Question

A media company wants to automatically tag news articles. Each article can belong to several topics simultaneously (for example, an article may be about both 'Politics' and 'Economy'), and the number of applicable topics varies per article. An ML engineer must choose a modeling approach that lets a single article receive multiple independent topic assignments. Which approach best fits this business problem?

Accepted Answer

Train a multi-label classifier that outputs an independent sigmoid probability per topic, applying a threshold to each label separately.

Answer

Train a multi-class classifier with softmax output so the model returns the single most probable topic per article.

Answer

Train a binary classifier that predicts whether an article is 'newsworthy' before assigning any topics.

Answer

Train a regression model that outputs a continuous topic-count value and round it to assign topics.