I have a large number of texts (each about 1000 words). Every text contains a various number of topics, usually expressed as a sentence. I've extracted and categorize 22 topics and found 5000 of them in 900 texts and now want to build a model, that would learn from this data.
So basically I have a sufficient amount of data for each topic. I was thinking about extracting keywords for each topic or making vector cluster. But how do I search for topics on texts? Should I take every sentence as an atomic piece to compare with the original categories?