Chapter 3: Recommendation System cold start and CB

3.1 Introduction to cold start:
The Cold Start Problem (cold start) is mainly divided into three types:
? Cold start
? Item cold start
? Cold start
Reference solution:
? Popular
? Use user information. (For example, gender, age, and Region)
? Use the social network information of the Logon account
? Require new users to provide some feedback when logging on
3.2 use User Registration Information
The following describes a simple recommendation algorithm based on user information. The core issue is to calculate the items that users like for each feature. That is to say, for each feature F, calculate the user's liking for each item with this feature P (F, I)
P (F, I) can be simply defined as the popularity of item I among users with F features:

N (I) is a set of users who like item I. U (f) indicates a set of users with feature F.
It can be seen that a relatively high N (I) is likely to have a relatively high P (F, I), so the launch result is likely to be a hot result. Therefore, we can define P (F, I) as the ratio of feature F among users who like item I:

Experiment on the dataset last. fm
Code to be written:
3.4 use item content information:
For user-CF, item cold start is not very sensitive. Because many websites provide users with more than recommendation-based content.
For item-CF, item similarity tables must be updated frequently when items are cold-started, with high time complexity.
If there is no cold start problem in the item content-based model, cold start can be properly solved. Generally, the item content can be represented by the vector space model. This model represents an item as a keyword vector. For text, the word segmentation technology in natural prediction processing may be used.

The vector space model may achieve better results in long text. (The short text cannot work. You can consider word2vec for further study)
Topic Model)
Representative Lda. LDA has three elements: Documents, topics, and words. Each document is a collection of words, called bag of words. Each word is a topic in one article.
To learn ....

Chapter 3: Recommendation System cold start and CB