个性化推荐算法python实现

基于ItemCF算法

# -*- coding: utf-8 -*-
"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei

"""

import math


# ItemCF算法
def ItemSimilarity(train):
    C = dict()
    N = dict()
    for u,items in train.items():
        for i in items.keys():
            N[i] += 1
            for j in items.keys():
                if i == j:
                    continue
                C[i][j] += 1
    W = dict()
    for i,related_items in C.items():
        for j,cij in related_items.items():
            W[i][j] = cij / math.sqrt( N[i] * N[j])
    return W


# ItemCF-IUF算法
def ItemSimilarity_v2(train):
    C = dict()
    N = dict()
    for u,items in train.items():
        for i in items.keys():
            N[i] += 1
            for j in items.keys():
                if i == j:
                    continue
                C[i][j] += 1 / math.log(1+len(items)*1.0)
    W = dict()
    for i,related_items in C.items():
        for j,cij in related_items.items():
            W[i][j] = cij / math.sqrt( N[i] * N[j])
    return W


def Recommend(train,user_id,W,K):
    rank = dict()
    ru = train[user_id]
    for i,pi in ru.items():
        for j,wj in sorted(W[i].items,key=itemgetter(1),reverse=True)[0:K]:
            if j in ru:
                continue
            rank[j] += pi*wj

    return rank

基于UserCF算法

# -*- coding: utf-8 -*-
"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei
"""

import math
'''
基于UserCF的推荐算法
'''
# UserCF算法
def UserSimilarity(train):
    item_users = dict()
    for u,items in train.items():
        for i in items.keys():
            if i not in item_users:
                item_users[i] = set()
            item_users[i].add(u)
    C = dict()
    N = dict()
    for i,users in item_users.items():
        for u in users:
            N[u] += 1
            for v in users:
                if u == v:
                    continue
                C[u][v] += 1
    W = dict()
    for u,related_users in C.items():
        for v,cuv in related_users.items():
            W[u][v] = cuv / math.sqrt(N[u] * N[v])
    return W


# User-IIF算法
def UserSimilarity_v2(train):
    item_users = dict()
    for u,items in train.items():
        for i in items.keys():
            if i not in item_users:
                item_users[i] = set()
            item_users[i].add(u)
    C = dict()
    N = dict()
    for i,users in item_users.items():
        for u in users:
            N[u] += 1
            for v in users:
                if u == v:
                    continue
                C[u][v] += 1 / math.log(1+len(users))
    W = dict()
    for u,related_users in C.items():
        for v,cuv in related_users.items():
            W[u][v] = cuv / math.sqrt(N[u] * N[v])
    return W


def Recommend(user,train,W):
    rank = dict()
    interacted_items = train[user]
    for v,wuv in sorted(W[u].items,key=itemgetter(1),reverse=True)[0:K]:
        for i,rvi in train[v].items:
            if i in interacted_items:
                continue
            rank[i] += wuv*rvi
    return rank

基于时间上下文的个性化推荐

# -*- coding: utf-8 -*-
"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei
"""

import math

def RecentPopularity(records,alpha,T):
    ret = dict()
    for user,item,tm in records:
        if tm >= T:
            continue
        addToDict(ret,item,1/(1.0+alpha*(T-tm)))
    return ret


def addToDict(dicts,item,value):
    pass


def ItemSimilarity(train,alpha):
    C = dict()
    N = dict()
    for u,items in train.items():
        for i,tui in items.items():
            N[i] += 1
            for j,tuj in items.items():
                if i == j:
                    continue
                C[i][j] += 1 / (1+alpha*abs(tui-tuj))
    W = dict()
    for i,related_items in C.items():
        for j,cij in related_items.items():
            W[i][j] = cij / math.sqrt(N[i] * N[j])
    return W


def RecommendItemCF(train,user_id,W,K,t0):
    rank = dict()
    ru = train[user_id]
    for i,pi in ru.items():
        for j,wj in sorted(W[i].items(),\
                key=itemgetter(1),reverse=True)[0:K]:
            if j,tuj in ru.items():
                continue
            rank[j] += pi * wj / (1 + alpha * (t0 - tuj))
    return rank




def UserSimilarity(train):
    item_users = dict()
    for u,items in train.items():
        for i,tui in items.items():
            if i not in item_users:
                item_users[i] = dict()
            item_users[i][u] = tui


