, and offer personalized
products.
Areas Of Improvement:
1. Providing more details on methodology.
2. Offering specific information on validation.
3. Deepening the discussion of results.
4. Addressing study limitations and proposing future research directions.
5. Ensuring language clarity and coherence.
6. Considering ethical considerations.
7. Incorporating visual aids for clarity.
8. Seeking peer review for feedback.
Paper 8:
Title:Customer Segmentation: Transformation from Data to Marketing Strategy
Author:Luciana Abednego, Cecilia Esti Nugraheni, Adelia Salsabina
Source:www.ieee.in
Year of publication:2023
Main Features:
Customer Segmentation: Explains why dividing customers into groups is
important for marketing.
1. RFM Model: Describes using recency, frequency, and monetary value to group
customers based on their purchases.
2. Dataset Overview: Briefly mentions the real customer data used.
3. RFM Components: Talks about why recency, frequency, and monetary value matter in
understanding customers.
4. Practical Use: Shows how RFM helps sort customers into groups practically.
5. Clustering Algorithms: Introduces K-Means and DBSCAN as tools to help with sorting.
6. Results: Summarizes what happened when these tools were used on the data.
7. Future Outlook: Discusses what's next in understanding and using customer
segmentation.
8. Machine Learning: Touches on using both supervised and unsupervised methods to
make sense of the data.
Areas Of Improvement:
1. Make study goals clearer.
2. Provide more dataset details.
3. Explain methods simply.
4. Clarify how results were checked.
5. Discuss findings in depth.
6. Address study limitations.
7. Check writing for clarity.
8. Consider ethical concerns.
9. Use visuals for better understanding.
Paper 9:
Title:How can algorithms help in segmenting users and customers?
A systematic review and research agenda for algorithmic customer
segmentation
Author:Joni Salminen,Mekhail Mustak, Muhammad Sufyan ,Bernard J. Jans
en
Source:www.ieee.in
Year of publication:2023
Main Features:
1.The research focuses on key questions in customer segmentation.
2.It employs a systematic literature review involving 172 articles.
3. Algorithmic segmentation, particularly K-means clustering, is prevalent.
4.Fourteen different evaluation metrics and various algorithms are
identified.
5. Most studies use segment size as the primary hyperparameter.
6. Typically, studies generate around four segments, rarely exceeding
twenty.
7. The study proposes seven goals and three practical implications.
8. Customer segmentation is vital for targeted marketing and improving
customer experience.
9. Firms struggle with understanding and implementing AI/ML methods for
segmentation.
Areas of improvement:
Provide clearer guidance on choosing algorithms and evaluating results.
• Encourage more diverse use of evaluation metrics.
• Include domain experts in outcome evaluation.
• Diversify hyperparameter selection beyond segment size.
• Simplify the segmentation process for easier implementation.
• Enhance understanding of AI/ML methods for segmentation.
• Foster broader application of segmentation research in practice.
Paper 10:
Title: Customer Segmentation in Food Retail Sector.An Approach
from Customer Behavior and Product Association Rules
Author: Juan Carlos Llivisaca
Source: https://www.researchgate.net
Year of publication:2023
Main Features:
Customer segmentation in the food retail sector, particularly when approached from
customer behavior and product association rules, typically involves the following main
features:
1. Behavioral Segmentation: Analyzing customers' purchasing beha-
vior, including frequency, volume, brand loyalty, preferred product
categories, and shopping channels (e.g., in-store, online).
2. Product Association Rules: Identifying patterns in customer pur-
chases, such as frequently bought together items or sequential pur-
chases, to optimize product placement, cross-selling, and upselling
strategies.
3. RFM Analysis: Segmenting customers based on Recency, Frequency,
and Monetary value of their purchases to identify high-value, loyal,
and at-risk customers for targeted marketing and retention efforts.
4. Market Basket Analysis: Examining the contents of customers' shop-
ping baskets to uncover common item associations and inform
product placement strategies and targeted promotions.
5. Predictive Analytics: Utilizing historical data and advanced analytics
techniques to forecast future purchasing behavior, demand patterns,
and potential churn risks for proactive decision-making and personal-
ized marketing strategies.
6. Segment Profiling: Understanding the characteristics, preferences,
and motivations of different customer segments in detail, including
lifestyle, values, and attitudes, to tailor marketing messages and
product offerings effectively.
Areas of improvement:
1. Concrete Examples: Incorporate specific examples or case studies il-
lustrating how each feature has been applied successfully in food re-
tail settings, providing clarity on its practical implementation and im-
pact.
2. Data Integration Strategies: Expand on the strategies and techno-
logies used to integrate data from various sources (e.g., POS systems,
online platforms, loyalty programs) to facilitate effective customer
segmentation.
3. Measurement Metrics: Emphasize the importance of defining and
measuring key performance indicators (KPIs) to evaluate the effect-
iveness of customer segmentation strategies, enabling continuous im-
provement and optimization.
4. Ethical Considerations: Address ethical considerations related to
customer data privacy, transparency, and consent in segmentation
practices, ensuring alignment with legal regulations and ethical stand-
ards to maintain customer trust.
