🛒 第26课：电商推荐项目

🎯 学习目标

🗂️ 课程定位

📖 核心理论

电商推荐是最复杂的推荐场景之一，需要处理多类型用户行为（浏览/点击/加购/购买）、丰富的商品特征、实时价格变化、库存限制等。电商推荐直接关联商业指标，是推荐系统最重要的落地方向。

📐 数学公式

💻 代码实现

import numpy as np
from collections import defaultdict
np.random.seed(42)
print("="*60+"\n电商推荐项目\n"+"="*60)
n_users=40;n_products=35;n_categories=5
cats=["数码","服饰","食品","家居","美妆"]
# 商品特征
product_cat=np.random.randint(n_categories,size=n_products)
product_price=np.random.lognormal(3,1,n_products).astype(int)+10
product_brand=np.random.randint(5,size=n_products)
product_quality=np.random.random(n_products)*2+3
print(f"用户:{n_users} 商品:{n_products} 类目:{n_categories}")
# 用户行为序列
user_sequences=defaultdict(list)
user_profiles=np.random.randint(n_categories,size=n_users)
for u in range(n_users):
    n_actions=np.random.randint(5,20)
    seq=[]
    for _ in range(n_actions):
        cat=user_profiles[u]
        cat_products=np.where(product_cat==cat)[0]
        if len(cat_products)==0:cat_products=np.arange(n_products)
        action=np.random.choice(cat_products)
        action_type=np.random.choice(["浏览","点击","加购","购买"],p=[0.4,0.3,0.2,0.1])
        seq.append((action,action_type))
    user_sequences[u]=seq
print(f"总行为数:{sum(len(s)for s in user_sequences.values())}")
# 多路召回
print("\n--- 多路召回 ---")
target=0
# 1. 协同过滤召回
co_occurrence=defaultdict(int)
for u,seq in user_sequences.items():
    purchased=[a for a,t in seq if t=="购买"]
    for i in range(len(purchased)):
        for j in range(i+1,len(purchased)):
            co_occurrence[(purchased[i],purchased[j])]+=1
            co_occurrence[(purchased[j],purchased[i])]+=1
# 基于用户历史购买召回
target_purchased=set(a for a,t in user_sequences[target]if t=="购买")
cf_candidates=set()
for p in target_purchased:
    for(q,c)in co_occurrence.items():
        if q[0]==p:cf_candidates.add(q[1])
print(f"CF召回:{len(cf_candidates)}个商品")
# 2. 类目召回
target_cat=user_profiles[target]
cat_candidates=set(np.where(product_cat==target_cat)[0])
print(f"类目召回:{len(cat_candidates)}个商品")
# 3. 热门召回
product_popularity=defaultdict(int)
for u,seq in user_sequences.items():
    for a,t in seq:
        weight={"浏览":1,"点击":2,"加购":3,"购买":5}[t]
        product_popularity[a]+=weight
hot_candidates=set([i for i,_ in sorted(product_popularity.items(),key=lambda x:-x[1])[:10]])
print(f"热门召回:{len(hot_candidates)}个商品")
# 合并召回
all_candidates=cf_candidates|cat_candidates|hot_candidates
all_candidates-=target_purchased
print(f"合并召回:{len(all_candidates)}个商品")
# 精排
print("\n--- 精排 ---")
# 特征:商品质量+类目匹配+价格匹配+热度
scores={}
for p in all_candidates:
    feat_quality=product_quality[p]/5
    feat_cat=1 if product_cat[p]==user_profiles[target] else 0.3
    feat_price=1-abs(product_price[p]-np.median(product_price))/np.max(product_price)
    feat_pop=product_popularity.get(p,0)/max(product_popularity.values())
    scores[p]=0.3*feat_quality+0.25*feat_cat+0.15*feat_price+0.3*feat_pop
ranked=sorted(scores.items(),key=lambda x:-x[1])
print(f"精排Top-10:")
for p,s in ranked[:10]:
    print(f"  商品{p}:分数={s:.3f} 类目={cats[product_cat[p]]} 价格={product_price[p]}")
# 重排(类目打散)
print("\n--- 重排(类目打散) ---")
reranked=[];cat_count=defaultdict(int)
for p,s in ranked:
    c=product_cat[p]
    if cat_count[c]<3:  # 每类目最多3个
        reranked.append((p,s));cat_count[c]+=1
print(f"重排Top-10:")
for p,s in reranked[:10]:
    print(f"  商品{p}:分数={s:.3f} 类目={cats[product_cat[p]]} 价格={product_price[p]}")
# 多样性评估
cat_coverage=len(set(product_cat[p]for p,_ in reranked[:10]))/n_categories
print(f"类目覆盖率:{cat_coverage:.2%}")

🔴 运行结果

✅ 验证通过

🔍 代码解读

💡 关键洞察

📝 本章小结

🧪 练习

📚 延伸阅读

💼 面试考点

🏭 工业案例