【Week 1】Predictions and Recommendations —— The output of recommender

内容来源：“Recommender Systems Specialization” course of 明尼苏达大学 (University of Minnesota) at Coursera

Learning Objectives

• To understand the ways in which recommender output can be used
• To understand the distinction between predictions and recommendations
• To understand the distinction between organic and explicit presentation
• To review examples and understand which presentation makes most sense in different applications.

Predictions

• Estimates of how much you’ll like an item
  • Often scaled to match some rating scale
  • Often tied to search or browsing for specific products

存在潜在风险，假设系统预测用户会喜欢该电影，但用户观看后不喜欢，会导致用户失去对系统推荐质量的信任。

Recommendations

Recommendation推荐用户可能喜欢的列表，而不做明确判断，避免了上述问题的产生。

• Recommendations are suggestions for items you might like (or might fit what you’re doing)
  • Often presented in the form of “top-n lists”（经常以top-n的形式返回）
  • Also sometimes just placed in front of you（有时仅是展示一些可能感兴趣的商品，而不做排序）

Prediction and Recommendation

• Often, the two come together
• Predictions:
  • Pro: helps quantify item
  • Con: provides something falsifiable（定量的预测可能会出错，用户发现该错误会降低对系统的信任度）
• Recommendations
  • Pro: provides good choices as a default
  • Con: if perceived as top-n, can result in failure to explore (if top few seem poor)（用户会认为推荐列表是按照偏好程度降序排列的，当用户对前几个推荐商品不满意/满意时，会放弃继续浏览，即放弃explore过程）

Another dimension to consider

具体推荐场景下，该如何利用推荐系统的输出来给用户做推荐。

• How explicit is the prediction or recommendation (vs. organic)?
  • Historical note: we paid for it, we’ll let you know （过往商家花了很多钱购买推荐系统服务，因此他们会将系统输出一股脑提供给用户）
  • Today: balance between explicit prediction (falsifiable) and coarser granularity (you might like this!)
    （直接给出explicit prediction，可能会造成预测错误，并且会很突兀，存在令人感到被强迫购买等问题。因此可以考虑更舒适的表达，例其它人认为这部电影可以达到4.5分，你可能会喜欢它）
  • Today: balance between theses are the best (top-n) and softer presentation (here are some that might be interesting)
    （直接推荐top-n会导致用户放弃explore过程，同样可以换更舒适的表达，例如这些商品可能你会比较喜欢）

That balance between organic and explicit is one that you have to strike as you design an application. And you should understand that there are advantages at every point along that spectrum. There isn’t a single correct that has to fit your application.