Julian McAuley of UCSD compiles a list of datasets for recommender systems. These are a diverse set of labelled text data fortified with additional measures available for various datasets (e.g. network relationships, geospatial data, categorical information, quantitative measures, etc.).
One dataset that has been frequently used in topic modelling and ...
In addition to the answer by @Erwan, I came across
Both are curated lists of text datasets for NLP tasks. A few datasets that may fit your requirement of temporal variation:
Cheng-Caverlee-Lee September 2009 - January 2010 Twitter Scrape: ...