Table 9 Comparison with existing disease-disease association extraction methods
From: A study on large-scale disease causality discovery from biomedical literature
Name | Principle | Advantage | Disadvantage | Accuracy |
|---|---|---|---|---|
Method in this study | To achieve an accurate extraction of disease causality by optimizing the disease causality semantic predicate list | Can achieve a more accurate and flexible disease causality extraction | Disease causality extraction is incomplete; the manual extraction of predicates is time-consuming and labour-intensive and the evaluation of predicates is subjective | Achieve an accuracy of 96.97% in disease causality extraction |
Method based on lexical semantics and document-clause frequency | Identify disease causality based on the lexicon-based causality term strength and document and clause frequencies in the literature | Reflect the direction and strength of disease causality | The results of disease causality pairs are limited; only 195 diseases are covered and the results are difficult to verify | Show higher correlation with associated diseases with the Spearman’s rank correlation coefficient of 0.83 |
LC-CNN | A neural network-based approach | Achieve more accurate DDA extraction by combining the hinge loss function of SVM with a convolutional neural network into a single neural network architecture | There are symptom/subclass errors, negation errors and co-occur errors | Achieve a Precision measure of 82.36% |