“协同过滤”是推荐系统中的常用技术,按照分析维度的不同可实现“基于用户”和“基于产品”的推荐。
以下是利用python实现电影推荐的具体方法,其中数据集源于《集体编程智慧》一书,后续的编程实现则完全是自己实现的(原书中的实现比较支离、难懂)。
这里我采用的是“基于产品”的推荐方法,因为一般情况下,产品的种类往往较少,而用户的数量往往非常多,“基于产品”的推荐程序可以很好的减小计算量。
其实基本的思想很简单:
首先读入数据,形成用户-电影矩阵,如图所示:矩阵中的数据为用户(横坐标)对特定电影(纵坐标)的评分。
其次根据用户-电影矩阵计算不同电影之间的相关系数(一般用person相关系数),形成电影-电影相关度矩阵。
其次根据电影-电影相关度矩阵,以及用户已有的评分,通过加权平均计算用户未评分电影的预估评分。例如用户对A电影评3分、B电影评4分、C电影未评分,而C电影与A电影、B电影的相关度分别为0.3和0.8,则C电影的预估评分为(0.3*3+0.8*4)/(0.3+0.8)。
最后对于每一位用户,提取其未评分的电影并按预估评分值倒序排列,提取前n位的电影作为推荐电影。
以下为程序源代码,大块的注释还是比较详细的,便于理解各个模块的作用。此外,程序用到了pandas和numpy库,实现起来会比较简洁,因为许多功能如计算相关系数、排序等功能在这些库中已有实现,直接拿来用即可。
import pandas as pd
import numpy as np
#read the data
data={'Lisa Rose': {'Lady in the Water': 2.5, 'Snakes on a Plane': 3.5,
'Just My Luck': 3.0, 'Superman Returns': 3.5, 'You, Me and Dupree': 2.5},
'Gene Seymour': {'Lady in the Water': 3.0, 'Snakes on a Plane': 3.5,
'Just My Luck': 1.5, 'The Night Listener': 3.0},
'Michael Phillips': {'Lady in the Water': 2.5, 'Snakes on a Plane': 3.0,
'Superman Returns': 3.5, 'The Night Listener': 4.0},
'Claudia Puig': {'Snakes on a Plane': 3.5, 'Just My Luck': 3.0,
'The Night Listener': 4.5, 'You, Me and Dupree': 2.5},
'Mick LaSalle': {'Just My Luck': 2.0, 'Lady in the Water': 3.0,'Superman Returns': 3.0, 'The Night Listener': 3.0, 'You, Me and Dupree': 2.0},
'Jack Matthews': {'Snakes on a Plane': 4.0, 'The Night Listener': 3.0, 'Superman Returns': 5.0, 'You, Me and Dupree': 3.5},
'Toby': {'Snakes on a Plane':4.5,'You, Me and Dupree':1.0,'Superman Returns':4.0}}
#clean&transform the data
data = pd.DataFrame(data)
#0 represents not been rated
data = data.fillna(0)
#each column represents a movie
mdata = data.T
#calculate the simularity of different movies, normalize the data into [0,1]
np.set_printoptions(3)
mcors = np.corrcoef(mdata, rowvar=0)
mcors = 0.5+mcors*0.5
mcors = pd.DataFrame(mcors, columns=mdata.columns, index=mdata.columns)
#calculate the score of every item of every user
#matrix:the user-movie matrix
#mcors:the movie-movie correlation matrix
#item:the movie id
#user:the user id
#score:score of movie for the specific user
def cal_score(matrix,mcors,item,user):
totscore = 0
totsims = 0
score = 0
if pd.isnull(matrix[item][user]) or matrix[item][user]==0:
for mitem in matrix.columns:
if matrix[mitem][user]==0:
continue
else:
totscore += matrix[mitem][user]*mcors[item][mitem]
totsims += mcors[item][mitem]
score = totscore/totsims
else:
score = matrix[item][user]
return score
#calculate the socre matrix
#matrix:the user-movie matrix
#mcors:the movie-movie correlation matrix
#score_matrix:score matrix of movie for different users
def cal_matscore(matrix,mcors):
score_matrix = np.zeros(matrix.shape)
score_matrix = pd.DataFrame(score_matrix, columns=matrix.columns, index=matrix.index)
for mitem in score_matrix.columns:
for muser in score_matrix.index:
score_matrix[mitem][muser] = cal_score(matrix,mcors,mitem,muser)
return score_matrix
#give recommendations: depending on the score matrix
#matrix:the user-movie matrix
#score_matrix:score matrix of movie for different users
#user:the user id
#n:the number of recommendations
def recommend(matrix,score_matrix,user,n):
user_ratings = matrix.ix[user]
not_rated_item = user_ratings[user_ratings==0]
recom_items = {}
#recom_items={'a':1,'b':7,'c':3}
for item in not_rated_item.index:
recom_items[item] = score_matrix[item][user]
recom_items = pd.Series(recom_items)
recom_items = recom_items.sort_values(ascending=False)
return recom_items[:n]
#main
score_matrix = cal_matscore(mdata,mcors)
for i in range(10):
user = input(str(i)+' please input the name of user:')
print recommend(mdata,score_matrix,user,2)