This article introduces NewsTSC, a manually annotated dataset to explore TSC on news articles . We find that sentiment in the news is expressed lessexplicitly, is more dependent on context and readership . Reasons include incorrectly resolved relation of target and sentiment-bearing phrases and off-context dependence . As a majorimprovement over previous news TSC, we find that BERT’s natural languageunderstanding capabilities capture the less explicit sentiment used in news articles as a major improvement in the state ofthe art in TSC performs worse than on other domains (averagerecall AvgRec = 69.8 on News TSC compared to [75.6, 82.2] onestablished TSC datasets). Reasons for TSC are cited include incorrectly resolving relation oftarget and . sentiment- bearing phrases and . off- context dependence and off . context dependence .

Author(s) : Felix Hamborg, Karsten Donnay, Bela Gipp

Links : PDF - Abstract

Code :
Coursera

Keywords : tsc - news - sentiment - context - articles -

Leave a Reply

Your email address will not be published. Required fields are marked *