Product Recommendations for a Food Delivery Marketplace

Problem

An online delivery marketplace sought to increase revenue by enhancing user experience through data-driven solutions. With over 205,000 food orders generating $6 million in revenue in 2023, the company identified untapped potential in its data. The existing user interface lacked personalized recommendations, prompting a proof-of-concept project to explore the impact of tailored food suggestions.

Solution

A recommendation pipeline combining content-based and collaborative filtering methods was developed to personalize food suggestions. Key steps included:

1. Data Preparation: Extracted user and product data from MongoDB and structured it into pandas DataFrames for analysis.
2. Exploratory Data Analysis (EDA): Uncovered patterns in user behavior and product attributes to refine recommendation strategies.
3. Content-Based Filtering: Created user and product profiles from features in product descriptions and preferences, using sklearn’s DictVectorizer to calculate similarity scores.
4. Collaborative Filtering: Leveraged purchase histories to identify trends among similar users and predict preferences.
5. Hybrid Approach: Combined both methods to deliver accurate recommendations while encouraging product discovery.

Tools Used

• MongoDB: Used for storing and retrieving user and product data.
• Python: Built the recommendation algorithms and performed data analysis.
• Pandas: Processed and analyzed data for actionable insights.
• Matplotlib: Visualized data trends and results.
• Sklearn: Encoded categorical data into numerical vectors for machine learning models.

Outcome

The proof-of-concept showcased the potential of personalized recommendations to enhance user experience and engagement. The hybrid pipeline delivered tailored food suggestions, improving both relevance and discovery. The project highlighted the importance of clean, structured data and revealed areas for improving the marketplace’s data pipelines.

The recommendation system is now being tested in a sandbox environment to simulate real-world usage, preparing for full-scale implementation. This initiative lays the groundwork for extending recommendation services to other marketplace categories, driving further growth and user satisfaction.