As long as developers follow NewsBlocks Protocol standards they can use any method to create NewsBlocks. For example, the Newslines application uses a combination of Machine Learning  (computer algorithms that do not need to be explicitly programmed, but “learn” to solve problems) , crowdsourcing and verification against graph data, to create a NewsBlock processing pipeline that reduces conversion time of news articles into data from minutes to seconds.

The pipeline extracts data from the article using a variety of processes depending on the data complexity. Simple data, such as time, date, and location is easy to extract with high accuracy.

Category data such as names and Event Types can be extracted and verified against existing graph data. For example, when the names ‘Tony’ and ‘Margaret’ are extracted from an article about Conor McGregor, a comparison to existing data about McGregor’s relationships will show that they are his parents, and not misidentify them as Margaret Thatcher and Tony Blair.

More complex textual data, such as News Event Summaries, requires significant processing, comparison against graph data and crowdsourced verification to ensure accuracy.

To catch automation errors, each Machine Learning module has a corresponding crowdsourcing module that is used to match the NewsBlocks editorial specification. Each piece of data is then verified and combined to create a complete NewsBlock.