These files are supplementary online material for the AAAI 2010 AI and the Web Track paper "Toward an Architecture for Never-Ending Language Learning" by Carlson, Betteridge, Kisiel, Settles, Hruschka, and Mitchell. This work is part of the Read the Web research project at Carnegie Mellon University.
This tab-separated text file lists all instances promoted by NELL during the 67-day run described in the paper:
These tables illustrate the promotion activity over time for all predicates:
These files contain the promoted instances that were sampled and labeled to estimate the accuracy of the overall pool of promoted instances as well as the accuracy of the promotions for specific predicates. There are separate files for iterations 1-22 and 23-66 and they are in slightly different file formats (the latter evaluation was done for the camera-ready paper in a slightly different way, thus the different file formats).
These spreadsheets contain the seed ontology information about categories and relations, including the seed instances for each predicate.
This is a web-browsable XML version of the knwoledge base learned by nell after its 67-day run described in the paper. For clairty, we have removed candidate relations that do not meet the domain and range type constraints necessary to become beliefs.