TomTBT commited on
Commit
3d6fc00
1 Parent(s): 09c2426

Doc update

Browse files
Files changed (2) hide show
  1. README.md +5 -20
  2. pmc_open_access_xml.py +4 -4
README.md CHANGED
@@ -77,7 +77,7 @@ The PMC Open Access Subset is one part of the PMC Article Datasets
77
 
78
  This version takes XML version as source, benefiting from the structured text
79
  to split the articles in parts, naming the introduction, methods, results,
80
- discussion and conclusion, and refers with keywords in the text to external or internal
81
  resources (articles, figures, tables, formulas, boxed-text, quotes, code, footnotes, chemicals, graphics, medias).
82
 
83
  The dataset was initially created with relation-extraction tasks in mind, between the references in text and the content of the
@@ -112,27 +112,11 @@ a corpus of pre-annotated text for other tasks (e.g. figure caption to graphic,
112
  - "box": List of \<boxed-text\> elements of the article.
113
  - "code": List of \<code\> elements of the article.
114
  - "quote": List of \<disp-quote\> and \<speech\> elements of the article.
115
- - "chem": List of \<chem-struct-wrap\> elements of the article.
116
  - "supplementary": List of \<supplementary-material\> and \<inline-supplementary-material\> elements of the article.
117
  - "footnote": List of \<fn-group\> and \<table-wrap-foot\> elements of the article.
118
  - "graphic": List of \<graphic\> and \<inline-graphic\> elements of the article.
119
  - "media": List of \<media\> and \<inline-media\> elements of the article.
120
-
121
- - "references": Lists of references in the order found in the corresponding categories
122
- - "pmid_ref": PMIDs of other articles. In text, replaced by " ##REF## "
123
- - "unknown_pub_ref": 0-indexed idx of the article in "unknown_pub". In text, replaced by " ##UREF## "
124
- - "figure_ref": 0-indexed idx of the figures in "figure" In text, replaced by " ##FIG## "
125
- - "table_ref": 0-indexed idx of the tables in "table" In text, replaced by " ##TAB## "
126
- - "formula_ref": 0-indexed idx of the formulas in "formula" In text, replaced by " ##FORMU## "
127
- - "box_ref": 0-indexed idx of the box in "box" In text, replaced by " ##BOX## "
128
- - "code_ref": 0-indexed idx of the code in "code" In text, replaced by " ##CODE## "
129
- - "quote_ref": 0-indexed idx of the quote in "quote" In text, replaced by " ##QUOTE## "
130
- - "chem_ref": 0-indexed idx of the chemical in "chem" In text, replaced by " ##CHEM## "
131
- - "supplementary_ref": 0-indexed idx of the figure in "supplementary" In text, replaced by " ##SUPPL## "
132
- - "footnote_ref": 0-indexed idx of the footnote in "footnote" In text, replaced by " ##FOOTN## "
133
- - "graphic_ref": 0-indexed idx of the graphics in "graphics" In text, replaced by " ##GRAPH## "
134
- - "media_ref": 0-indexed idx of the medias in "medias" In text, replaced by " ##MEDIA## "
135
- - "references_text": For each reference in "references", the corresponding text that the reference keywords replaced.
136
  - "glossary": Glossary if found in the XML
137
  - "unknown_references": JSON of a dictionnary of each "tag":"text" for the reference that did not indicate a PMID
138
  - "n_references": Total number of references and unknown references
@@ -142,6 +126,7 @@ a corpus of pre-annotated text for other tasks (e.g. figure caption to graphic,
142
  - "citation": Citation of the article
143
  - "package_file": path to the folder containing the graphics and media files of the article (to append to the base URL: ftp.ncbi.nlm.nih.gov/pub/pmc/)
144
 
 
145
  ### Data Splits
146
 
147
  [Needs More Information]
@@ -204,13 +189,13 @@ future version.
204
 
205
  [Needs More Information]
206
 
207
- ### Preprocessing advices
208
 
209
  - Filter out empty contents.
210
  - Remove unwanted references from the text, and replace either by the "references_text" or by the reference content itself.
211
  - Unescape HTML special characters: `import html; html.unescape(my_text)`
212
  - Remove superfluous line break in text.
213
- - Remove XML tags / replace by special tokens (\<italic\>, \<sup\>, \<sub\>, ...).
214
  - Join the items of the contents' lists.
215
 
216
  ## Additional Information
 
77
 
78
  This version takes XML version as source, benefiting from the structured text
79
  to split the articles in parts, naming the introduction, methods, results,
80
+ discussion and conclusion, and reference with keywords in the text to external or internal
81
  resources (articles, figures, tables, formulas, boxed-text, quotes, code, footnotes, chemicals, graphics, medias).
82
 
83
  The dataset was initially created with relation-extraction tasks in mind, between the references in text and the content of the
 
112
  - "box": List of \<boxed-text\> elements of the article.
113
  - "code": List of \<code\> elements of the article.
114
  - "quote": List of \<disp-quote\> and \<speech\> elements of the article.
115
+ - "chemical": List of \<chem-struct-wrap\> elements of the article.
116
  - "supplementary": List of \<supplementary-material\> and \<inline-supplementary-material\> elements of the article.
117
  - "footnote": List of \<fn-group\> and \<table-wrap-foot\> elements of the article.
118
  - "graphic": List of \<graphic\> and \<inline-graphic\> elements of the article.
119
  - "media": List of \<media\> and \<inline-media\> elements of the article.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
120
  - "glossary": Glossary if found in the XML
121
  - "unknown_references": JSON of a dictionnary of each "tag":"text" for the reference that did not indicate a PMID
122
  - "n_references": Total number of references and unknown references
 
