Data cleaning is a hugely important part of data science, but it can be hard to find "good" messy datasets to practice your cleaning skills. This site includes datasets that need clearning/organizing/reformatting to be most useful, along with a brief overview of what needs to be fixed in each dataset.
A collection of datasets and data generators used by the machine learning community. Currently has 452 datasets, searchable by data type, task of interest, domain area, and other attributes.
Contains nearly 6 million Yelp reviews, plus photos and business attributes. Check out the suggested challenges for photo classification and natural language processing and sentiment analysis.