scikit-learn - 直接使用Python构建电影推荐系
2017-11-03 本文已影响0人
forestgang
1. 参考资料
http://blog.csdn.net/u013185349/article/details/61192218
2. 算法
2.1 所有电影之间的相似度
计算相似度,就要涉及相似距离度量,这里列举两种:欧氏距离sim_distance ,皮尔逊sim_pearson 。距离越大,越相似
2.2 推荐用户没看过的电影
某一部未看过电影分数= sum(该部未看过的电影与每一部已看电影之间相似度*已看电影的评分) /sum(未看电影与每一部已看电影之间相似度)
例如:未看电影A,已看电影B,C:则,
电影A分数 = [sim(A,B)*rating(B) +sim(A,C)*rating(C)] / [ sim(A,B) + sim(A,C)]
3. 代码
"""
==============
构建电影推荐系统
==============
http://blog.csdn.net/u013185349/article/details/61192218
"""
print(__doc__)
# 电影打分数据
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,
'The Night Listener': 3.0},
'Gene Seymour': {'Lady in the Water': 3.0, 'Snakes on a Plane': 3.5,
'Just My Luck': 1.5, 'Superman Returns': 5.0, 'The Night Listener': 3.0,
'You, Me and Dupree': 3.5},
'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, 'Superman Returns': 4.0,
'You, Me and Dupree': 2.5},
'Mick LaSalle': {'Lady in the Water': 3.0, 'Snakes on a Plane': 4.0,
'Just My Luck': 2.0, 'Superman Returns': 3.0, 'The Night Listener': 3.0,
'You, Me and Dupree': 2.0},
'Jack Matthews': {'Lady in the Water': 3.0, '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}
}
#
# 因为计算以电影物品为主,所以先将上面data[user][movie]数据转换成newdata[movie][user]格式,
# 也即二维矩阵进行行列对换
#
def transformdata(data):
'''
物品之间的相似度 与 用户之间的相似度 求解 一样。故只需要将用户换成物品即可
'''
newdata = {}
for person in data:
for movie in data[person]:
# 初始化
newdata.setdefault(movie, {})
# 物品与用户对调
newdata[movie][person] = data[person][movie] # 字典可以直接写[key],就表示插入key值了。非常简便
return newdata
print("1. 电影打分:")
print(transformdata(data))
print("")
from math import sqrt
def sim_distance(data, person1, person2):
'''欧氏距离求相似度,距离越大,越相似'''
commonmovies = [movie for movie in data[person1] if movie in data[person2]]
if len(commonmovies) == 0: return 0
# 平方和
sumSq = sum([pow(data[person1][movie] - data[person2][movie], 2) for movie in commonmovies])
# 使最终结果是,越相似,距离越大。所以将上面距离取倒数即可
sim = 1 / (1 + sqrt(sumSq))
return sim
def sim_pearson(data, person1, person2):
'''
计算上面格式的数据 里的 两个用户 相似度.
基于用户过滤思路:找出两个用户看过的相同电影的评分,从而进行按pearson公式求值。那些非公共电影不列入求相似度值范围。
基于物品过滤思路:找过两部电影相同的观影人给出的评分,从而按pearson公式求值
返回:评分的相似度,[-1,1]范围,0最不相关,1,-1为正负相关,等于1时,表示两个用户完全一致评分
这里的data格式很重要,这里计算相似度是严格按照上面data格式所算。
此字典套字典格式,跟博客计算单词个数 存储格式一样
'''
# 计算pearson系数,先要收集两个用户公共电影名单
# commonmovies = [ movie for movie in data[person1] if movie in data[person2]] 分解步骤为如下:
commonmovies = [] # 改成列表呢
for movie in data[person1]: # data[person1]是字典,默认第一个元素 in (字典)是指 key.所以这句话是指 对data[person1]字典里遍历每一个key=movie
if movie in data[person2]: # data[person2]也是字典,表示该字典有key是movie.
