Utilize user tag data

Except as mentioned before UserCF and ItemCF The two methods , The third important way is through some features (feature ) Contact users and items , Recommend items with features that users like . The features here have different ways of expression , For example, it can be expressed as a collection of attributes of items ( For example, for books , The attribute set includes the author 、 Press. 、 Topics and keywords, etc ), It can also be expressed as a vector of argot meaning (latent factor vector), This can be learned from the previous implicit semantic model . This chapter will discuss an important feature expression " label ".

What is a label

A tag is a non hierarchical structure 、 Keywords used to describe information , It can be used to describe the semantics of objects . Depending on who labels the item , Label applications are generally divided into two types ： One is to ask the author or expert to label the item ; The other is to let ordinary users label items , It's just MUGC (User Generated Content, User generated content ) Label application .UGC Tag system is an important way to express user interest and item semantics . When a user puts a label on an item , On the one hand, this tag describes the user's interests , On the other hand, it represents the semantic meaning of objects , Thus linking users and items .、

Recommendations in the labeling system

Tagging is an important user behavior , Contains a lot of user interest information , Therefore, deep research and use of user marking
Signing can guide us to improve the recommendation quality of personalized recommendation system . meanwhile , The label representation is very simple , Convenient for many algorithms .
There are two main recommendation problems in the labeling system .
How to use users' labeling behavior to recommend items for them （ Tag based recommendation ）？
How to recommend a label suitable for the item when the user labels the item （ Tag recommendation ）？
In order to study the above two problems , We need to answer the following questions first 3 A question .
Why do users label ？
How users label ？
What kind of label does the user type ？

Label based recommendation system

Get the user tag behavior data , I believe everyone can think of the simplest personalized recommendation algorithm . This algorithm
The description is as follows .

1. Count the most commonly used tags of each user .
2. For each tag , Count the items that have been labeled the most times .
3. For a user , First, find his commonly used tags , Then find the most popular items with these labels and recommend them to this
   Users .