    C = dict()
    N = dict()
    for i,users in item_users.items():
        for u,tui in users.items():
            N[u] += 1
            for v,tvi in users.items():
                if u == v:
                    continue
                C[u][v] += 1 / (1 + alpha * abs(tui - tvi))
    W = dict()
    for u,related_users in C.items():
        for v,cuv in related_users.items():
            W[u][v] = cuv / math.sqrt(N[u] * N[v])
    return W


def RecommendUserCF(user,T,train,W):
    rank = dict()
    interacted_items = train[user]
    for v,wuv in sorted(W[u].items,key=itemgetter(1),\
            reverse=True)[0:K]:
        for i,tvi in train[v].items:
            if i in interacted_items:
                continue
            rank[i] += wuv / (1 + alpha * (T - tvi))
    return rank

基于LFM算法的个性化推荐

# -*- coding: utf-8 -*-

"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei
"""

'''
items => {'12':'PHP','1203':'Storm','123':'Ubuntu'}
items_pool => [12,32,121,324,532,123,53,1203,429,2932]
user_items => {'1010':[12,1203,123,429]}
'''
def RandomSelectNagativeSample(items):
    ret = dict()
    for i in items.keys():
        ret[i] = 1
    n = 0
    for i in range(0,len(items)*3):
        item = items_pool[random.randint(0,len(items_pool)-1)]
        if item in ret:
            continue
        ret[item] = 0
        n += 1
        if n > len(items):
            break
    return ret




def InitModel(user_items,F):
    P = dict()
    Q = dict()
    for u in user_items.keys():
        if u not in P:
            P[u] = {}
        for f in range(0,F):
            P[u][f] = 1


    items = user_items.values()
    itemLen = len(items[0])
    i = 0
    while i< itemLen:
        ii = items[0][i]
        if ii not in Q:
            Q[ii] = {}
        for f in range(0,F):
            Q[ii][f] = 1
        i += 1
    return [P,Q]




def LatentFactorModel(user_items,F,N,alpha,lambda1):
    [P,Q] = InitModel(user_items,F)
    for setup in range(0,N):
        for user,items in user_items.items():
            samples = RandomSelectNagativeSample(items)
            for item,rui in samples.items():
                eui = rui - Predict(user,item)
                for f in range(0,F):
                    P[user][f] += alpha * (eui * Q[item][f] - lambda1 * P[user][f])
                    Q[item][f] += alpha * (eui * P[user][f] - lambda1 * Q[item][f])
        alpha *= 0.9
    return [P,Q]




def Recommend(user,P,Q):
    rank = dict()
    for f,puf in P[user].items():
        for i,pfi in Q[f].items():
            if i not in rank:
                rank[i] += puf * qfi
    return rank






def PersonalRank(G,alpha,root,maxsetup):
    rank = dict()
    #rank = {x:0 for x in G.keys()}
    rank = rank.fromkeys(G.keys(),0)
    rank[root] = 1
    for k in range(maxsetup):
        tmp = dict()
        #tmp = {x:0 for x in G.keys()}
        tmp = tmp.fromkeys(G.keys(),0)
        for i,ri in G.items():
            for j,wij in ri.items():
                if j not in tmp:
                    tmp[j] = 0
                tmp[j] += alpha * rank[i]/(1.0*len(ri))
                if j == root:
                    tmp[j] += 1 - alpha
        rank = tmp


        print 'iter:' + str(k) + "\t",
        for key,value in rank.items():
            print "%s:%.3f,\t" % (key,value),
        print
    return rank


if __name__ == '__main__':
    G = {'A':{'a':1,'c':1},
     'B':{'a':1,'b':1,'c':1,'d':1},
     'C':{'c':1,'d':1},
     'a':{'A':1,'B':1},
     'b':{'B':1},
     'c':{'A':1,'B':1,'C':1},
     'd':{'B':1,'C':1}}
    PersonalRank(G,0.85,'A',20)

'''

