Most sentiment analysis tools miss nuance—they can't detect that a single message contains both praise for one group and criticism for another. This work enables fine-grained tracking of who is being helped, harmed, supported, or opposed in online discourse.
This paper introduces a new method to detect mixed positive and negative sentiments directed at different targets within the same message. Instead of labeling text as simply positive or negative, the approach identifies specific targets (like people or groups) and scores them across three dimensions: advocacy vs. opposition, aid vs. harm, and support vs. victimization.