An article is a news article that we collected and imported into Event Registry. Articles are collected using RSS feeds and contain properties such as title, body, date, time, source (news publisher), image, a list of concepts, a list of categories, a list of extracted dates (dates mentioned in the article), and other available metadata.

Event Registry categorizes the news into a taxonomy of categories from DMOZ. The DMOZ taxonomy has over 1 million categories organized into several levels. In Event Registry, we use only the top 3 levels of taxonomy which amounts to about 50,000 categories – from generic ones (like Business) to very specific ones (like Business/Investing/Derivatives).

Categories are assigned to articles and events using machine learning models that are trained for each category separately. Categories do not represent a particular mention in the articles or events, but instead represent what topic the content is about. Categories are currently assigned only to content in the English language.

Each article has a news source (like The New York Times, The Wall Street Journal, The Guardian, etc.). A news source can have properties such as URI (host name), title, description, location (geographical location), and importance.

A location in Event Registry stands for a geographical location and can be associated with articles, events, news sources (location of the publisher), and certain concepts (those that represent locations). Information about geographical locations are obtained from GeoNames. Details that can be returned about a geographical location include GeoNames ID, wiki URL (URL of the corresponding Wikipedia page), population, geo location, area, continent, and others.

...or phrases (multiple words) that should be mentioned in your results.

You can search using keywords and phrases by typing them into the search box and selecting the first option in the auto-suggest drop-down menu.

For each event, we might identify articles that report about the event in different languages.

Each concept has various properties, such as labels in different languages, synonyms, image, and description. Concepts are very important because they allow you to search for things based on the thing itself and not based on the keyword. For example, if you want to find all Spanish articles that mention the White House, using the keyword "White House" will not give you desired results as in Spanish the White House translates to "Casa Blanca". If you use the concept URI for the White House instead of the keyword, you can find mentions of the White House in all languages. You can use concepts by typing in the search input box a part of the concept name and selecting the correct concept from the list of autosuggestions.

Furthermore, searching with concepts provides you with more accurate results. Event Registry is able to read the context in the news articles and distinguish for example, if the news article is about Apple – the company, or an apple – the fruit. By using the concepts, Event Registry will also find the results that mention various synonyms that are commonly used when referring to it (like Apple Inc., Apple, Apple Computer Company, and Apple Computer, Inc.).

If you find that you are obtaining less search results than expected, we suggest that you try searching by the keyword or phrase instead of the concept. This can be especially important in cases when the concept is not easy to recognize or is highly ambiguous.

The related articles feature represents a list of articles that are not assigned to the current event but are potentially related to it. You can adjust the settings to select which properties should be used when determining related articles.

Top concepts in articles represent a list of entities (people, organizations, and locations) and non-entities that best describe what articles forming an event are about.

The computation of top concepts can be done in 3 ways - by importance, frequency, or uniqueness.

Importance

The importance criterion means that the concepts which are more important in the results and appear more frequently in the results will get a higher score. The most important concept will get a score of 100, while the remaining ones get proportionally lower scores.

Frequency

In this case, the concepts are scored based on how frequently they appear in the results. The value that is shown for each concept is the percentage of the results in which the concept appears.

Uniqueness

The uniqueness criterion identifies the concepts that are more likely mentioned in the resulting articles and much less likely mentioned in all other articles. This criterion is useful for identifying unique concepts that are otherwise not frequently present in the articles but are more specific to the topic of your search.

Tag cloud displays top keywords extracted from the articles. The larger the keyword, the more frequently it was mentioned in the news articles.

The timeline of articles displays the publishing times of articles describing the event. The height of the curve indicates the number of articles published in last 8 hours and therefore shows how intensive the reporting on that event was. Each differently colored curve represents a set of articles in a different language. The size of the point indicates how frequently the article was shared on social media.

When analyzing individual news articles we also look at which dates are being mentioned in the article. The visualization displays how many times individual dates are mentioned in the articles about the event.

The map shows the geographical locations of the news sources that published articles about the selected event. When there is a high density of news sources at a nearby location, the map shows a cluster that displays the number of sources that are located in that geographical area. By zooming in, you can then find the specific locations of the sources.

The timeline below the map represents the time distribution of articles. It is an interactive chart and you can click and drag to select a specific time period of interest and highlight only the news sources that reported in that time period.

Events are typically not completely unrelated to other events. The similar events feature displays a list of other events that relate to a similar topic. The related events can help the user get a more complete picture of an event and what potentially led to it.

When identifying related events, you can choose which properties the related events should have in common and set time constraints if necessary.

Similar events can be displayed as a simple event list, on an event timeline, or by showing daily trends. In the event list, the most related events are displayed first. The event timeline, on the other hand, sorts the events by date and allows the user to see the progression of related events.

The daily trends also show the events on the timeline, but here the emphasis is on displaying the intensity of reporting about the event. Each related event is displayed with a different color and the height represents the number of articles that reported about the event on a particular day.

Each event consists of a list of news articles that report about the event. The articles are grouped by language. If the event is reported in multiple languages you can switch to another language by clicking the drop-down menu with the language name. You can open an article by clicking its title.

Article sorting

Displayed articles about the event can be sorted in different ways, depending on your needs. Available options are:

Date

The latest articles about the event will be displayed first.

Story centrality

Articles that describe the overall event best will be displayed first.

Source importance

Articles will be sorted from those that come from higher-ranking news sources to lower-ranking ones.

Shares on social media

Articles will be ranked based on how many times they were shared on social media. We analyze how frequently each article is shared on Facebook, Google+, LinkedIn, Pinterest, and Instagram and use the aggregated count as a measure for ranking the articles.