Integrating User Behavior with Product Data: Building a Foundational Personalized Recommendation System

1. Recommendation Systems Are More Than "You Might Like"

Traditional recommendation systems are often perceived as black-box algorithms that simply output a list of “you might like” items. However, for early-stage projects or lightweight shopping guide applications, a more practical approach is:

• Accurately identifying the user's behavioral intent, such as search terms, click paths, or favorite actions;

• Performing basic matching based on product attribute tags (e.g., brand, price, category);

• Leveraging existing structured data to respond quickly with relevant results.

By starting with “near real-time behavior sensing + static product tagging + structured API data,” you can progressively build your own lightweight personalized recommendation engine without needing complex models initially.

2. Core Concept: The Three Pillars of Recommendation Systems

An effective recommendation system is generally driven by three types of data:

For example, in the scenario where a user searches for “Lenovo laptop,” the flow can be:

User searches “Lenovo laptop” → System records behavior → Recommendation page shows similar products from Dell, Huawei → Encourage user to favorite or add to cart

3. Practical Guide: Collecting User Behavior Data

Collecting and storing user behavior data is the foundation of a recommendation system. With frontend event tracking, you can instantly capture user interactions and send them to the backend for storage and analysis.

Example: Track User Clicks

// Example: User clicks a product
function trackClick(skuId, platform) {
fetch('/api/track/click', {
method: 'POST',
body: JSON.stringify({ skuId, platform }),
headers: { 'Content-Type': 'application/json' }
});
}

Example of a record stored in MongoDB or Redis:

{
"user_id": "u12345",
"type": "click",
"sku_id": "10001234",
"platform": "jd",
"timestamp": "2025-05-15T15:00:00"
}

This behavioral data forms the base for future personalized recommendations.

4. Building Recommendation Logic (No Machine Learning Required)

At the initial stage, there's no need for machine learning models. You can implement a configurable rule-based system to achieve useful recommendations.

1. Similar Product Recommendation

When a user clicks on a product, you can recommend others that share the following:

• Same brand

• Similar price range

• Popular on the same platform (LuckData provides sales volume field)

def recommend_by_behavior(sku_id):
item = get_item_by_sku(sku_id)  # Lookup clicked item
# Find same brand + similar price products
return luckdata_search({
"brand": item.brand,
"min_price": item.price - 200,
"max_price": item.price + 200
})

LuckData’s API supports flexible field filtering, enabling direct retrieval of relevant product recommendations.

2. Historical Keyword-Based Recommendation

If a user’s historical search keywords are:

["Huawei phone", "Mate60", "Honor90"]

You can build a “keyword → related brand/category” mapping to drive recommendations:

def recommend_by_keywords(user_keywords):
hot_keywords = get_related_keywords(user_keywords)
result = []
for kw in hot_keywords:
items = luckdata_search({"keyword": kw})
result.extend(items[:3])
return result

This method is especially useful for new users without click data (cold start).

5. Tag-Based Recommendation System

To improve relevance and control, you can implement a tagging system for each product, connecting user preferences with product attributes:

Tags can be extracted from the API’s product fields:

• Platform category (category)

• Keywords in product titles (e.g., “lightweight”, “gaming”)

• Price segment classification (low, mid, high)

• Review and sales metrics (some platforms supported by LuckData include ratings)

If a user shows a preference for "lightweight laptops," products with matching tags can be prioritized in the recommendation.

6. Enhancing Data Quality with LuckData

Data quality and consistency are critical for the success of any recommendation engine. LuckData offers several advantages as a product data provider:

Advantages of LuckData:

• Supports unified APIs across major platforms (JD, Taobao, Pinduoduo, Xiaohongshu), enabling easier product normalization;

• Offers comprehensive product fields: price, brand, platform, title, sales, rating, etc.;

• Search API supports conditional filters (keyword, brand, price range), making it highly flexible for custom queries;

• Well-suited as a backend service for recommendation systems, minimizing the need to build and maintain data crawlers.

With LuckData, you focus on building logic and user experience, not on data acquisition.

7. Recommendation Output Example (Demo)

User Behavior:

• User A searches for “Lenovo lightweight laptop,” clicks on SKU: JD-123456

Recommended Output:

[
{
"title": "Lenovo Xiaoxin Air14 Plus",
"platform": "taobao",
"price": 4899,
"reason": "Same brand + similar price"
},
{
"title": "Huawei MateBook D14",
"platform": "jd",
"price": 4999,
"reason": "User preference: lightweight laptop"
}
]

This approach ensures recommendations are both relevant and action-oriented.

8. Summary and Practical Tips

You don’t need complex models to start building a valuable recommendation system. A data-driven, rule-based approach is often the best place to begin.

✅ Start with collecting and storing user behavior logs
✅ Build a product tagging system and rule-based recommendation logic
✅ Leverage structured product data from LuckData to reduce development effort
✅ Integrate recommendations with search results to enhance conversion rates

A recommendation system acts as a long-term amplifier of data value. The earlier you start designing it, the sooner you can accumulate meaningful behavioral insights and user preference data.

Articles related to APIs :

• Building a Unified API Data Layer: A Standardized Solution for Cross-Platform Integration

• Enterprise-Level API Integration Architecture Practice: Unified Management of JD.com and Third-Party APIs Using Kong/Tyk Gateway

• Practical Guide to API Debugging and Automated Testing Platforms: Postman, Hoppscotch, Swagger Full Workflow

• JD API Third-Party SDKs and Community Libraries: Selection Strategies and Best Practices

• In-depth Guide to JD.com APIs: After-Sales Services and Consumer Rights Protection

For seamless and efficient access to the Jingdong API, please contact our team : support@luckdata.com