RS,7th,July-Tutorials

• Link：https://cambridgespark.com/content/tutorials/implementing-your-own-recommender-systems-in-Python/index.html

1. Collabrative Filtering 

I guess, it is more about recommend things that people similar to you likes

while content-based relies on the content of the stuff, the features of the stuff itself

　　　2.Similarity：

user-similarity and item-similarity are m by m marix and n by n matrix,　(suppose there are m users and n items)

　　　3.About this :"

These two users could have a very similar taste but treat the rating system differently.

"

In the formula of user-based CF:(as follows)

It includes the mean value of the previous rating .This is to solve the rating problem in different rating standard patially,but it cannot solve the problem in similarity... Because in our similarity rating system, users who are actuallly similar are classified as unsimilar.

I think the same problem exists in item-based CF. The personal differences in how they give the points. (I wonder whether here is something we can do in this? Maybe build model about the Gaussian distribution people rate things, and put some resarch in it ? )

　　　　4.Cold-start problem

Memory-based CF cannot solve any cold-start problem, it relies on similarity which requires existing data.

　　　　5.something about the split of data :

Notice that it only diveide the data into two parts randomly, in test data and train data, there are the same number of users and movies! (Misunderstood the spilit in the first place)

　　　　6.Model-based problem 

　　　　To be continued