Abstract:
The paper presents the development of a personalized recommendation system for
Beplob Mart, an e-commerce website. The system leverages Matrix Factorization
(MF), a collaborative filtering technique, to recommend products to users based on their
past behavior and product characteristics. The system architecture utilizes a
combination of technologies: HTML, CSS, and Javascript for a user-friendly interface,
PHP for server-side logic, and Python libraries (Pandas, NumPy, scikit-learn) for data
manipulation, numerical computations, and implementing the MF model. The
recommendation process involves collecting user data (purchase history, browsing
behavior, ratings) and product data (descriptions, attributes, category). Both datasets
undergo preprocessing to ensure quality. User-item interaction matrices are then
created, capturing interactions between users and products. MF plays a key role in
uncovering hidden patterns within these matrices. The chosen technique decomposes
the user-item interaction matrix into user and item factor matrices with a predefined
number of latent factors. These factors represent underlying user preferences and
product characteristics. By multiplying the user and transposed item factor matrices,
the system predicts potential user interactions with various products. Recommendations
are generated for each user based on the highest predicted scores, suggesting items
they're most likely to be interested in. The website integrates the recommendation logic,
displaying personalized suggestions alongside product browsing and search results.
This approach aims to enhance the user experience by offering relevant product
recommendations, potentially leading to increased engagement and sales for Beplob
Mart. The paper highlights the importance of model evaluation and real-time updates
for optimal performance. Additionally, it emphasizes the potential benefits of
incorporating explainability techniques to build trust and transparency with users.
