With the rise of personalized beauty, recommender systems have become essential for skincare retailers, brands, and apps. But building one isn’t just about fancy algorithms — it starts with clean, structured data.
In this guide, we’ll walk through how to build a skincare recommender system using real product data from BeautyFeeds.io.
Step 1: Get the Right Dataset
Before you write a single line of code, you need solid data. BeautyFeeds provides structured .xlsx datasets with fields like:
- Product name & brand
- Retailer source (e.g., Dermstore, Mecca)
- Category (serum, moisturizer, cleanser, etc.)
- Ingredients list (INCI format)
- Price, SKU, image URL
- Optional: customer reviews, ratings
Step 2: Clean & Preprocess
Typical cleaning steps include:
- Normalize ingredients: Convert INCI names to lowercase, split by commas, remove duplicates.
- Standardize categories: Group similar products (e.g., “night cream” = “moisturizer”).
- Filter active products: Drop discontinued or placeholder entries.
Optional NLP: Use TF-IDF or Word2Vec on ingredients to represent products in vector space.
Step 3: Choose a Recommendation Strategy
There are 3 main types of recommenders. We’ll focus on content-based filtering, which is ideal when user data is limited.
Content-Based Example: Ingredient Similarity
- Vectorize ingredient lists (e.g., TF-IDF, BERT embeddings)
- Calculate cosine similarity between products
- Recommend products with the most similar ingredient profiles
Bonus: Use product categories and skin type tags (if available) to refine relevance.
Step 4: (Optional) Add Personalization
If you have user data (e.g., skin type, past purchases):
- Create user profiles based on preferred ingredients or product types
- Blend content-based + collaborative filtering (hybrid model)
You can also onboard users with a quiz to help cold-start recommendations.
Step 5: Evaluate and Iterate
Use metrics like:
- Precision @ K: How many recommended items were relevant?
- Diversity: Are you recommending different brands/categories?
- User feedback: Are recommendations improving engagement or conversions?
Tip: A/B test changes when possible to measure impact.
What You Can Build
With real product and ingredient data, you can create:
- Interactive skincare quizzes or product finders
- Ingredient-based filters (e.g., fragrance-free, paraben-free)
- AI-powered product advisors for ecommerce or mobile apps
Ready to Build?
Get high-quality, structured skincare product data from BeautyFeeds.io. Whether you’re prototyping a personal project or launching a production-grade AI tool, the data is clean, rich, and ready to power your beauty-tech ideas.
Leave a Reply