commonmovies.append(movie) # commonmovie是 两个用户的公共电影名的列表
# 看过的公共电影个数
n = float(len(commonmovies))
if n == 0:
return 0
'''下面正是计算pearson系数公式 '''
# 分布对两个用户的公共电影movie分数总和
sum1 = sum([data[person1][movie] for movie in commonmovies])
sum2 = sum([data[person2][movie] for movie in commonmovies])
# 计算乘积之和
sum12 = sum([data[person1][movie] * data[person2][movie] for movie in commonmovies])
# 计算平方和
sum1Sq = sum([pow(data[person1][movie], 2) for movie in commonmovies])
sum2Sq = sum([pow(data[person2][movie], 2) for movie in commonmovies])
# 计算分子
num = sum12 - sum1 * sum2 / n
# 分母
den = sqrt((sum1Sq - pow(sum1, 2) / n) * (sum2Sq - pow(sum2, 2) / n))
if den == 0: return 0
return num / den
def topmatches(data, givenperson, returnernum=5, simscore=sim_pearson):
'''
用户匹配推荐:给定一个用户,返回对他口味最匹配的其他用户
物品匹配: 给定一个物品,返回相近物品
输入参数:对person进行默认推荐num=5个用户(基于用户过滤),或是返回5部电影物品(基于物品过滤),相似度计算用pearson计算
'''
# 建立最终结果列表
usersscores = [(simscore(data, givenperson, other), other) for other in data if other != givenperson]
# 对列表排序
usersscores.sort(key=None, reverse=True)
return usersscores[0:returnernum]
moviedata = transformdata(data)
print("2. 找出跟“超人回归”这电影相关的电影:")
print(topmatches(moviedata, 'Superman Returns'))
print("")
def calSimilarItems(data, num=10):
# 以物品为中心,对偏好矩阵转置
moviedata = transformdata(data)
ItemAllMatches = {}
for movie in moviedata:
ItemAllMatches.setdefault(movie, [])
# 对每个电影 都求它的匹配电影集,求电影之间的距离用pearson距离
ItemAllMatches[movie] = topmatches(moviedata, movie, num, simscore=sim_pearson)
return ItemAllMatches
print("3. 列出所有电影之间的相关性:")
print(calSimilarItems(data))
print("")
"""
推荐用户没看过的电影
某一部未看过电影分数= sum(该部未看过的电影与每一部已看电影之间相似度*已看电影的评分)/sum(未看电影与每一部已看电影之间相似度)
例如:未看电影A,已看电影B,C:
则,电影A分数 = [sim(A,B)*rating(B) +sim(A,C)*rating(C)] / [ sim(A,B) + sim(A,C)]
"""
def getrecommendations(data, targetperson, moviesAllsimilarity):
'''
输入movieAllSimilarity就是上面calsimilarItems已经计算好的所有物品之间的相似度数据集:
'''
# 获得所有物品之间的相似数据集
scoresum = {}
simsum = {}
# 遍历所有看过的电影
for watchedmovie in data[targetperson]:
rating = data[targetperson][watchedmovie]
# 遍历与当前电影相近的电影
for (similarity, newmovie) in moviesAllsimilarity[watchedmovie]: # 取一对元组
# 已经对当前物品评价过,则忽略
if newmovie in data[targetperson]: continue
scoresum.setdefault(newmovie, 0)
simsum.setdefault(newmovie, 0)
# 全部相似度求和
simsum[newmovie] += similarity
# 评价值与相似度加权之和
scoresum[newmovie] += rating * similarity
rankings = [(score / simsum[newmovie], newmovie) for newmovie, score in scoresum.items()]
rankings.sort(key=None, reverse=True)
return rankings
itemsAllsim = calSimilarItems(data) # 这个值会事先计算好
print('4. 基于物品过滤,为用户Toby推荐的电影是:')
print(getrecommendations(data, 'Toby', itemsAllsim))
print()
print("5. 为用户Toby推荐品味相当的用户:")
print(topmatches(data, 'Toby', 3))
print()
"""
推荐未看过的电影:
未看过电影分数=sum(被推荐用户与其他用户之间相似度*用户对该电影评分)/sum(被推荐用户与其他用户之间相似度)
"""
def recommendItems(data, givenperson, num=5, simscore=sim_pearson):
'''
物品推荐:给定一个用户person,默认返回num=5物品
要两个for,对用户,物品 都进行 遍历
'''
# 所有变量尽量用字典,凡是列表能表示的字典都能表示,那何不用字典
itemsimsum = {}
# 存给定用户没看过的电影的其他用户评分加权
itemsum = {}
# 遍历每个用户,然后遍历该用户每个电影
for otheruser in data:
# 不要和自己比较
if otheruser == givenperson: continue
# 忽略相似度=0或小于0情况
sim = simscore(data, givenperson, otheruser)
if sim <= 0: continue
for itemmovie in data[otheruser]:
# 只对用户没看过的电影进行推荐,参考了其他用户的评价值(协同物品过滤是参考了历史物品相似度值)
if itemmovie not in data[givenperson]:
# 一定要初始化字典:初始化itemsum与itemsimsum
itemsum.setdefault(itemmovie, 0)
itemsimsum.setdefault(itemmovie, 0)
