Benchmaking biorxiv

Header metadata

Evaluation on 2000 random PDF files out of 1998 PDF (ratio 1.0).

Strict Matching (exact matches)

Field-level results

label	precision	recall	f1	support
authors	84.48	83.59	84.03	1999
first_author	96.41	95.49	95.95	1997
title	77.18	75.95	76.56	2000
all fields (micro avg.)	86.04	85.01	85.52	5996
all fields (macro avg.)	86.02	85.01	85.52	5996

Soft Matching (ignoring punctuation, case and space characters mismatches)

Field-level results

label	precision	recall	f1	support
authors	84.93	84.04	84.49	1999
first_author	96.66	95.74	96.2	1997
title	79.37	78.1	78.73	2000
all fields (micro avg.)	87	85.96	86.48	5996
all fields (macro avg.)	86.99	85.96	86.47	5996

Levenshtein Matching (Minimum Levenshtein distance at 0.8)

Field-level results

label	precision	recall	f1	support
authors	92.21	91.25	91.73	1999
first_author	96.92	95.99	96.45	1997
title	91.77	90.3	91.03	2000
all fields (micro avg.)	93.64	92.51	93.07	5996
all fields (macro avg.)	93.63	92.51	93.07	5996

Ratcliff/Obershelp Matching (Minimum Ratcliff/Obershelp similarity at 0.95)

Field-level results

label	precision	recall	f1	support
authors	88.17	87.24	87.7	1999
first_author	96.41	95.49	95.95	1997
title	87.6	86.2	86.9	2000
all fields (micro avg.)	90.73	89.64	90.18	5996
all fields (macro avg.)	90.73	89.65	90.18	5996

Instance-level results

Total expected instances:   2000
Total correct instances:    1346 (strict) 
Total correct instances:    1381 (soft) 
Total correct instances:    1701 (Levenshtein) 
Total correct instances:    1570 (ObservedRatcliffObershelp) 

Instance-level recall:  67.3    (strict) 
Instance-level recall:  69.05   (soft) 
Instance-level recall:  85.05   (Levenshtein) 
Instance-level recall:  78.5    (RatcliffObershelp) 

Evaluation metrics produced in 12.073 seconds