Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -71,13 +71,44 @@ Each document in the dataset is a dictionary with the following fields:
|
|
71 |
|
72 |
------
|
73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
## peS2o V1
|
75 |
|
76 |
### Key Facts
|
77 |
|
78 |
- *Knowledge cutoff*: 2023-01-03
|
79 |
- *Number of documents*: 67.56M
|
80 |
-
- *Number of whitespace-separated tokens*: 47.
|
81 |
|
82 |
### Processing
|
83 |
|
@@ -129,33 +160,7 @@ Unfiltered, the corpus contains 91.1M papers and 15.5B whitespace-separated toke
|
|
129 |
|s2ag | valid | 111,228 | 24,398,512 |
|
130 |
|
131 |
|
132 |
-
------
|
133 |
-
|
134 |
-
## peS2o V2
|
135 |
-
|
136 |
-
|
137 |
-
### Key Facts
|
138 |
-
|
139 |
-
- *Knowledge cutoff*: 2023-01-03
|
140 |
-
- *Number of documents*: 38.97M
|
141 |
-
- *Number of whitespace-separated tokens**: 42.01B
|
142 |
|
143 |
-
### Processing
|
144 |
-
|
145 |
-
peS2o V2 is largely the same as V1, but it includes additional heuristics s2ag aimed at filtering out OCR errors from abstract.
|
146 |
-
|
147 |
-
First, we check if the abstract was obtained from Semantic Scholar sources that are likely to contain OCR'ed content. For any abstract derived from those sources, we count how often the text contains subsequences matching `\b([A-Za-z]\s)([a-z]\s)*[A-Za-z]\b`, i.e. individual alpha letters separated by a space. This heuristic matches cases such as `A b stra ct` (2 matching subsequences), where the OCR parser inserted erroneous spaces.
|
148 |
-
Any abstract with more than 4 matching subsequences is removed.
|
149 |
-
|
150 |
-
|
151 |
-
#### Statistics
|
152 |
-
|
153 |
-
| Dataset | Split | # Documents | # Words |
|
154 |
-
|:-------:|:-----:|------------:|---------------:|
|
155 |
-
| s2orc | train | 8,242,162 | 36,088,195,908 |
|
156 |
-
| s2orc | valid | 51,323 | 255,139,074 |
|
157 |
-
| s2ag | train | 30,569,017 | 5,920,099,207 |
|
158 |
-
| s2ag | valid | 109,709 | 24,029,459 |
|
159 |
|
160 |
[1]: https://aclanthology.org/2020.acl-main.447/
|
161 |
[2]: https://github.com/allenai/s2orc
|
|
|
71 |
|
72 |
------
|
73 |
|
74 |
+
|
75 |
+
## peS2o V2 (Latest)
|
76 |
+
|
77 |
+
|
78 |
+
### Key Facts
|
79 |
+
|
80 |
+
- *Knowledge cutoff*: 2023-01-03
|
81 |
+
- *Number of documents*: 38.97M
|
82 |
+
- *Number of whitespace-separated tokens**: 42.01B
|
83 |
+
|
84 |
+
### Processing
|
85 |
+
|
86 |
+
peS2o V2 is largely the same as V1, but it includes additional heuristics s2ag aimed at filtering out OCR errors from abstract.
|
87 |
+
|
88 |
+
First, we check if the abstract was obtained from Semantic Scholar sources that are likely to contain OCR'ed content. For any abstract derived from those sources, we count how often the text contains subsequences matching `\b([A-Za-z]\s)([a-z]\s)*[A-Za-z]\b`, i.e. individual alpha letters separated by a space. This heuristic matches cases such as `A b stra ct` (2 matching subsequences), where the OCR parser inserted erroneous spaces.
|
89 |
+
Any abstract with more than 4 matching subsequences is removed.
|
90 |
+
|
91 |
+
|
92 |
+
#### Statistics
|
93 |
+
|
94 |
+
| Dataset | Split | # Documents | # Words |
|
95 |
+
|:-------:|:-----:|------------:|---------------:|
|
96 |
+
| s2orc | train | 8,242,162 | 36,088,195,908 |
|
97 |
+
| s2orc | valid | 51,323 | 255,139,074 |
|
98 |
+
| s2ag | train | 30,569,017 | 5,920,099,207 |
|
99 |
+
| s2ag | valid | 109,709 | 24,029,459 |
|
100 |
+
|
101 |
+
|
102 |
+
|
103 |
+
-------
|
104 |
+
|
105 |
## peS2o V1
|
106 |
|
107 |
### Key Facts
|
108 |
|
109 |
- *Knowledge cutoff*: 2023-01-03
|
110 |
- *Number of documents*: 67.56M
|
111 |
+
- *Number of whitespace-separated tokens*: 47.37B
|
112 |
|
113 |
### Processing
|
114 |
|
|
|
160 |
|s2ag | valid | 111,228 | 24,398,512 |
|
161 |
|
162 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
163 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
164 |
|
165 |
[1]: https://aclanthology.org/2020.acl-main.447/
|
166 |
[2]: https://github.com/allenai/s2orc
|