人工智能推荐系统测验

当您参加测验时，就会出现答案和解释。

1. “协同过滤”依靠什么来进行推荐？
   • 许多具有相似品味的用户的模式
   • 仅对每个项目进行文字描述
   • 产品价格
   • 固定的编辑前 10 名列表
2. 基于内容的过滤主要通过以下方式推荐项目：
   • 目录随机抽样
   • 将项目功能与您过去的偏好相匹配
   • 计算全球总销售额
   • 要求其他用户投票
3. What did the Netflix Prize seek to improve by 10% over the Cinematch baseline?
   • Click-through rate for every recommended film
   • The number of films watched per user
   • Root mean squared error (RMSE) on the competition’s held-out ratings
   • A user’s average watch time on the service
4. 在矩阵分解中，用户和项目表示为：
   • 单个是/否标志
   • 纯文本评论
   • 日历日期
   • 隐藏“潜在因素”的向量
5. What mechanism did the Deep Interest Network paper propose for modeling a user’s different interests across candidate ads?
   • A local activation unit that forms a candidate-specific interest representation
   • A fixed user vector that is identical for every candidate
   • A rule that selects only the most popular ad
   • A graph algorithm that proves every ad is relevant
6. “混合”推荐器是这样的：
   • 只在周末工作
   • 结合协作和基于内容的方法
   • 完全不使用任何数据
   • 只推荐最新的商品
7. What is one limitation of collaborative filtering for a new item with no interactions?
   • It cannot use ratings
   • It requires item descriptions only
   • It may lack the interaction data needed to recommend the item
   • It always recommends every new item
8. What should a recommender evaluation measure besides click-through rate?
   • 仅计算模型的参数
   • Only training loss
   • Only the number of recommendations
   • User outcomes and harms such as diversity, exposure, or unwanted engagement