Collaborative filtering algorithm summary article

(Note: there is no longer tired formula algorithm)
1. Similarity Algorithm

1.1Jaccard distance
proportion set to measure the different elements of the two sets of discrimination, but there are obvious problems can not be concerned about the right elements in a set weight value (score)

1.2 cosine similarity
with the vector space to solve the weight value (score) into the issue of similarity calculation, a very common similarity algorithm, to make up for the lack of Jaccard distance calculation

1.3Pearson similarity (also known as the center of cosine similarity)
Pearson cosine similarity is the similarity improvement, it is simple to heavy weight value (score) minus weighted average, it can be understood as the weight value relative to the overall proportion brought into features, the central processing mean some improvement of the accuracy of the algorithm

2. singular value decomposition (SVD) dimension reduction
when an excessive amount of data when the similarity calculation load increases significantly, this time using SVD dimension reduction can effectively reduce the computational burden