Building Better Recommenders: A Data Scientist’s Journey

Understanding Recommendation Systems

3 min readSep 28, 2025

In the early phase of my data science fellowship at Buildables, I began by understanding the fundamentals of recommendation systems. I discovered how these algorithms power personalized suggestions by analysing user behaviour and item attributes. I took careful notes and even wrote a detailed summary to cement my understanding, which was challenging because I hadn’t done structured writing in a while. Reflecting on these core ideas helped me grasp why recommendation systems matter and how they deliver personalized content by using techniques like collaborative filtering and content analysis.

Collaborative Filtering Techniques

Next, I dove into collaborative filtering, a popular technique in recommendation systems. I learned about two main approaches: user-user and item-item. In user-based filtering, the system finds people with tastes similar to yours and uses their preferences to suggest new items. In item-based filtering, it finds items similar to those you already like and recommends those. We also discussed combining both perspectives into a hybrid collaborative approach to refine suggestions by leveraging multiple similarity angles. Working through examples helped me see how patterns in user behaviour can translate into personalized recommendations.

Collaborative Filtering Visual Explanation

Content-Based Recommendation Systems

Next, I turned my attention to content-based recommendation systems, which rely on analysing item features. These systems recommend items by looking at the attributes or content of each item and matching them to a user’s past preferences. For example, if I liked articles about machine learning, a content-based system might suggest other articles tagged similarly or with related keywords. In a group assignment, we worked together to apply these concepts and model how item features drive recommendations. Collaborating with peers on this project helped me see how analysing content in detail can match users with relevant items more effectively.

Rule-Based Recommendation Systems

I also explored rule-based recommendation systems, which rely on predefined logic rather than statistical patterns. These systems use explicit conditions — for example, “if a user is interested in product X, then recommend product Y.” I considered their scope by learning when rule-based approaches might be useful, such as in domains with clear logic defined by experts. Evaluating a colleague’s answer on this topic and providing feedback helped me solidify my understanding by explaining the concepts to someone else. Considering rule-based methods alongside statistical approaches gave me a well-rounded view of possible solutions.

Rule-based Recommendation Engines Visual Example.

Hybrid Recommender Systems

Finally, I learned about hybrid recommender systems that combine multiple approaches to improve recommendations. Hybrid systems might merge collaborative filtering with content analysis, or blend two collaborative methods, aiming to leverage the strengths of each technique. The idea is to create a more personalized and efficient system by compensating for one approach’s weaknesses with another. This concept excited me because it shows how real-world recommendation engines often mix different techniques to better capture user preferences and context.

Hybrid Recommendation Systems Visual Explanation.

Reflections and Next Steps

Looking back on what I’ve covered so far, I feel increasingly confident in my understanding of recommendation techniques. Each concept built on the previous one, from basic filters to hybrid models, giving me a clear sense of progression. I still have much to learn, but I appreciate how each concept connects to the others. I’m excited to apply these ideas in upcoming projects and continue refining my skills as a data scientist.