Dominant topic in a document

[dineshdisprz]

Is there any way that I can get the dominant topic from the documnets?

[GeorgeDittmar]

When you say dominant topic do you mean the one with the most members or something else?

[MaartenGr]

After training the model, you can access the assigned topics for each document with topic_model.topics_. These are ordered with the input. Technically, these are the dominant topic in each document since each document is assigned to a single document.

However, if you want to model the distribution of topics in the documents, it might be worthwhile to use .approximate_distribution instead. You can find more about that here.

[dineshdisprz]

When you say dominant topic do you mean the one with the most members or something else?

The document is represented by a set of topics. The topic that is most talked about in the document is the "dominant topic".

[dineshdisprz]

After training the model, you can access the assigned topics for each document with topic_model.topics_. These are ordered with the input. Technically, these are the dominant topic in each document since each document is assigned to a single document.

However, if you want to model the distribution of topics in the documents, it might be worthwhile to use .approximate_distribution instead. You can find more about that here.

In the documentation with topic_model.topics_ most recent topics are tracked. Is it the same as the dominant topic? Because when I checked the probabilities of the topics with the documents ( by setting the parameter " calculate_probabilities=True " ). The topic that got max probability differs from the results I get from the topic_model.topics_ .

[MaartenGr]

In the documentation with topic_model.topics_ most recent topics are tracked. Is it the same as the dominant topic?

Yes, there are the dominant topics per document.

Because when I checked the probabilities of the topics with the documents ( by setting the parameter " calculate_probabilities=True " ). The topic that got max probability differs from the results I get from the topic_model.topics_ .

The probabilities that are calculated are approximations as a result of how HDBSCAN generates these probabilities. As a result, they indeed may be different and are used as just that, an approximation. I have had this question a couple of times before, so I'll definitely make sure to make this a bit more clear in the documentation.