Context Simple classification task using the titles of post and the labels are that of the corresponding subreddits it is taken from. This was done as a course project at USC. The complete report can be found here: https://github.com/TheShadow29/subreddit-classification-dataset/blob/master/Subreddit_Classifier_Report.pdf There is also code to reproduce TF-IDF and ULMFiT here: Content CSV files with labels and text. I also created two categories one with 17 popular subreddits and the other one with 1416 subreddits, so the latter is much more difficult. Inspiration The task with 1416 is quite interesting, it is basically a form of fine-grained classification. Use Cases: A model good at this can perhaps be used to recommed similar subreddits and possible subreddits a user would want to post to.