ACELA

ACELA streamlines the creation of comprehensive, high-quality annotated corpora for named entity (NE) recognition by combining probabilistic NE tagging with an iterative human-in-the-loop annotation process to reduce manual annotation effort.

Key Features:

• Active Learning-Like Framework: Employs an active learning–inspired approach that systematically annotates all occurrences of target named entities within a corpus to avoid sampling bias.
• Iterative and Interactive Process: Facilitates repeated interaction between human annotators and a probabilistic NE tagger to refine annotations over successive rounds.
• Reduction in Human Effort: Simulates annotation and leverages the probabilistic tagger to minimize the number of sentences requiring manual examination, especially for sparse target named entities.
• Probabilistic Tagging: Uses a probabilistic model to predict named entities within text and guide annotator focus.
• Iterative Refinement: Updates the probabilistic model based on human feedback to improve predictive accuracy across iterations.
• Comprehensive Coverage: Ensures annotation of all instances of target entities to eliminate biases associated with selective sampling.

Scientific Applications:

• Biomedical informatics: Enables creation of gold-standard annotated datasets for NE recognition used in biomedical text mining and extraction tasks.
• Genomics: Supports extraction and annotation of genomic entities from large text corpora to improve entity recognition models in genomics research.
• Information extraction from large corpora: Applies to domains requiring detailed entity extraction and high-quality annotated corpora for training NE recognizers.

Methodology:

Probabilistic tagging of text using a probabilistic NE model; iterative refinement of the model based on human annotator feedback; systematic annotation to ensure comprehensive coverage of target entities.