BMC Bioinformatics

Table 2 GeneRIF prediction results

From: GeneRIF indexing: sentence selection based on machine learning

	NB			SVM			ABM1
	p	r	f	p	r	f	p	r	f
pos	0.6052	0.3256	0.4234	0.6052	0.3256	0.4234	0.6594	0.7691	0.7100
posf	0.6705	0.5358	0.5956	0.6798	0.5196	0.5890	0.7218	0.7252	0.7235
text	0.5941	0.6051	0.5995	0.6322	0.6351	0.6336	0.8250	0.0762	0.1395
gene	0.5533	0.6952	0.6162	0.5533	0.6952	0.6162	0.5533	0.6952	0.6162
dis	0.6960	0.8037	0.7460	0.6755	0.8268	0.7435	0.7284	0.6628	0.6941
posf + dis	0.6974	0.8568	0.7689	0.6755	0.8268	0.7435	0.7323	0.7390	0.7356
posf + dis + gene	0.6996	0.8337	0.7608	0.6976	0.8152	0.7519	0.7875	0.7275	0.7563
posf + dis + go	0.6972	0.8614	0.7707	0.6755	0.8268	0.7435	0.7323	0.7390	0.7356
posf + dis + go + text	0.6751	0.7968	0.7309	0.7282	0.6559	0.6902	0.7342	0.7529	0.7434
disg	0.6061	0.9630	0.7440	0.6061	0.9630	0.7440	0.6061	0.9630	0.7440
posf + disg	0.6798	0.7552	0.7155	0.7250	0.7667	0.7452	0.7259	0.7644	0.7447
posf + disg + gene	0.7047	0.7991	0.7489	0.7886	0.7321	0.7593	0.7810	0.6836	0.7291
posf + disg + go	0.6708	0.7575	0.7115	0.7249	0.7667	0.7452	0.7259	0.7644	0.7447
posf + disg + go + text	0.6759	0.7321	0.7029	0.7802	0.6559	0.7127	0.7393	0.7206	0.7298

The results are show for each feature or their combination on the test set after training a Naïve Bayes (NB), Support Vector Machine (SVM) or AdaBoostM1 (ABM1). For each feature or feature combination, the precision (p), recall (r) and F-measure (f) are shown. The individual features are: sentence position from the beginning of the abstract (pos), sentence position from the end of the abstract (posf), text features from the sentence (text), gene mention and normalization (gene), discourse features predicted by the AdaBoostM1 classifier (dis), discourse features predicted by the CRF model (disg) and Gene Ontology score (go).

Back to article page

ISSN: 1471-2105

Contact us

General enquiries: journalsubmissions@springernature.com