| A | B | C | D | E | F | |
|---|---|---|---|---|---|---|
1 | CATEGORY | SUB-CATEGORY | WHAT IS DISCLOSED | DESCRIPTION | EXAMPLES | PRIMARY STAKEHOLDERS |
2 | Documentation | User-specific | Explanations | Show users why they are seeing a specific item recommended to them | Facebook's "Why am I seeing this post?" for in-feed recommendations Amazon's "Frequently bought together" | Individual users |
3 | Inferred interests and attributes | Shows users what interests and attributes the platform has inferred about them based on a combination of their user activity and other data (e.g. location) | Twitter's user "interests" based on the user's profile, activity, and topics the user follows Facebook's "interest categories" under ad settings | Individual users | ||
4 | System Level | Prioritized parameters and signals (alternatively, "the rules" of the system) | Details about the conditions under which the system is operating, what the underlying algorithms are optimizing for and prioritizing: page growth vs. "meaningful social interactions", friends and family over news and politics, etc. | Facebook's downranking political content | General public | |
5 | Content-specific ranking decisions and interventions | Details about the type of content the platform demotes, discourages, or deletes, including announcing updates as and when they happen | Facebook's "Types of Content We Demote" Updates by various platforms on their policy changes for Russian state-controlled media in February 2022 | General public | ||
6 | Change log | Details of new functionality introduced and bugs addressed as well as changes to their community guidelines and speech policies | Reddit's changelog Google Search updates | General public | ||
7 | Transparency reports | A dashboard that shows high-level statistics about various parts of the platforms, including government requests, legal requests, security and integrity initiatives (including taking down networks of malicious actors), etc. | Transparency hubs set up by various platforms (Google, Apple, Twitter) providing aggregated statistics into a subset of policies and actions taken by the platform due to both internal and external factors | General public | ||
8 | Data | Selective datasets in machine-readable format | Provides the general public — or vetted researchers — with a subset of data that illustrates the platform goings-on during a specific time-period or event | Twitter's information operations archives SS1, a consortium of academics looking at building industry partnerships | Academics, researchers, and journalists | |
9 | Recommender outcomes at subgroup levels | Shows how the recommendation engine behaves at a subgroup level: which demographic sees posts about which topics or from which users | General public | |||
10 | APIs | Provides the general public — or vetted researchers — with an API to access selective data | Twitter Developer API Meta's CrowdTangle Reddit API | Academics, researchers, and journalists Independent auditors | ||
11 | Code | Open-source code | Provides a way for people to look at the code that underpins the various feeds users interact with — with relevant weights and signals, it can be a useful way to understand how the recommendation system operates | Twitter's BirdWatch BlueSky Reddit (no longer open-sourced as of 2017) | Academics, researchers, and journalists Independent auditors | |
12 | Research | Internal research | Research conducted internally at the platforms that show platform behavior and how various changes affects the platform | Papers submitted by platforms to various conferences or official blogposts that detail approaches to recommendation engines, algorithms used and how they are tuned, what the system is prioritizing, etc. | Academics, researchers, and journalists Independent auditors General public | |
13 | Industry/academic collaborations | Research conducted with platform data in collaboration with trusted third parties | Twitter's partnership with OpenMined to "advance algorithmic transparency" Facebook's Election 2020 partnerships | Academics, researchers, and journalists General public | ||
14 | Third-party audits | Allow external auditors access to code and workflows to "signal" the quality of the system to regulators and end users, which can declare whether the system has been assessed and deemed trustworthy | Independent auditors appointed by regulators or the organization itself |