Event Registry data

Classification of AI news by AI Principle and policy area 

A list of keywords was created for each AI Principle and policy area by selecting the most relevant related concepts in Event Registry’s search engine. To be classified under a certain AI Principle or policy area, a news article must a) be related to AI, and b) contain at least one of the concepts specified for that AI Principle or policy area in its title and/or abstract. 

Selection of related online news by AI Principle and policy area 

The most relevant news events from the past six months are shown in the “Related online news from Event Registry” lists in both AI Principle and policy areas dashboards. A news “event” is defined as a collection of news articles covering the same news. It serves as a measure of news impact.

News sentiment analysis 

A ‘sentiment’ score is assigned to each news article and event to determine whether an article or event is ‘positive’, ‘neutral’, or ‘negative’. The sentiment score is calculated using the “VADER” (Valence Aware Dictionary and sEntiment Reasoner) open-source service, “a lexicon and rule-based sentiment analysis tool that is specifically attuned to sentiments expressed in social media.”

VADER contains a lexicon of words for which a sentiment score has been computed. The sentiment score is a floating value between -1 (negative) and 1 (positive). Additionally, VADER contains a set of rules to determine how sentiment is emphasised with individual words (for instance, “very” emphasises the sentiment, while negations inverse it). The sentiment score of an article results from the analysis of the first five sentences of the article. In practice, news articles rarely have extreme values close to 1 and -1; thus, scores below -0.1 are classified as negative; between -0.1 and 0.1 as neutral; and above 0.1 as positive.