Estimation of the quality based on absence/presence of words in the TagParser lexicon.

year	nb of papers from the metadata	nb of papers in PDF	nb of papers in XML (= output of PDFBox)	nb of non empty papers as extraction result	nb of papers with an abstract (from extraction)	nb of papers with references (from extraction)	nb of unknown words	nb of known words	nb of words of the content	evaluation of noise = pourcentage of nb of known words / nb of words of the content	evaluation of silence = pourcentage of non empty papers as extraction result / PDF docs	combined evaluation of noise and silence	nb of English papers	nb of French papers	nb of papers in another language (es+de+ru)
2009	103	103	100	100	100	97	7221	449880	457101	98.420	97.087	97.749	100	0	0
2011	139	139	139	139	138	139	10800	594389	605189	98.215	100.000	99.100	139	0	0
2013	121	121	121	121	119	121	9960	534494	544454	98.171	100.000	99.077	121	0	0
total	363	363	360	360	357	357	27981	1578763	1606744	98.259	99.174	98.714	360	0	0

Note#1: the unknowns with initiale lower-case letter denote some full-size unknown words but often words which have been cut due to bad interpreation of PDF multi-columns by PDFBox.

Note#2: a paper without any content (i.e. without any body) is not taken in the pipeline. This situation may be the consequence of a processing problem or may be it is an invited presentation without any text.In constrast, a paper without abstract is taken.

Note#3: a paper without any content holds an entry in the metadata, and normally, each paper has an entry in the metadata.

Note#4: the combined evaluation is computed as: 2*EvalNoise*EvalSilence / EvalNoise+EvalSilence

total elapsed time (read and display included)= 0.8119833333333334 minutes with 8 cores

computed on: Sun Sep 25 19:33:54 CEST 2016 from: METADONNEES_RANLP_140429.txt