Skip to content

Latest commit

 

History

History
65 lines (40 loc) · 1.81 KB

overview.md

File metadata and controls

65 lines (40 loc) · 1.81 KB

Overview

How it works

Diagram of how the tool works

1. Identification

Trusted sources identify and flag terrorist and violent extremist content (TVEC) using their own methods:

2. Matching

Altitude integrates with each of the trusted sources, and consolidates their TVEC signals into one view.

Platforms install Altitude, instead of checking each individual source separately.

Platforms use Altitude to check whether any content on their platform matches the flagged content that’s been consolidated into Altitude.

3. Review & Action

When there’s a match that needs review, small platform moderators click for more detail, review and choose what action they want to take on the content.

Architecture

Diagram of the architecture

Entities

Target

The thing that clients are concerned about and want to “target”, i.e. bring in scope of the system. For example, an image.

Targets are represented as a collection of features. Some of those features are provided by the client (e.g. the raw bytes), while others are fetched or calculated later (e.g. OCR'ed text from an image).

Signal

An input from an external or an internal source that can be used to identify new content for review.

An external source could be URLs provided by TCAP or an ML model that detects violent content. An internal source could be a previously reviewed piece of “bad” content.

Case

A request for a human to visualize a target and its features, and to generate a review.

Review

A single action taken by a human on a case.