# 用户相似度*评价值
itemsum[itemmovie] += sim * data[otheruser][itemmovie]
itemsimsum[itemmovie] += sim
# 最终结果列表,列表包含一元组(item,分数)
rankings = [(itemsum[itemmovie] / itemsimsum[itemmovie], itemmovie) for itemmovie in itemsum]
# 结果排序
rankings.sort(key=None, reverse=True);
return rankings
# 调用此方法如下:
print("6. 为用户Toby推荐未看过的电影:")
print(recommendItems(data, 'Toby', 3))
print()
执行上述代码,屏幕输出:
==============
构建电影推荐系统
==============
http://blog.csdn.net/u013185349/article/details/61192218
1. 电影打分:
{'Lady in the Water': {'Lisa Rose': 2.5, 'Gene Seymour': 3.0, 'Michael Phillips': 2.5, 'Mick LaSalle': 3.0, 'Jack Matthews': 3.0}, 'Snakes on a Plane': {'Lisa Rose': 3.5, 'Gene Seymour': 3.5, 'Michael Phillips': 3.0, 'Claudia Puig': 3.5, 'Mick LaSalle': 4.0, 'Jack Matthews': 4.0, 'Toby': 4.5}, 'Just My Luck': {'Lisa Rose': 3.0, 'Gene Seymour': 1.5, 'Claudia Puig': 3.0, 'Mick LaSalle': 2.0}, 'Superman Returns': {'Lisa Rose': 3.5, 'Gene Seymour': 5.0, 'Michael Phillips': 3.5, 'Claudia Puig': 4.0, 'Mick LaSalle': 3.0, 'Jack Matthews': 5.0, 'Toby': 4.0}, 'You, Me and Dupree': {'Lisa Rose': 2.5, 'Gene Seymour': 3.5, 'Claudia Puig': 2.5, 'Mick LaSalle': 2.0, 'Jack Matthews': 3.5, 'Toby': 1.0}, 'The Night Listener': {'Lisa Rose': 3.0, 'Gene Seymour': 3.0, 'Michael Phillips': 4.0, 'Claudia Puig': 4.5, 'Mick LaSalle': 3.0, 'Jack Matthews': 3.0}}
2. 找出跟“超人回归”这电影相关的电影:
[(0.6579516949597695, 'You, Me and Dupree'), (0.4879500364742689, 'Lady in the Water'), (0.11180339887498941, 'Snakes on a Plane'), (-0.1798471947990544, 'The Night Listener'), (-0.42289003161103106, 'Just My Luck')]
3. 列出所有电影之间的相关性:
{'Lady in the Water': [(0.7637626158259785, 'Snakes on a Plane'), (0.4879500364742689, 'Superman Returns'), (0.3333333333333333, 'You, Me and Dupree'), (-0.6123724356957927, 'The Night Listener'), (-0.9449111825230676, 'Just My Luck')], 'Snakes on a Plane': [(0.7637626158259785, 'Lady in the Water'), (0.11180339887498941, 'Superman Returns'), (-0.3333333333333333, 'Just My Luck'), (-0.5663521139548527, 'The Night Listener'), (-0.6454972243679047, 'You, Me and Dupree')], 'Just My Luck': [(0.5555555555555556, 'The Night Listener'), (-0.3333333333333333, 'Snakes on a Plane'), (-0.42289003161103106, 'Superman Returns'), (-0.4856618642571827, 'You, Me and Dupree'), (-0.9449111825230676, 'Lady in the Water')], 'Superman Returns': [(0.6579516949597695, 'You, Me and Dupree'), (0.4879500364742689, 'Lady in the Water'), (0.11180339887498941, 'Snakes on a Plane'), (-0.1798471947990544, 'The Night Listener'), (-0.42289003161103106, 'Just My Luck')], 'You, Me and Dupree': [(0.6579516949597695, 'Superman Returns'), (0.3333333333333333, 'Lady in the Water'), (-0.250000000000002, 'The Night Listener'), (-0.4856618642571827, 'Just My Luck'), (-0.6454972243679047, 'Snakes on a Plane')], 'The Night Listener': [(0.5555555555555556, 'Just My Luck'), (-0.1798471947990544, 'Superman Returns'), (-0.250000000000002, 'You, Me and Dupree'), (-0.5663521139548527, 'Snakes on a Plane'), (-0.6123724356957927, 'Lady in the Water')]}
4. 基于物品过滤,为用户Toby推荐的电影是:
[(3.610031066802183, 'Lady in the Water'), (3.5313950341859766, 'The Night Listener'), (2.960999860724268, 'Just My Luck')]
5. 为用户Toby推荐品味相当的用户:
[(0.9912407071619299, 'Lisa Rose'), (0.9244734516419049, 'Mick LaSalle'), (0.8934051474415647, 'Claudia Puig')]
6. 为用户Toby推荐未看过的电影:
[(3.3477895267131017, 'The Night Listener'), (2.8325499182641614, 'Lady in the Water'), (2.530980703765565, 'Just My Luck')]