Machine-readable news is news delivered as a structured data feed that a computer can process without a person reading it. Each item arrives with fields that software can act on at once: a precise timestamp, codes identifying the companies, currencies or commodities concerned, topic tags such as earnings or merger, and often scores for relevance, novelty and tone. Feeds are sold by financial news agencies and specialist analytics firms and are delivered through an application programming interface. Economic releases are offered in the same form.
The feeds exist to serve news-based trading and other forms of algorithmic trading. A program can widen its quotes, pause or place orders within milliseconds of a headline, long before a human could respond, which makes low latency delivery a selling point. The tone scores are produced by sentiment analysis, a branch of natural language processing. Automated reading can misfire on false, stale or ambiguous stories.