Recommender System: Analytical Framework

 5th July 2017 at 10:07am

Dimensions of Analysis

  • Domain
  • Purpose
  • Recommendation Context
  • Whose Opinions
  • Personalization Level
  • Privacy and Trustworthiness
  • Interfaces
  • Recommendation Algorithms


A domain may refer to:

  • News: the news in 今日头条
  • Products: a book recommended by Amazon
  • Matchmaking: online matching system in DotA 2 / 王者荣耀
  • Sequences: music playlists recommended by Netease Music

Recommender should consider choosing whether a new item or new or a re-recommend old one.


  • Sales or other benefits for recommendations themselves (most of the time)
  • Education of user / customer (very rarely)
  • Some other nonsences

Recommendation Context

What's the user doing at the time of recommendation? Shopping, listening to music or reading news?

Whose Opinions

  • Experts: wine shop with comprehensive reviews that only experts can provide
  • Ordinary "phoaks"
  • People like you: movie recommendations on Douban are based-on users' whose taste is similar with you

Privacy and Trustworthiness

  • Privacy: who knows what about me and my data on the recommender system?
  • Trustworthiness: Is the recommendation honest? Or business rules built-in by operator for commercial benefits?


Predictions? Recommendations? Explicit or implicit input?

Recommendation Algorithms

See Recommendation Algorithms.