126
  - "citation": Citation of the article
127
  - "package_file": path to the folder containing the graphics and media files of the article (to append to the base URL: ftp.ncbi.nlm.nih.gov/pub/pmc/)
128
 
129
+ In text, the references are in the form ##KEYWORD##IDX_REF##OLD_TEXT##, with keywords (REF, UREF, FIG, TAB, FORMU, BOX, CODE, QUOTE, CHEM, SUPPL, FOOTN, GRAPH, MEDIA) referencing respectively to "pubmed articles" (external), "unknown_references", "figure", "table", "formula", "box", "code", "quote", "chem", "supplementary", "footnote", "graphic" and "media".
130
  ### Data Splits
131
 
132
  [Needs More Information]
 
189
 
190
  [Needs More Information]
191
 
192
+ ### Preprocessing recommendations
193
 
194
  - Filter out empty contents.
195
  - Remove unwanted references from the text, and replace either by the "references_text" or by the reference content itself.
196
  - Unescape HTML special characters: `import html; html.unescape(my_text)`
197
  - Remove superfluous line break in text.
198
+ - Remove XML tags (\<italic\>, \<sup\>, \<sub\>, ...), replace by special tokens?
199
  - Join the items of the contents' lists.
200
 
201
  ## Additional Information
pmc_open_access_xml.py CHANGED
@@ -75,7 +75,7 @@ REFS_KEYS = ["pmid_ref", "unknown_pub_ref", "figure_ref", "table_ref", "form
75
  "quote_ref", "chem_ref", "supplementary_ref", "footnote_ref", "graphic_ref", "media_ref"]
76
  CONTENT_KEYS = ["introduction", "methods", "results", "discussion", "conclusion",
77
  "front", "body", "back", "figure", "table", "formula", "box",
78
- "code", "quote", "chem", "supplementary", "footnote"]
79
  begin_doc_rgx = re.compile("""<!DOCTYPE.*""")
80
  def clean_raw(xml_text):
81
  """
@@ -434,7 +434,7 @@ class OpenAccessXML(datasets.GeneratorBasedBuilder):
434
  "box": datasets.features.Sequence(datasets.Value("string")),
435
  "code": datasets.features.Sequence(datasets.Value("string")),
436
  "quote": datasets.features.Sequence(datasets.Value("string")),
437
- "chem": datasets.features.Sequence(datasets.Value("string")),
438
  "supplementary": datasets.features.Sequence(datasets.Value("string")),
439
  "footnote": datasets.features.Sequence(datasets.Value("string")),
440
  "graphic": datasets.features.Sequence(datasets.Value("string")),
@@ -571,7 +571,7 @@ class OpenAccessXML(datasets.GeneratorBasedBuilder):
571
  "box": content_d["box"],
572
  "code": content_d["code"],
573
  "quote": content_d["quote"],
574
- "chem": content_d["chem"],
575
  "supplementary": content_d["supplementary"],
576
  "footnote": content_d["footnote"],
577
  "graphic": content_d["graphic"],
@@ -633,7 +633,7 @@ class OpenAccessXML(datasets.GeneratorBasedBuilder):
633
  "box": content_d["box"],
634
  "code": content_d["code"],
635
  "quote": content_d["quote"],
636
- "chem": content_d["chem"],
637
  "supplementary": content_d["supplementary"],
638
  "footnote": content_d["footnote"],
639
  "graphic": content_d["graphic"],
 
75
  "quote_ref", "chem_ref", "supplementary_ref", "footnote_ref", "graphic_ref", "media_ref"]
76
  CONTENT_KEYS = ["introduction", "methods", "results", "discussion", "conclusion",
77
  "front", "body", "back", "figure", "table", "formula", "box",
78
+ "code", "quote", "chemical", "supplementary", "footnote"]
79
  begin_doc_rgx = re.compile("""<!DOCTYPE.*""")
80
  def clean_raw(xml_text):
81
  """
 
434
  "box": datasets.features.Sequence(datasets.Value("string")),
435
  "code": datasets.features.Sequence(datasets.Value("string")),
436
  "quote": datasets.features.Sequence(datasets.Value("string")),
437
+ "chemical": datasets.features.Sequence(datasets.Value("string")),
438
  "supplementary": datasets.features.Sequence(datasets.Value("string")),
439
  "footnote": datasets.features.Sequence(datasets.Value("string")),
440
  "graphic": datasets.features.Sequence(datasets.Value("string")),
 
571
  "box": content_d["box"],
572
  "code": content_d["code"],
573
  "quote": content_d["quote"],
574
+ "chemical": content_d["chemical"],
575
  "supplementary": content_d["supplementary"],
576
  "footnote": content_d["footnote"],
577
  "graphic": content_d["graphic"],
 
633
  "box": content_d["box"],
634
  "code": content_d["code"],
635
  "quote": content_d["quote"],
636
+ "chemical": content_d["chemical"],
637
  "supplementary": content_d["supplementary"],
638
  "footnote": content_d["footnote"],
639
  "graphic": content_d["graphic"],