Development and use of a gold-standard data set for subjectivity classifications

This paper presents a case study of analyzing and improving intercoder reliability in discourse tagging using statistical techniques. Bias-corrected tags are formulated and successfully used to guide a revision of the coding manual and develop an automatic classifier.