5. Emerging Trends: Discuss emerging trends and innovations shaping
customer segmentation in the food retail sector, such as the adoption
of AI and machine learning algorithms for predictive analytics, to stay
ahead of industry developments and opportunities for improvement.
Paper 11:
Title: Data mining in customer segmentation
Author: Drazena Gaspar, Mirela Mabić, Ivica Ćorić
Source: https://www.researchgate.net
Year of publication:2016
Main Features:
1. Pattern Recognition: Data mining techniques are used to identify
patterns and relationships within large datasets, allowing businesses
to uncover meaningful insights into customer behavior, preferences,
and characteristics.
2. Segmentation Model Development: Data mining enables the cre-
ation of sophisticated segmentation models by clustering customers
based on similar attributes or behaviors. These models can identify
distinct customer segments with unique needs and preferences.
3. Predictive Analytics: Data mining algorithms can be used to predict
future customer behavior, such as purchasing patterns and product
preferences. By leveraging historical data, businesses can anticipate
customer needs and tailor marketing strategies accordingly.
4. Market Basket Analysis: Data mining techniques like association
rule mining are utilized to analyze customers' purchase histories and
identify frequently co-occurring products. This information is valuable
for cross-selling, product bundling, and optimizing product placement
strategies.
5. Personalization and Targeting: Data mining enables businesses to
create personalized marketing campaigns and offers tailored to spe-
cific customer segments. By understanding customer preferences and
behaviors, businesses can deliver relevant and timely messages, in-
creasing engagement and conversion rates.
Areas of improvement:
1. Tangible Application Scenarios: Include practical examples or case
studies showcasing how data mining techniques have been specific-
ally applied to segment customers within the food retail industry.
Real-world instances help readers grasp the utility and effectiveness
of these techniques.
2. Algorithm Transparency: Provide more detailed insights into the
workings of the data mining algorithms used for customer segmenta-
tion. Explain the underlying principles and mechanics of each tech-
nique, making it easier for readers to understand how they contribute
to segmentation outcomes.
3. Data Quality Assurance: Emphasize the significance of data quality
and integrity in data mining processes. Discuss strategies for ensuring
data cleanliness, completeness, and reliability, as well as addressing
issues like missing values and outliers that can affect segmentation
accuracy.
4. Interpretability and Explainability: Highlight the importance of in-
terpretability and explainability in data mining models. Discuss meth-
ods for making segmentation results interpretable to stakeholders,
such as visualizations, feature importance rankings, or decision rule
summaries.
5. Continuous Model Refinement: Stress the need for ongoing refine-
ment and optimization of data mining models for customer segmenta-
tion. Discuss strategies for iteratively improving segmentation accur-
acy and relevance over time, incorporating feedback from business
stakeholders and adjusting models accordingly.
Paper 12:
Title: Customer segmentation approach in Commercial Banking
Author: Vesela Mihova, Velisar Pavlov
Source: https://www.researchgate.net
Year of publication:2018
Main Features:
1. Demographic Segmentation: Segmenting customers based on
demographic factors such as age, income, occupation, and family
size. This helps banks tailor their products and services to different
customer groups with varying financial needs and preferences.
2. Behavioral Segmentation: Analyzing customers' banking behavior,
including transaction frequency, account balances, online banking us-
age, and product usage patterns. This allows banks to identify high-
value customers, understand their preferences, and personalize their
banking experience.
3. Risk Segmentation: Segmenting customers based on their credit risk
profiles, including credit scores, repayment history, and debt levels.
This helps banks manage risk effectively by offering appropriate credit
products and setting risk-based pricing.
4. Channel Preference Segmentation: Identifying customers' pre-
ferred banking channels, such as in-branch, online banking, mobile
banking, or ATMs. This enables banks to optimize their channel mix,
improve customer service, and enhance the overall banking experi-
ence.
5. Lifecycle Stage Segmentation: Segmenting customers based on
their lifecycle stage, such as new customers, active customers,
dormant customers, or churned customers. This allows banks to tailor
their marketing strategies and retention efforts according to each cus-
tomer's stage in the customer lifecycle.
Areas of improvement:
1. Enhanced Data Integration: Improve integration of data from vari-
ous sources including transactional data, CRM systems, online bank-
ing platforms, and external sources. This would provide a more hol-
istic view of customer behavior and preferences, leading to more ac-
curate segmentation.
2. Advanced Analytics Techniques: Incorporate more advanced ana-
lytics techniques such as machine learning algorithms for predictive
modeling and clustering methods for segmentation. These techniques
can uncover complex patterns in customer data and enhance the ac-
curacy of segmentation.
3. Dynamic Segmentation Models: Develop dynamic segmentation
models that can adapt to changes in customer behavior and market
conditions in real-time. This would enable banks to respond quickly to
evolving customer needs and preferences.
4. Personalization Strategies: Implement personalized marketing and
product recommendations based on segmentation insights. Utilize tar-
geted messaging and customized product offerings to enhance cus-
tomer engagement and satisfaction.
5. Ethical and Transparent Practices: Ensure transparency and eth-
ical use of customer data in segmentation activities. Implement ro-
bust data privacy measures and adhere to regulatory guidelines to
maintain customer trust and confidence.
