Extracting causal relations from the literature with word vector mapping.

Area of Science:

• Biomedical informatics
• Computational biology
• Natural language processing

Background:

• Causal graphs are crucial for understanding causality in biology and medicine.
• Traditional data-driven methods for causal graph construction often lack accuracy.
• Existing literature-based causality extraction methods face limitations with labeled data or predefined patterns.

Purpose of the Study:

• To develop an advanced causality extraction framework integrating rule-based and unsupervised learning.
• To enhance the accuracy and efficiency of establishing causal graphs from scientific publications.
• To overcome the data dependency of supervised methods and the pattern rigidity of rule-based systems.

Main Methods:

• A three-module framework: data preprocessing, syntactic pattern matching, and causality determination.
• Sentence simplification and part-of-speech tagging for syntactic pattern matching.
• Unsupervised learning with word vectors and verb similarity comparison for causality determination.

Main Results:

• The proposed method demonstrated superior performance compared to existing approaches (Bui's and Alashri's methods).
• Significant improvements were observed in precision, recall, specificity, accuracy, and F-score.
• The F-score increased by 8.29% and 5.37% over the compared methods.

Conclusions:

• The integrated framework effectively extracts causal relations from text, overcoming limitations of prior methods.
• This approach offers a more robust and accurate way to build causal graphs using biomedical literature.
• The method shows promise for advancing causal discovery in complex biological and medical research.