Architectural Components

Architecture

Let’s have a look at the high-level architecture of the system. There will be two main actors involved in our ad prediction system - platform users and advertiser. Let’s see how they fit in the architecture:

1. Advertiser flow

Advertisers create ads containing their content as well as targeting, i.e., scenarios in which they want to trigger their ads. A few examples are:

• Query-based targeting: This method shows ads to the user based on the query terms. The query terms can be a partial match, full match, expansion, etc.
  For example, whenever a user types a query “machine learning course”, the system shows the ML course on educative.io.

• User-based targeting: The ads will be subjective to the user based on a specific region, demographic, gender, age, etc.

• Interest-based targeting: This method shows interest-based ads. Assume that on Facebook, the advertiser might want to show ads based on certain interest hierarchies. For example, the advertiser might like to show sports-related ads to people interested in sports.

• Set-based targeting: This type shows ads to a set of users selected by the advertisers. For example, showing an ad to people who were previous buyers or have spent more than ten minutes on the website. Here, we can expand our set and do seed audience expansion.

2. User flow

As the platform user queries the system, it will look for all the potential ads that can be shown to this user based on different targeting criteria used by the advertiser.

So, the flow of information will have two major steps as described below:

• Advertisers create ads providing targeting information, and the ads are stored in the ads index.
• When a user queries the platform, ads can be selected from the index based on their information (e.g., demographics, interests, etc.) and run through our ads prediction system.

Let’s briefly look at each component here. Further explanation will be provided in the following lessons.

Ad selection

The ad selection component will fetch the top k ads based on relevance (subject to the user context) and bid from the ads index.

Ad prediction

The ad prediction component will predict user engagement with the ad (the probability that an action will be taken on the ad if it is shown), given the ad, advertiser, user, and context. Then, it will rank ads based on relevance score and bid.

Auction

The auction mechanism then determines whether these top K relevant ads are shown to the user, the order in which they are shown, and the price the advertisers pay if an action is taken on the ad.

For every ad request, an auction takes place to determine which ads to show. The top relevant ads selected by the ad prediction system are given as input to Auction. Auction then looks at total value based on an ad’s bid as well as its relevance ...