README File content

This README file describes the COMPARATIVE commonsense database: whose construction is explained in the AAAI 2014 paper: comparative.pdf , comparative-slides.pdf

COMPARATIVE contains commonsense knowledge of comparing two concepts with respect to different attributes/ relations.
The compressed data (60MB) can be downloaded at The compressed file inflates into a 117MB file.


This file contains several comparative relations in the following format:

  Column   |       Type        | Modifiers
 x         | character varying | ambiguous arg1 e.g. car
 r         | character varying | ambiguous attribute e.g. fast
 y         | character varying | ambiguous arg2 e.g. bike
 awps      | character varying | disambiguated attribute e.g. fast#a#1
 xwps      | character varying | disambiguated arg1 e.g. car#n#1
 rnorm     | character varying | normalized comparing relation 
                                 e.g. be faster than
 ywps      | character varying | disambiguated arg2 e.g. plant#n#2
 direction | character varying | fwd or backward direction e.g. 
                                 car faster bike is forward whereas bike slower car is backward direction 
 freq      | real              | corpus frequency of the triple.
 normid    | character varying | id of the direction-noramlized triple
                                 e.g. car fast bike; bike slow car -- 12345
 observid  | character varying | id of the observation  
                                 e.g. car faster bike 1234 
 sources   | character varying | list of the sources 
                                 e.g. ngram, wordnet

The details of these synset-ids can be looked up at: noun.gloss.txt