LLM-based annotation and token-augmented modeling for emotional tone classification in online cancer peer-support posts.

The collapsed comparison remained highly significant (chi-square = 2645.39, p < 0.001) with a Jensen–Shannon divergence of 0.0351, indicating that while most four-class disagreement was absorbed by the collapse, the LLM still assigned substantially more posts to Negative and fewer to Neutral/Positive than the human labels.

also in 132,142 other papers