基于淘宝点击及购买记录的口碑商家推荐——基于物品的协同过滤

Table_2: Users’shopping records at brick-and-mortar stores before Dec. 2015. (ijcai2016_koubei_train)
User_id、Merchant_id、Location_id 、Time_Stamp
Table_3: Merchant information. (ijcai2016_merchant_info)
Merchant_id 、Budget(budget constraints imposed on the merchant) 、Location_id_list
Table_4: Prediction result. (ijcai2016_koubei_test)
User_id、Location_id、Merchant_id_list

首先统计出下面几个字典：
{用户:[商家]}
{商家:[用户]}
{商家:商家热度} （热度用用户访问次数来表示）
{商圈:{商家:商家热度}} （方便后面的推荐）
商家间的相似度矩阵（其实是存储在字典中）

1、对于某用户，若其所在商圈内商家少于10个，则全部推荐，并且按照商家热度从高到低排序。
2、冷启动问题，若该用户之前没有使用过口碑，即没有历史商家记录，则推荐所在商圈热门的前十个商家（需排序）。
3、如果有历史商家记录，分别对消费过的商家找到相似商家和相似度，对于所有消费过的商家设置等权，累计同一商家的相似度，按照相似度从大到小排序，推荐前十个。

# -*- coding:utf-8 -*-
__author__ = 'Bai'
import os,math,csv
os.chdir('c:/Bai/datasets/tianchi_ijcai2016')
f1 = open('ijcai2016_merchant_info')
context1 = f1.readlines()
f2 = open('ijcai2016_koubei_train')
context2 = f2.readlines()
f3 = open('ijcai2016_koubei_test')
context3 = f3.readlines()

csvfile = file('forcast.csv','wb')
writer = csv.writer(csvfile)

#统计字典: {用户:[商家]}、{商家:[用户]} 及 {商家:商家热度}
user_merchant = {}
merchant_user = {}
merchant_popular = {}
for x in context2:
    x = x.replace('\n','').split(',')
    if x[0] not in user_merchant:
        user_merchant[x[0]] = set()
    user_merchant[x[0]].add(x[1])
    if x[1] not in merchant_user:
        merchant_user[x[1]] = set()
        merchant_popular[x[1]] = 0
    merchant_user[x[1]].add(x[0])
    merchant_popular[x[1]] += 1

#统计字典： {商圈:{商家:商家热度}}
district_merchant = {}
for x in context1:
    x = x.replace('\n','').split(',')
    y = x[2].split(':')
    for j in y:
        if j not in district_merchant:
            district_merchant[j] = {}
        district_merchant[j][x[0]] = merchant_popular[x[0]]

#统计商家间的相似度矩阵
C = {}
N = {}
for user in user_merchant:
    for i in user_merchant[user]:
        if i not in N:
            N[i] = 0
        N[i] += 1
        if i not in C:
            C[i] = {}
        for j in user_merchant[user]:
            if i == j:
                continue
            if j not in C[i]:
                C[i][j] = 0
            C[i][j] += 1
W = {}
for i in C:
    if i not in W:
        W[i] = {}
    for j in C[i]:
        W[i][j] = C[i][j]/math.sqrt(N[i] * N[j])

for i in context3:
    i = i.replace('\n','').split(',')
    #若商圈内商家数量少于十个，则全部推荐
    if len(district_merchant[i[1]]) <= 10:
        #将商家按照热度排序
        s = sorted(district_merchant[i[1]].iteritems(),key=lambda t:t[1],reverse=True)
        for j in range(len(s)):
            if j == 0:
                a = s[j][0]
            else:
                a = a + ':' + s[j][0]
    #冷启动：若用户没有口碑商家记录，则推荐所在商区热门商家前十个，按照商家热度从大到小的顺序
    elif i[0] not in user_merchant:
        s = sorted(district_merchant[i[1]].iteritems(),key=lambda t:t[1],reverse=True)
        for j in range(10):
            if j == 0:
                a = s[j][0]
            else:
                a = a + ':' + s[j][0]
    else:
        e = set()
        sim_merchant = {}
        for u in user_merchant[i[0]]:
            for v in W[u]:
                if v not in sim_merchant:
                    sim_merchant[v] = 0
                sim_merchant[v] += W[u][v]

        s2 = sorted(sim_merchant.iteritems(),key=lambda t:t[1],reverse=True)
        j = 0
        for k in s2:
            e.add(k[0])
            if j == 10:
                break
            if j == 0:
                a = k[0]
            else:
                a = a + ':' + k[0]
            j += 1
        if j < 10:
            s = sorted(district_merchant[i[1]].iteritems(),key=lambda t:t[1],reverse=True)
            l = 0
            n = 0
            while n < 10-j:
                if s[l][0] in e:
                    continue
                else:
                    if len(a) == 0:
                        a = s[l][0]
                    else:
                        a = a + ':' + s[l][0]
                    n += 1
                l += 1
    i.append(a)
    writer.writerow(i)
csvfile.close()