Chemical Entity Recognition and Resolution to ChEBI

Table 4

Evaluation of entity resolution, subset of the gold standard composed by 9,696 chemical entities that contain a mapping to ChEBI. Results of entity resolution for each assessment and method are shown in this table. Have been considered for this evaluation only the entities successfully recognized by both methods. For an exact matching assessment, the amount of entities successfully recognized by both methods was 3,668. For the left, right, left/right, and partial matching assessments, that amount was correspondingly 4,022, 4,082, 4,455, and 5,286 entities. True Positives (TP) is the amount of those entities for which the resolution was correct, that is, the mapping agrees with the gold standard. Values of precision, recall, and F-measure are presented.

Assessment	Method	TP	Precision	Recall	F-measure
Exact matching	Dictionary	3,079	83.94	31.76	46.08
	Machine learning	3,206	87.40	33.07	47.98
Left matching	Dictionary	3,215	79.94	33.16	46.87
	Machine learning	3,381	84.06	34.87	49.29
Right matching	Dictionary	3,191	78.17	32.91	46.32
	Machine learning	3,467	84.93	35.76	50.33
Left/right matching	Dictionary	3,327	74.68	34.31	47.02
	Machine learning	3,650	81.93	37.64	51.59
Partial matching	Dictionary	3,861	73.04	39.82	51.54
	Machine learning	4,273	80.84	44.07	57.04