Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -56,12 +56,14 @@ marking 'supporting evidence' for the label, following how the task is defined b
|
|
56 |
all the words, in the sentence, they think shows
|
57 |
evidence for their chosen label.
|
58 |
|
59 |
-
#### Our annotations
|
60 |
negative 1555 |
|
61 |
positive 1435 |
|
62 |
no sentiment 470
|
63 |
Total 3460
|
64 |
|
|
|
|
|
65 |
|
66 |
### SST2
|
67 |
|
@@ -81,12 +83,14 @@ sentiment that they do not see.
|
|
81 |
*The Zuco data contains eye-tracking data for 400 instances from SST. By annotating some of these with rationales,
|
82 |
we add an extra layer of information for future research.
|
83 |
|
84 |
-
#### Our annotations
|
85 |
positive 1027 |
|
86 |
negative 900 |
|
87 |
no sentiment 163
|
88 |
Total 2090
|
89 |
|
|
|
|
|
90 |
|
91 |
### CoS-E
|
92 |
|
@@ -105,9 +109,11 @@ think that removing it will decrease your
|
|
105 |
confidence toward your chosen label,
|
106 |
please mark it.’
|
107 |
|
108 |
-
#### Our annotations
|
109 |
Total 3760
|
110 |
|
|
|
|
|
111 |
|
112 |
### Dataset Sources
|
113 |
|
@@ -116,15 +122,16 @@ Total 3760
|
|
116 |
- **Repository:** https://github.com/terne/Being_Right_for_Whose_Right_Reasons
|
117 |
- **Paper:** [Being Right for Whose Right Reasons?](https://aclanthology.org/2023.acl-long.59/)
|
118 |
|
119 |
-
|
120 |
|
121 |
<!-- Address questions around how the dataset is intended to be used. -->
|
122 |
In our paper, we present a collection of three
|
123 |
existing datasets (SST2, DynaSent and Cos-E) with demographics-augmented annotations to enable profiling of models, i.e., quantifying their alignment (or agreement) with rationales provided
|
124 |
by different socio-demographic groups. Such profiling enables us to ask whose right reasons models are being right for and fosters future research on performance equality/robustness.
|
125 |
|
126 |
-
For each dataset, we provide the data under a unique **'
|
127 |
-
|
|
|
128 |
|
129 |
|
130 |
## Dataset Structure
|
|
|
56 |
all the words, in the sentence, they think shows
|
57 |
evidence for their chosen label.
|
58 |
|
59 |
+
#### >Our annotations:
|
60 |
negative 1555 |
|
61 |
positive 1435 |
|
62 |
no sentiment 470
|
63 |
Total 3460
|
64 |
|
65 |
+
Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
|
66 |
+
|
67 |
|
68 |
### SST2
|
69 |
|
|
|
83 |
*The Zuco data contains eye-tracking data for 400 instances from SST. By annotating some of these with rationales,
|
84 |
we add an extra layer of information for future research.
|
85 |
|
86 |
+
#### >Our annotations:
|
87 |
positive 1027 |
|
88 |
negative 900 |
|
89 |
no sentiment 163
|
90 |
Total 2090
|
91 |
|
92 |
+
Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
|
93 |
+
|
94 |
|
95 |
### CoS-E
|
96 |
|
|
|
109 |
confidence toward your chosen label,
|
110 |
please mark it.’
|
111 |
|
112 |
+
#### >Our annotations:
|
113 |
Total 3760
|
114 |
|
115 |
+
Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
|
116 |
+
|
117 |
|
118 |
### Dataset Sources
|
119 |
|
|
|
122 |
- **Repository:** https://github.com/terne/Being_Right_for_Whose_Right_Reasons
|
123 |
- **Paper:** [Being Right for Whose Right Reasons?](https://aclanthology.org/2023.acl-long.59/)
|
124 |
|
125 |
+
<a id="uses">## Uses</a>
|
126 |
|
127 |
<!-- Address questions around how the dataset is intended to be used. -->
|
128 |
In our paper, we present a collection of three
|
129 |
existing datasets (SST2, DynaSent and Cos-E) with demographics-augmented annotations to enable profiling of models, i.e., quantifying their alignment (or agreement) with rationales provided
|
130 |
by different socio-demographic groups. Such profiling enables us to ask whose right reasons models are being right for and fosters future research on performance equality/robustness.
|
131 |
|
132 |
+
For each dataset, we provide the data under a unique **'train'** split due to the current limitation of not being possible to upload a dataset with a single *'test'* split.
|
133 |
+
Note, however, that the original itended used of these collection of datasets was to **test** the quality & alignment of post-hoc explainability methods.
|
134 |
+
If you use it following different splits, please clarify it to ease reproducibility of your work.
|
135 |
|
136 |
|
137 |
## Dataset Structure
|