Chemical name extraction based on automatic training data generation and rich feature set.

Area of Science:

• Biomedical Informatics
• Computational Chemistry
• Natural Language Processing

Background:

• Automating chemical name extraction is crucial for biomedical research but hindered by the lack of quality training data and reliance on domain expertise.
• Developing reliable entity extraction models requires extensive, well-annotated datasets and deep knowledge of chemical nomenclature.

Purpose of the Study:

• To develop a method for automatically generating realistic training data for chemical name extraction using random text generation techniques.
• To propose novel, domain-independent features for chemical name identification.
• To evaluate the effectiveness of the proposed approach against state-of-the-art methods.

Main Methods:

• Utilized random text generation, starting with an incomplete chemical name dictionary, to create synthetic training documents.
• Statistically analyzed chemical name structures to derive informative features without prior chemistry knowledge.
• Trained and evaluated entity extraction models on real-world data using both generated and manually labeled datasets.

Main Results:

• The proposed method achieved comparable or superior performance to state-of-the-art models trained on manually labeled data.
• The approach significantly reduced the human effort required for data preparation.
• Demonstrated that both structural and semantic aspects of chemical names follow a Zipfian distribution, similar to natural languages.

Conclusions:

• Automated training data generation is a viable and efficient strategy for improving chemical name extraction.
• The developed feature engineering approach effectively captures chemical name characteristics without domain expertise.
• The findings suggest underlying linguistic principles govern chemical nomenclature, offering avenues for future research.