How to programmatically determine if a phrase belongs to a certain topic?

Question

Machine learning

How to determine if a phrase belongs to a certain topic?
For example, the phrase "painting the walls" refers to the topic of construction, but "magnetic storm in Moscow" does not.
What algorithms are used in such tasks? What to dig? Machine learning? Neural networks? Maybe sooner!?

Answer 1

Purely as an option - to analyze the issue in Yandex.Catalog.
Which headings are more - the phrase refers to that.
https://yandex.ru/yaca/?text=%D0%BF%D0%BE%D0%BA%D1...

Answer 2

In any case, this is segmentation, and segmentation is a tree with a search for occurrences of chains or NS.
If it's quite simple and "on-the-fingers": take a category tree and tags from any site and make a segmentation tree: category -> tag-> weight.

Answer 3

https://github.com/loguntsov/bayes - train and classify :)
You need to make as many training pairs as possible, but then it's a matter of technique.

Answer 4

Topic Modeling (topic modeling) is an interesting and math-rich topic. It is better to get to know each other through the video of K.V. Vorontsova, for example, is up to date with Yandex and MIPT on Coursera.