Ralston commited on
Commit
b1d9c2a
·
verified ·
1 Parent(s): 6d0e5a1

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +74 -3
README.md CHANGED
@@ -1,3 +1,74 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ This file will introduce the format of data and ontology in SciEvents.
2
+
3
+ About the data `train.json`, `dev.json`, and `test.json`:
4
+
5
+ ```
6
+ # A single document
7
+ {
8
+ "venue": ACL, # all ACL in current SciEvents
9
+ "title": str, # paper title
10
+ "abstract": str, # paper original abstract
11
+ "doc_id": str,
12
+ "publication_year": int, # 2019-2022 in current SciEvents
13
+ "sentences": [str, ...], # split from abstract
14
+ "events": SciEvents's format events, # will be introduced following
15
+ "document": [str, ...], # token-level presentation of the abstract. it can be used to locate offset
16
+ }
17
+ ```
18
+
19
+ ```
20
+ # SciEvents's format events
21
+ [
22
+ { # a certain event
23
+ "event_type": str, # constrained by ontology
24
+ "arguments": [
25
+ { # a certain argument
26
+ "text": str,
27
+ "nugget_type": str, # constrained by ontology, not necessary
28
+ "argument_type": str, # constrained by ontology
29
+ "tokens": [str, ...] # include every single token. argument["text"].split(' ')
30
+ "offsets": [int, ...] # the same length as argument["tokens"]
31
+ },
32
+ ...
33
+ ],
34
+ "trigger": {
35
+ "text": str,
36
+ "tokens": [str, ...], # include every single token. trigger["text"].split(' ')
37
+ "offsets": [int, ...], # the same length as trigger["tokens"]
38
+ }
39
+ },
40
+ ...
41
+ ]
42
+ ```
43
+
44
+ About the ontology `ontology.json`:
45
+
46
+ ```
47
+ {
48
+ "nugget_types": [str, ...], # In SciEvents, we add nugget types for all arguments (see constrains following). Those nugget types are not necessary for Event Extraction (EE) task. They can be used to enhance EE performance or other NLP tasks.
49
+ "event_types": {
50
+ "a certain event type": {
51
+ "argument_1": contrained nugget types,
52
+ "argument_2": contrained nugget types,
53
+ ...
54
+ },
55
+ ...
56
+ }
57
+ }
58
+ ```
59
+
60
+ If you want to use EXCEEDS & SciEvents formats for your own datasets, nugget types are not necessary. Therefore, for compatibility purposes, your customized `ontology.json` can be like:
61
+
62
+ ```
63
+ {
64
+ "event_types": {
65
+ "a certain event type": {
66
+ "argument_1": [], # empty list is ok
67
+ "argument_2": [],
68
+ ...
69
+ },
70
+ ...
71
+ }
72
+ }
73
+ ```
74
+