Topics

How is a topic defined in DeepNews?

Topics are implicitly defined by the set (mathematically: cluster) of news items that are similar to each other. Explicit definitions of topics do not exist in DeepNews.

In DeepNews, topics are formed automatically via the similarity of news items. Topic formation in DeepNews is a self-organizing process. You can think of topic formation as analogous to crystals crystallizing. Whenever a new news item is suddenly identified, it forms a new topic with subsequent similar news items.

How many topics does DeepNews know?

The amount of topics is unlimited. Every time something new is covered, and possibly more news items match it, a new topic defines itself (see above self-organization / crystallization).

Some numbers:

Are topics fixed once formed?

The topics that form in DeepNews are neither rigid nor predefined. With each new article added, a topic 'moves' or 'shifts' a little bit thematically.

How is the relevance of a topic determined?

The size and thus the relevance of a topic is determined on the one hand by the amount of assigned news items and on the other hand by their temporal weighting. For example, a topic with 5 articles published in the last hour has a much higher relevance than a topic with 20 articles, but all of them are one or two days old.

How is the article representing a topic determined?

The article representing a topic is redetermined at short intervals as time progresses. This incorporates similarity to all other articles in the topic, as well as the novelty and quality of the article.

Can topics overlap?

Topics can overlap in content, but at any given time an article belongs to either one topic or another, depending on the similarity in content. That doesn't mean that individual news items in one topic can't have broad similarities to other topics.