From: Automatic RadLex coding of Chinese structured radiology reports based on text similarity ensemble
Method | Precision (%) | Recall (%) | F1-score (%) |
---|---|---|---|
Levenshtein distance | 70.91 | 50.93 | 59.28 |
Jaccard | 56.64 | 53.28 | 58.05 |
Word2vec CBOW | 81.75 | 81.16 | 84.18 |
WordNet Wup | 86.78 | 84.59 | 85.67 |
Levenshtein distance + Jaccard | 66.90 | 50.85 | 57.78 |
Levenshtein distance + Word2vec CBOW | 72.61 | 64.18 | 68.14 |
Levenshtein distance + WordNet Wup | 78.52 | 67.38 | 72.52 |
Jaccard + Word2vec CBOW | 68.29 | 66.47 | 67.37 |
Jaccard + WordNet Wup | 70.22 | 67.72 | 68.95 |
Word2vec CBOW + WordNet Wup | 83.95 | 82.41 | 83.17 |
Levenshtein distance + Jaccard + Word2vec CBOW | 71.81 | 62.93 | 67.08 |
Levenshtein distance + Jaccard + WordNet Wup | 73.17 | 71.43 | 72.29 |
Levenshtein distance + Word2vec CBOW + WordNet Wup | 84.21 | 77.49 | 80.71 |
Jaccard + Word2vec CBOW + WordNet Wup | 72.83 | 73.92 | 73.37 |
MLP weighing | 91.78 | 88.59 | 90.15 |