ISAAC
An open, modular English corpus released by the University of Illinois, comprising over 527 million Reddit posts. Organized around six social group dimensions including race, sexual orientation, and age, it provides multi-dimensional semantic annotations covering moralization, sentiment, and emotion, supporting cross-category, temporal, and spatial studies of social discourse.
Timeline (last 90 days)
ISAAC (Illinois Social Attitudes Aggregate Corpus) was introduced by a team at the University of Illinois.
ISAAC is an open, modular, and easily accessible English corpus containing more than 527 million Reddit posts.
The temporal span of the ISAAC corpus covers 17 years from 2007 to 2023.
ISAAC filters content along six major social group dimensions: race, sexual orientation, age, ability (disability), weight, and skin color.
ISAAC employs a multi-step, human-reviewed filtering process that keeps irrelevant content below 10% both overall and in each social group subset.
Each post in ISAAC is algorithmically annotated with the user's estimated geographic region and accompanied by validated semantic labels for moralization, sentiment, emotion, and linguistic generalization.
The ISAAC team developed a custom model for linguistic generalization to enable large-scale automatic identification.
The research team confirmed ISAAC's validity through convergent evidence by comparing the corpus signals against macro-level indicators such as web search behavior, temporal peaks of major social events, and long-term shifts in public attitudes.
ISAAC offers three research capabilities: cross-category comparison, long-term temporal tracking (17 years of continuous data), and spatial variation mapping.
ISAAC provides multiple access methods: a point-and-click website, an annotator web application, a SQL playground, a Python toolkit, and datasets on HuggingFace.
3 more timeline events
All Facts (13)
By fixing data sources and pipelines and making the full pipeline public, ISAAC provides NLP and computational social science with a shared benchmark analogous to the role ImageNet once played in computer vision.
50%UnverifiedISAAC is an open, modular, and easily accessible English corpus containing more than 527 million Reddit posts.
50%UnverifiedThe temporal span of the ISAAC corpus covers 17 years from 2007 to 2023.
50%UnverifiedISAAC filters content along six major social group dimensions: race, sexual orientation, age, ability (disability), weight, and skin color.
50%UnverifiedISAAC employs a multi-step, human-reviewed filtering process that keeps irrelevant content below 10% both overall and in each social group subset.
50%UnverifiedEach post in ISAAC is algorithmically annotated with the user's estimated geographic region and accompanied by validated semantic labels for moralization, sentiment, emotion, and linguistic generalization.
50%UnverifiedThe ISAAC team developed a custom model for linguistic generalization to enable large-scale automatic identification.
50%UnverifiedThe research team confirmed ISAAC's validity through convergent evidence by comparing the corpus signals against macro-level indicators such as web search behavior, temporal peaks of major social events, and long-term shifts in public attitudes.
50%UnverifiedISAAC offers three research capabilities: cross-category comparison, long-term temporal tracking (17 years of continuous data), and spatial variation mapping.
50%UnverifiedISAAC provides multiple access methods: a point-and-click website, an annotator web application, a SQL playground, a Python toolkit, and datasets on HuggingFace.
50%UnverifiedISAAC's fully open, modular pipeline supports expansion to new platforms, new languages, and new social categories.
50%UnverifiedISAAC (Illinois Social Attitudes Aggregate Corpus) was introduced by a team at the University of Illinois.
50%UnverifiedThe ISAAC paper was published on arXiv.
50%