table 1 & table 2 : user rating movie history data <user : movie_id0 = score, movie_id1=score>

table 3 & table 4 : normalized co-occurrence matrix, normalize is becuase each movie has differnet rating feedback, need to balanced weight of contribution

table 5 : rating matrix per user

MapReduce task :

• Data :

user0 : movie0=rating

user0 : movie1=rating

user10 : movie 5=rating

• data pre-process :

<user_id, movie0=rating, movie1=rating ......>

• build co-occurrence matrix :

input : <user_id, movie0=rating, movie1=rating ......>

output : <movieA : movieB, count>

• Normalized co-occurrence matrix :

input : <movieA : movieB, count>

output : <movieB, movieA = count/total count>

• Rating matrix :

input : user0 : movie0=rating data

output : <user_id, movie = rating>

• Multiplication by cell :

input : rating matrix & co-occurrence matrix

output : rating matrix cell * co-occurrence matrix cell

• Sum up cell (which belong to the same row(key)) ==> prediction value
• use this prediction value to determine recommend or not