Why Are Zhihu and Douyin So Addictive? A First Look at How Recommendation Systems Work

Introduction to Recommendation Systems

A recommendation system (Recommender system) is an important tool that helps users discover content and overcome information overload. By analyzing user behavior and modeling user interests, it predicts users' interests and makes recommendations.

Clay Shirky, an American writer who studies the internet's impact on society and the economy, once said: "The problem facing humanity is not information overload, but filter failure." Taobao, Amazon, NetEase Cloud Music, Instagram, and major video sites all use recommendation systems to recommend various products users might be interested in—books, movies, music, electronics, and more. For users, recommendation systems are like "smart" filters that help them quickly find what they desire. For businesses, recommendation systems are content-driven and generate revenue.

Classification of Recommendation Systems

Based on whether they recommend different data to different users

1. Based on mass behavior (recommended by the site administrator themselves, or currently popular items calculated from feedback statistics of all users in the system)
2. Personalized recommendation engines (help you find like-minded friends with similar tastes, then make recommendations on that basis)

Based on their data sources

1. Demographic-based (users of the same age or gender are judged to be similar users)
2. Content-based (items have the same keywords and tags; human factors are not considered)
3. Collaborative filtering-based recommendations (discover correlations among items, content, or users for recommendations; divided into three subclasses, discussed below)

With the development of Web 2.0, websites have increasingly encouraged user participation and user contributions, and thus collaborative filtering-based recommendation mechanisms emerged. The principle is simple: based on users' preferences for items or information, discover correlations among the items or content themselves, or discover correlations among users, and then make recommendations based on these associations.

Based on how they are built

1. Based on items and users themselves (a user-item two-dimensional matrix describes user preferences; clustering algorithms)
2. Association rule-based (The Apriori algorithm is one of the most influential algorithms for mining frequent itemsets of Boolean association rules)
3. Model-based recommendations (machine learning; so-called machine learning means letting computers continuously learn like the human brain, and is a subfield of artificial intelligence)

Classified by recommendation method

• Content-based recommendation
• Collaborative filtering recommendation
• Association rule-based recommendation
• Utility-based recommendation
• Knowledge-based recommendation
• Hybrid recommendation

Recommendation System Workflow

Welcome to Discuss

I've compiled a series of technical articles and materials. Reply with linux, flutter, c#, netcore, android, java, python, and so on in the backend of the WeChat official account "Programming Lab" to get related technical articles and materials. If you have any questions, you can also leave a message in the official account backend~