MIT Press

Skip to Main Content

Table 1:

The data sets, the split sizes (train, dev, test), and the available categories and their properties. Numbers inside parentheses are the number of unique category labels.

Data Set	Splits	Categories	Properties
Yelp 2013	62,522 / 7,773 / 8,671	• users (1.6k) • products (1.6k)	Categories can be sparse (i.e., there may not be enough reviews for each user/product).
AAPR	33,464 / 2,000 / 2,000	• author (48k) • research area (144)	Authors are sparse and have many category labels. Categories can have multiple labels (e.g., multiple authors, multidisciplinary fields).
PolMed	4,500 / 0 / 500	• politician (505) • media source (2) • audience (2) • political bias (2)	The data set has more categories. Categories with binary labels may not be diverse enough to be useful.

Data Set	Splits	Categories	Properties
Yelp 2013	62,522 / 7,773 / 8,671	• users (1.6k) • products (1.6k)	Categories can be sparse (i.e., there may not be enough reviews for each user/product).
AAPR	33,464 / 2,000 / 2,000	• author (48k) • research area (144)	Authors are sparse and have many category labels. Categories can have multiple labels (e.g., multiple authors, multidisciplinary fields).
PolMed	4,500 / 0 / 500	• politician (505) • media source (2) • audience (2) • political bias (2)	The data set has more categories. Categories with binary labels may not be diverse enough to be useful.