学习协同过滤推荐 w 100行Python代码

python3
import operator
prefs_str = '''
david 百年孤独
david 霍乱时期的爱情
david 从0到1
andy 霍乱时期的爱情
jack 浪潮之巅
jack 失控
jack 创业维艰
michale 寻路中国:从乡村到工厂的自驾之旅
michale 背包十年:我的职业是旅行
ann 活着
ann 霍乱时期的爱情
ann 迟到的间隔年
joel 巨人的陨落:世纪三部曲
joel 中国历代政治得失
joel 人类简史:从动物到上帝
joel 失控
jim 背包十年:我的职业是旅行
jim 迟到的间隔年
ray 霍乱时期的爱情
ray 迟到的间隔年
ray 枪炮、病菌与钢铁:人类社会的命运
'''

# {'andy': {'霍乱时期的爱情': 1},...}
def read_prefs(prefs_str):
    prefs = {}
    for line in prefs_str.split('n'):
        parts = line.rstrip().split()
        if len(parts) == 2:
            userId, itemId = parts
            prefs.setdefault(userId, {})
            prefs[userId].update({itemId:1})
    return prefs

prefs = read_prefs(prefs_str)

def jaccard_distance(prefs, user1, user2):
    s1 = set(prefs[user1].keys())
    s2 = set(prefs[user2].keys())
    return 1.0 * len(s1.intersection(s2)) / len(s1.union(s2))

# 找出与 user1 兴趣最相近的用户 user1 -> [(1.0, user3), (0.8, user4), ...]
def top_matches(prefs, user1, similarity, n = 5):
    scores = [(similarity(prefs, user1, user2), user2) for user2 in prefs if user1 != user2]
    scores.sort()
    scores.reverse()
    return scores[0:n]

# prefs -> "用户 xx 根据 xx 推荐 xx 书籍 xxx"
def calculate_user_cf(prefs, similarity, n = 10):
    ret = {}
    for user in prefs.keys():
        scores = top_matches(prefs, user, similarity, n)
        ret[user] = scores
    return ret

def print_recomendation(prefs, similiar_users, min_score = 0.1):
    # 对每个用户
    for target_user in similiar_users:
        itemId_cnt = {}
        # 找出兴趣相近的用户
        for score, similiar_user in similiar_users[target_user]:
            if score > min_score:
                   # 统计兴趣相近用户最喜欢的书，排序
                for itemId in set(prefs[similiar_user]) - set(prefs[target_user]):
                    itemId_cnt.setdefault(itemId, 0)
                    itemId_cnt[itemId] += 1
        recommends_itemId_cnt = sorted(itemId_cnt.items(), key=operator.itemgetter(1), reverse=True)
        print('n用户:%snt喜欢:%snt相似用户:%snt推荐:%s' % (target_user, list(prefs[target_user].keys()), list(filter(lambda score_user:score_user[0] > min_score, similiar_users[target_user])), recommends_itemId_cnt))

print('n基于用户的协同过滤推荐')
similiar_users = calculate_user_cf(prefs, jaccard_distance, n = 10)
print_recomendation(prefs, similiar_users)

用户:jim
	喜欢:['背包十年:我的职业是旅行', '迟到的间隔年']
	相似用户:[(0.3333333333333333, 'michale'), (0.25, 'ray'), (0.25, 'ann')]
	推荐:[('霍乱时期的爱情', 2), ('活着', 1), ('寻路中国:从乡村到工厂的自驾之旅', 1), ('枪炮、病菌与钢铁:人类社会的命运', 1)]
...

# prefs[userId][itemId] -> prefs[itemId][userId]
def transpose_prefs(prefs):
    ret = {}
    for userId in prefs:
        for itemId in prefs[userId]:
            ret.setdefault(itemId, {})
            ret[itemId][userId] = prefs[userId][itemId]
    return ret

def calculate_item_cf(prefs, similarity, n = 10):
    ret = {}
    itemPrefs = transpose_prefs(prefs)
    # 对于每个物品，找出用户评价向量距离相近的物品
    for item in itemPrefs:
        scores = top_matches(itemPrefs, item, similarity, n)
        ret[item] = scores
    return ret

def print_similiar_items(similiar_items, min_score = 0.1):
    for target_itemId in similiar_items:
        print('n根据 %s 推荐:' % target_itemId)
        score_items = similiar_items[target_itemId]
        for score_item in score_items:
            score, itemId = score_item
            if score > min_score: print('t%s %f' % (itemId, score))

print('n基于书籍的协同过滤推荐')
similiar_items = calculate_item_cf(prefs, jaccard_distance, n = 10)
print_similiar_items(similiar_items)

根据 创业维艰 推荐:
	浪潮之巅 1.000000
	失控 1.000000
...