Conceptual
Browse Instruct Curate

LLM Embedding Features for Hyperpartisan News Detection

Detecting whether a news article is hyperpartisan (extremely one-sided political content intended to deepen division) by encoding the article as a document embedding produced by a large language model and classifying that vector. Benchmarked against n-gram, sentiment, and ELMo-with-bidirectional-LSTM baselines, the LLM-embedding approach raises detection accuracy to roughly 92 percent.