I have different texts, some of them have been manually labelled some others have not. Texts
“The modelling dataset were rebuilt yesterday” “This movie is awful” “Trump said that ufo exist”
I would need to classify texts as fake or not fake. I have thought to tokenise them first, in order to extract the more significant words, then to use them for predicting further upcoming texts. However, I realised that probably I would need to distinguish texts by topic and extract word frequency distinguishing between the words used for texts labelled as fake and words used for texts labelled as not fake. But my question is: should I label the texts manually? It would be ok if they were 100, but when there are thousands of texts to analyse it would sound impossible. Do you have any example or suggestion on how to approach the problem and which type of analis would be useful to run such analysis?