How to start a research that aims at identifying and describing Facebook’s strategies in mapping our desires?
When dealing with such a huge network of data, tools, categories and involved parties, it does not take much to get lost. Our preliminary intuition was to have a look at and register the most basic entry points: the tools that Facebook makes available for advertisers and developers. Facebook’s tools for developers include Products, SDKs and APIs.
Through APIs, a developer is able to retrieve, save or delete data, assuming that the required permission for that is granted by the user to the developer’s app.
The Graph API
The entire ecosystem of Facebook API’s revolves around the Graph API. Named after the social graph, it is the primary way to get data out of, and put data into, Facebook’s platform. The basic components of this API are nodes, edges and fields.
Here is how Facebook defines these components in the Graph API Overview page:
- nodes – basically “things” such as a User, a Photo, a Page, a Comment
- edges – the connections between those “things”, such as a Page’s Photos, or a Photo’s Comments
- fields – info about those “things”, such as a person’s birthday, or the name of a Page
Components apply to user, page, post and profile (which can be user, page, group, event, application).
Those “things” could also be called “objects”, if one were to use programmatic language. Facebook provides a quick way to interact with the Graph API and have a glimpse of data as objects, i.e. in JSON format: the Graph API Explorer.
By default, apps (including the Graph API Explorer) have access to the user’s public profile, which is defined by the user’s privacy settings, and friends list. Additional data requires express authorization by the user. If you’d like to have a look at your data, follow the following steps:
- Log in, authorize the app and select the permissions you grant to it.
- Make some calls to the API. Try, for example: me/feed
What you will see after clicking the Submit button is a piece of text representing the latest posts you published in your timeline, which are themselves represented by the following objects:
- “message” (the text you published on your timeline),
- “created_time” (date and time),
- “id” (a unique identifier)
- and possibly other objects, such as “story” (interactions involving this object, such as sharing).
The Marketing API
One point that will be further investigated, certainly, is the creation of target audiences, as already mentioned by Lídia. There are two ways of defining who will see an ad. With targeting, the admin provides attributes of the audience (such as age, or location) and Facebook will deliver the ad to users that meet those attributes.
With audience, the admin builds a group of people who see the ad.
(Custom) Audiences can be created based on:
- CRM data
- Mobile App
- Facebook Pixel (very much like Google Analytics, but connected to the user’s FB profile)
- Lookalike Audiences (the details forging the concept of ‘similarity’ are completely opaque so far)
- Offline Custom Audiences
- Messenger Custom Audiences
Targeting can be done using:
- Demographics and Events (gender, workplace, education, job title types and relationship status, life events, such as “Recently Moved”)
- Location (country, country group, city, state, zip code, locale, DMA code, electoral districts, geo locations metadata, radius suggestions)
- Interests (interests from someone’s timeline, from Pages liked or from keywords associated with Pages or apps someone uses. Interests names may change anytime)
- Behaviors (user’s actions, digital activities, devices people use, past or intended purchases, and travel)
- Partner Categories (not clear)
Other API’s, such as Atlas and Ads Insight, were also – briefly – investigated but at this point it does not seem that they have many insights to offer concerning the main objective of this research.
Here one can find the research clusters containing the backbones of Facebook’s AI machinery. Publications and code are available.
Topics: Applied Machine Learning, Data Science, Human Computer Interaction – UX, Systems & Networking, Connectivity, Economics & Computation, Natural Language Processing & Speech, Virtual Reality, Computer Vision, Security & Privacy, Facebook AI Research (FAIR). A deeper look at this material can be revealing, but should probably be done at a later stage.