Recommendation algorithm-user recommendation (usercf) and item recommendation (itemcf) Comparison

I. Definition

1. Usercf
2. Itemcf: Recommended items that are similar to those he liked before.

According to user recommendations, the focus is to reflect the hot spots of small groups similar to user interests. Based on the item recommendations, the focus is on the historical interests of users, that is:

• Usercf is the popularity of an item in a certain group
• Itemcf reflects my interests and is more personalized.

Ii. Reasons Why news websites use usercf:

1. Most users like hot news, and the fine-grained personalization is negligible.
2. Personalized news recommendations emphasize hot spots. Popularity and effectiveness are the focus of recommendations, while the importance of personalization can be reduced.
3. Itemcf needs to maintain a table of item relevance. When the item quantity is updated too quickly, it is technically difficult to maintain this table. News websites can directly recommend hot news to new users.
4. For websites such as e-commerce, music, and books, itemcf has the following advantages:
   1. Users' interests are fixed and persistent;
   2. You don't need to think too much about popularity. You just need to help the user discover related items in his research field.
5. Technical Considerations
   1. Usercf needs to maintain a user similarity Matrix
   2. Itemcf needs to maintain an item similarity Matrix

Iii. Comparison of advantages and disadvantages

Project	Usercf	Itemcf
Performance	Suitable for scenarios with few users. If there are too many users, the cost of calculating the user similarity matrix is high.	This method is applicable when the number of items is significantly smaller than the number of users. If there are many items, the cost of calculating the item similarity matrix is high.
Fields	High requirement on effectiveness and low requirement on user personalized interests	A wide range of long-tail items and strong user personalized needs
Real-time	The recommendation results do not need to change immediately because the user has new behaviors.	New user behaviors will inevitably lead to real-time changes in recommendation results
Cold start	After a new user behaviors a small number of items, personalized recommendations cannot be made to the user immediately, because the user similarity is calculated offline. After a new item is launched, a user can recommend the item to other users once the item has behaviors.	A new user can recommend related items to an item as long as he has behaviors on the item, but cannot recommend the item to the user without updating the item similarity table offline.
Reason for recommendation	Hard to provide	The reason for recommendation can be summarized based on historical user behaviors.

Recommendation algorithm-user recommendation (usercf) and item recommendation (itemcf) Comparison