Filter Bubble effect in recommender system: case study in e-commerce
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Faculty of Sciences
Abstract
In the rapidly growing e-commerce landscape, recommender systems have emerged as a core
component of creating personalized user interactions, but this can lead to unintended
consequences, such as "filter bubbles," where users only see repeated content with similar qualities
determined by their historical data. Filter bubbles can damage long run user satisfaction by limiting
diversity.
This project presented the development of a real-time hybrid recommender system combining
collaborative filtering (using ALS) with content-based filtering (using TF-IDF and cosine
similarity), primarily to reduce the filter bubble effect while optimizing relevance and diversity of
recommendations.
The methodology consists of two main phases:
• An offline phase, where models are trained and evaluated using a large-scale Amazon
dataset.
• An online phase, where the trained system is integrated into a Django-based e-commerce
platform that dynamically updates recommendations in response to real-time user
interactions.
The results of the experiment show substantial improvements in precision, recall, and
recommendation diversity. It was also shown that the system used in the experiment is adaptable
to user behavior as needed, which improves engagement and user satisfaction with the
recommendations.
This work demonstrates the efficacy of a hybrid and adaptive solution to solving modern
recommendation challenges, specifically as it relates to providing recommendations with a balance
of personalization and content exploration and diversity in dynamic real-time environments