IDS 200 Lecture Notes - Lecture 4: Nosql, Management System, Pagerank
12 Feb 2020
School
Department
Course
Professor
Document Summary
Initially, facebook (or thefacebook) worked in directory mode, like myspace. Users read what others had on their profiles. News feed launched in 2006, when facebook didn"t have nearly as much traffic as now. Initially, it was viable to show every user the complete, chronological set of recent activity. With scale, it became necessary to prioritize and effectively throttle the feed. More friends per user, posting more often. Advertisers want to reach users but users aren"t interested in most ads. Borrowing from google"s pagerank, used a multi-factor formula to rank posts: All else equal, post ranks continually decline. This decay can, for a time, be outweighed by user interactions with the post. Facebook continually gathers data about user actions and interactions. Given enough data, more accurate predictions of user behavior are possible. Typically, only a few variables are strong predictors. Combining the many less-relevant variables can materially improve prediction success. Now, facebook uses many variables and the mix is intermittently changes.