#items_pool = {'12':'PHP','32':'Nginx','121':'Apache','324':'Erlang','532':'Linux','123':'Ubuntu','53':'Java','1203':'Storm','429':'Kafka','2932':'Flume'}
items_pool = [12,32,121,324,532,123,53,1203,429,2932]
items = {'12':'PHP','1203':'Storm','123':'Ubuntu'}
user_items = {'1010':[12,1203,123,429]}


#print RandomSelectNagativeSample(items)
print InitModel(user_items,4)
'''

基于图的推荐算法

# -*- coding: utf-8 -*-
"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei
"""

'''
    基于图的推荐算法,二分图
'''




def PersonalRank(G,alpha,root,maxsetup):
    rank = dict()
    #rank = {x:0 for x in G.keys()}
    rank = rank.fromkeys(G.keys(),0)
    rank[root] = 1
    for k in range(maxsetup):
        tmp = dict()
        #tmp = {x:0 for x in G.keys()}
        tmp = tmp.fromkeys(G.keys(),0)
        for i,ri in G.items():
            for j,wij in ri.items():
                if j not in tmp:
                    tmp[j] = 0
                tmp[j] += alpha * rank[i]/(1.0*len(ri))
                if j == root:
                    tmp[j] += 1 - alpha
        rank = tmp


        print 'iter:' + str(k) + "\t",
        for key,value in rank.items():
            print "%s:%.3f,\t" % (key,value),
        print
    return rank


if __name__ == '__main__':
    G = {'A':{'a':1,'c':1},
     'B':{'a':1,'b':1,'c':1,'d':1},
     'C':{'c':1,'d':1},
     'a':{'A':1,'B':1},
     'b':{'B':1},
     'c':{'A':1,'B':1,'C':1},
     'd':{'B':1,'C':1}}
    PersonalRank(G,0.85,'C',20)

基于标签的推荐算法

# -*- coding: utf-8 -*-
"""
Created on Wed Sep 30 15:36:15 2015


@author: tanyouwei
"""

import math


#标签流行度算法
def TagPopularity(records):
    tagfreq = dict()
    for user,item,tag in records:
        if tag not in tagfreq:
            tagfreq[tag] = 1
        else:
            tagfreq[tag] += 1
    return tagfreq


#物品相似度余弦算法
def CosineSim(item_tags,i,j):
    ret  = 0
    for b,wib in item_tags[i].items():
        if b in item_tags[j]:
            ret += wib * item_tags[j][b]
    ni = 0
    nj = 0
    for b,w in item_tags[i].items():
        ni += w * w
    for b,w in item_tags[j].items():
        nj += w * w
    if ret == 0:
        return 0
    return ret / math.sqrt(ni * nj)


#推荐物品的多样性算法
def Diversity(item_tags,recommend_items):
    ret = 0
    n = 0
    for i in recommend_items.keys():
        for j in recommend_items.keys():
            if i == j:
                continue
            ret += CosineSim(item_tags,i,j)
            n += 1
    return ret / (n * 1.0)




def addValueToMat(dicts,index,k,v):
    if index not in dicts:
        dicts[index] = dict()
        dicts[index][k] = v
    else:
        if k not in dicts[index]:
            dicts[index][k] = v
        else:
            dicts[index][k] += v


def InitStat(records):
    user_tags = dict() #存储 user_tags[u][b] = n(u,b)
    tag_items = dict() # tag_items[b][i] = n(b,i)
    user_items = dict()
    for user,item,tag in records.items():
        addValueToMat(user_tags,user,tag,1)
        addValueToMat(tag_items,tag,item,1)
        addValueToMat(user_items,user,item,1)




def Recommend(user):
    recommend_items = dict()
    tagged_items = user_items[user]
    for tag,wut in user_tags[user].items():
        # wut = wut*1.0/math.log(1+len(tag_users[tag])) #TagBasedTFIDF and TagBasedTFIDF++
        for item,wti in tag_items[tag].items():
            # wti = wti*1.0/math.log(1+len(user_items[user])) #TagBasedTFIDF++
            if item in tagged_items:
                continue
            if item not in recommend_items:
                recommend_items[item] = wut * wti
            else:
                recommend_items[item] += wut * wti
    return recommend_items




if __name__ == "main":
    user_tags = dict()
    user_items = dict()
    tag_items = dict()


    records = dict()
    user = '1220';


    InitStat(records)
    rec_items = Recommend(user)