Extractive CNN-BiLSTM Summarizer Prevents Hallucinations in Clinical Text

A hybrid 1D-CNN-BiLSTM architecture performs extractive summarization for biomedical documents to eliminate generative hallucinations. The model uses multi-kernel convolutions and bidirectional LSTMs to score sentence importance, ensuring all summary outputs are direct copies from the source text.