This is my current experiments on truth discovery with dataset population.
Data set population is from https://cogcomp.org/page/resource_view/16, source paper is :Knowing What to Believe (when you already know something), Jeff Pasternack and Dan Roth, COLING - 2010
I implemented popular truth discovery methods CRH, CATD, GTM, TruthFinder, Sums, Average.log, investment, pooled-investment.
Folder src has java codes.
Folder data has preprocessed files.
First, run PreProcessDataNewYear.java, which preprocesses the data like (A Probabilistic Model for Estimating Real-valued Truth from Conflicting Sources). Then, for each algorithm, just run the main function with different parameters. MAE, RMSE, error rate (1%) is reported.
After preprocessing, 4183 claims about 1172 population entities from 2415 sources. 277 city-year pairs appear in evaluation. Compared with "4119 claims about 1148 population entities from 2415 sources. 274 city-year pairs" in (A Probabilistic Model for Estimating Real-valued Truth from Conflicting Sources).
The experiment results:
