engineering architecture direction are more or less timid when seeing algorithm-related technologies such as recommendation/search/advertising. But after careful study, I found that it is not so difficult.
Today’s 1-minute series, I will introduce you to the "collaborative filtering" in the recommendation system. There is absolutely no formula, so everyone can understand.
What is Collaborative Filtering?
Answer
: By finding groups with similar interests or common experiences, we recommend information of interest to users. For example,
, how to use collaborative filtering to recommend movies to user A?
Answer
: The brief steps are as follows:
find the hobbies of user A (user_id_1)
find the set of user groups that have the same movie hobbies as user A (user_id_1) Set
find the set of movies that the group likes Set
recommend these movies Set
?
Answer
: The brief steps are as follows:
(1) Draw a big table,
abscissa
is all movie_id,
ordinate
all user_id,
cross
represents this user likes this movie
z47zz as above horizontal table: Coordinates, assuming there are 10w movies, so the abscissa has 10w movie_ids, the data comes from thedatabase
the ordinate, assuming there are 100w users, so the ordinate has 100w user_ids, and the data also comes from the intersection of
database
, "1 "Means that the user likes the movie. The data comes from the
log
voice-over: What is "like" needs to be artificially defined, such as browsed, searched, and liked. Anyway, there are these data in the log
(2) find user A( User_id_1)’s hobbies
are as shown in the table above, you can see that user A likes movies {m1, m2, m3}
(3) find the user group set Set
that has the same movie hobbies as user A (user_id_1) as shown in the table above, you can see So, users who like {m1, m2, m3}, in addition to u1, there are {u2, u3}
(4) find the group’s favorite movie collection Set
as shown in the table above, in the user group with the same preferences {u2, u3} , And the favorite movie collection is {m4, m5}
voice-over: "Synergy" is reflected here.
(5) When user A (use_id_1) visits the website in the future, he must recommend the movie {m4, m5} to him. The general principle of
collaborative filtering is as above, and I hope everyone can gain something.