1.
INTRODUCTION
Modern e-commerce platforms offer users an overwhelming array of products, which can
lead to decision fatigue. Recommendation systems have become essential tools to filter
these choices and suggest items that align with a user’s preferences. Recommendation
systems help users navigate large catalogs and discover relevant products.
In our context, there is a growing interest in promoting domestic products: government
initiatives like Vocal for Local and Make in India encourage consumers to choose Indian-
made goods. The proposed project addresses this by creating a web system where users
searching for foreign goods (in categories like electronics, automobiles, pharmaceuticals,
and medical devices) are recommended equivalent or alternative products made in India.
This not only improves user choice but also supports local industries. India’s e-commerce
market is rapidly expanding, making effective recommendations even more critical.
Customers often face confusion when selecting from many similar items. By analyzing
transaction data or product information, recommendation engines can match user needs
with suitable items. For instance, if a user looks at a foreign smartphone model, the
system might recommend a comparable smartphone from an Indian brand, turning 'Vocal
for Local' into practical shopping guidance through personalized suggestions.
2. LITERATURE REVIEW
Recommender systems have been extensively studied in the literature. Traditional
approaches fall into two main categories: collaborative filtering and content-based
filtering. Collaborative filtering recommends items by leveraging the preferences of
similar users or by relating similar items, without necessarily analyzing item content. It
excels at discovering patterns from user behavior but struggles with new or sparse data
(the cold-start problem) and scalability issues. By contrast, content-based filtering
suggests items whose attributes match those of products a user has liked in the past. It
relies on item metadata and can recommend novel items, but it requires rich item profiles
and may miss serendipitous suggestions. Hybrid systems combine multiple methods to
capture their complementary strengths. As Çano and Morisio note, hybrid recommenders
“combine two or more recommendation strategies in different ways to benefit from their
complementary advantages.”
In e-commerce specifically, association rule mining is a widely used technique to identify
item correlations that can drive recommendations. Association methods (like Apriori or
FP-Growth) analyze historical transactions to find sets of items frequently bought
together. Padhy et al. describe using Apriori and FP-Growth algorithms on large e-
commerce datasets to extract such rules for product suggestions. Such multi-method
approaches can leverage pattern mining while also using ratings or similarity.
Association rule mining has proven effective in practice. Hunyadi et al. emphasize that
frequent patterns and associations from transaction data support key applications such as
real-time product recommendation systems. Their study compares Apriori and FP-
Growth and confirms that FP-Growth consistently outperforms Apriori for large-scale
datasets. Combining methods into hybrid recommenders is considered state-of-the-art. In
practical e-commerce systems, this often means using both user-behavior data and item
attributes. This supports the use of Apriori and FP-Growth in a hybrid system for our use
case of promoting Indian alternatives.
3. RESEARCH GAPS AND OBJECTIVES
Our project initially explored several recommendation strategies. We implemented the
Apriori and FP-Growth association mining algorithms on sample e-commerce transaction
data to identify related product sets. Consistent with the literature, we observed that FP-
Growth handled larger data volumes more efficiently, while Apriori was simpler but
slower. Collaborative filtering alone could not connect foreign items to local equivalents,
so we decided to use a hybrid recommendation model. This model combines association
rule mining with content-based filtering to capture multiple types of signals.
A major challenge was data acquisition. No public dataset exists that directly pairs
foreign products with their Indian counterparts. To address this, we conducted a
comprehensive data collection using web scraping and API calls. We collected product
metadata such as brand, model, category, features, and price from both international and
Indian listings. Using AI tools, we normalized product names and attributes to ensure
consistency and usability. This approach helped us build a dataset tailored for
recommendation logic.
Our objectives are:
• Develop a hybrid recommendation engine suggesting Indian alternatives for foreign
products.
• Prepare a structured dataset from online sources using AI-assisted enrichment.
• Use Apriori and FP-Growth to discover item associations and use them in rule-based
filtering.
• Evaluate recommendation performance using accuracy metrics such as precision and
recall.
REFERENCES
1. N. Padhy, S. Suman, T. S. Priyadarshini, and S. Mallick, “A Recommendation System
for E-Commerce Products Using Collaborative Filtering Approaches,” Eng. Proc., vol.
67, no. 1, p. 50, 2024.
2. O. Doğan, “A Recommendation System in E-Commerce with Profit-Support Fuzzy
Association Rule Mining (P-FARM),” J. Theor. Appl. Electron. Commer. Res., vol. 18,
no. 2, pp. 831–847, 2023.
3. Çano, E., & Morisio, M. “Hybrid Recommender Systems: A Systematic Literature
Review,” Intell. Data Anal., 21(6), 1487–1524, 2017.
4. Hunyadi, I. D., Constantinescu, N., & Țicleanu, O.-A., “Efficient Discovery of
Association Rules in E-Commerce,” Appl. Sci., vol. 15, art. no. 5498, 2025.
