Links: http://www.zhihu.com/question/20326697/answer/58148605
The proposed algorithm can be broadly divided into three categories: Content-based recommendation algorithm, collaborative filtering recommendation algorithm and knowledge-based recommendation algorithm.
Content-based recommendation algorithm, the principle is that users like and their attention to the item in the content similar to the item, such as you read the Harry Potter I, Content-based recommendation algorithm found Harry Potter ii-vi, and you previously watched the content above (a lot of keywords) have a very strong relevance, the latter recommended to you, This method avoids the cold boot problem of item (Cold boot: If an item has never been followed, other recommended algorithms are seldom recommended, but the content-based recommendation algorithm can analyze the relationship between items, implementation recommendations), the disadvantage is that the recommended item may be duplicated, typical news recommendation , if you read a news about MH370, it is likely to recommend the news and you have viewed the content of the same, the other drawback is that some multimedia recommendations (such as music, movies, pictures, etc.) because it is difficult to mention the content characteristics, it is difficult to recommend, one solution is to manually tag these items.
Collaborative filtering algorithm, the principle is that users like those who have similar interests like the products, such as your friends like the movie Harry Potter I, then will recommend to you, this is the simplest user-based collaborative filtering algorithm (user-based collaborative filtering), There is also an item-based collaborative filtering algorithm (item-based collaborative filtering), both of which read all of the user's data into memory for operation and therefore become memory-based collaborative Filtering, the other is model-based collaborative Filtering, including aspect Model,plsa,lda, clustering, Svd,matrix factorization, etc., This method of training is relatively long, but after the training is completed, the recommendation process is faster.
The last method is based on the recommendation algorithm of knowledge, and some people classify this method as content-based recommendation, which is more typical of building domain ontology, or establishing certain rules and recommending.
Hybrid recommendation algorithm, it will fuse the above methods, weighted or series, parallel, and so on.
Of course, the recommendation system also includes many methods, in fact, machine learning or data mining methods, many can be applied in the recommendation system, such as LR, GBDT, RF (these three methods are often used in some e-commerce recommendations), social network inside the graph structure, can be said to be recommended method. Remark: Recommended book "Recommendation System Practice"
What are the more familiar recommended algorithms for "turn"?
