How does content personalization work for streaming services?

There are two dimensions to this answer:

1) How personalization is done:

a) Demography based- This is usually a qualitative analysis. Based on distinct data sets, users are categorized into demographic segments e.g. 30-35 years of age, married, based out of midwest. Once this is done, each user is mapped into one of these demographic and its assumed that people within a demographic segment behave alike

b) Content based- This a classification process- It is beleived that categories of content go together e.g. While watching a family program, an ad that shows families shopping together will be well received

c) Contextual signals- This is based on certain characters, words on a screen at any given point of time. This will be used to predict the users preferences and thereby used to select an ad

2) What is personalized

a) Page personalization- Choosing videos to recommend, personalization of artwork using collaborative, neural, Bayesian, bandit, or combinatorial to create pages that reliably lead to members finding something great to watch.

b) Search algorithms- using natural language processing, text analytics, machine learning as well as collaborative filtering to  discover new connections between a huge number of textual queries and the many titles in the catalog.

c) Personalization algorithms that decide what to send, when and to whom. Budget allocation algorithms that decide what to advertise, to whom, and for how much. Both of these areas rely on algorithms to determine the delicate balance of cost and incremental reward.