Update files from the datasets library (from 1.16.0)
Browse filesRelease notes: https://github.com/huggingface/datasets/releases/tag/1.16.0
- README.md +93 -33
- dataset_infos.json +1 -1
- irc_disentangle.py +25 -15
README.md
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
annotations_creators:
|
3 |
- expert-generated
|
4 |
language_creators:
|
5 |
-
-
|
6 |
languages:
|
7 |
- en
|
8 |
licenses:
|
@@ -10,18 +10,18 @@ licenses:
|
|
10 |
multilinguality:
|
11 |
- monolingual
|
12 |
size_categories:
|
13 |
-
-
|
14 |
source_datasets:
|
15 |
- original
|
16 |
task_categories:
|
17 |
-
- sequence-modeling
|
18 |
- structure-prediction
|
19 |
task_ids:
|
20 |
-
-
|
21 |
-
- dialogue-modeling
|
22 |
paperswithcode_id: irc-disentanglement
|
|
|
23 |
---
|
24 |
|
|
|
25 |
# Dataset Card for IRC Disentanglement
|
26 |
|
27 |
## Table of Contents
|
@@ -47,12 +47,13 @@ paperswithcode_id: irc-disentanglement
|
|
47 |
- [Licensing Information](#licensing-information)
|
48 |
- [Citation Information](#citation-information)
|
49 |
- [Contributions](#contributions)
|
|
|
50 |
|
51 |
## Dataset Description
|
52 |
|
53 |
- **Homepage:** https://jkk.name/irc-disentanglement/
|
54 |
- **Repository:** https://github.com/jkkummerfeld/irc-disentanglement/tree/master/data
|
55 |
-
- **Paper:** https://
|
56 |
- **Leaderboard:** NA
|
57 |
- **Point of Contact:** jkummerf@umich.edu
|
58 |
|
@@ -60,9 +61,16 @@ paperswithcode_id: irc-disentanglement
|
|
60 |
|
61 |
Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. This new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. The dataset is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context.
|
62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
### Supported Tasks and Leaderboards
|
64 |
|
65 |
-
Conversational Disentanglement
|
66 |
|
67 |
### Languages
|
68 |
|
@@ -130,59 +138,92 @@ data["train"][50]
|
|
130 |
|
131 |
### Data Splits
|
132 |
|
133 |
-
ubuntu: This data split is a new dataset introduced by the authors which labels connected messages in an online chatroom about ubuntu.
|
134 |
|
135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
## Dataset Creation
|
138 |
|
139 |
### Curation Rationale
|
140 |
|
141 |
-
|
|
|
|
|
|
|
|
|
|
|
142 |
|
143 |
### Source Data
|
144 |
|
145 |
#### Initial Data Collection and Normalization
|
146 |
|
147 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
|
149 |
#### Who are the source language producers?
|
150 |
|
151 |
-
|
152 |
|
153 |
### Annotations
|
154 |
|
155 |
#### Annotation process
|
156 |
|
157 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
158 |
|
159 |
#### Who are the annotators?
|
160 |
|
161 |
-
|
|
|
162 |
|
163 |
### Personal and Sensitive Information
|
164 |
|
165 |
-
|
|
|
166 |
|
167 |
## Considerations for Using the Data
|
168 |
|
169 |
### Social Impact of Dataset
|
170 |
|
171 |
-
|
172 |
|
173 |
### Discussion of Biases
|
174 |
|
175 |
-
|
|
|
176 |
|
177 |
### Other Known Limitations
|
178 |
|
179 |
-
|
|
|
180 |
|
181 |
## Additional Information
|
182 |
|
183 |
### Dataset Curators
|
184 |
|
185 |
-
Jonathan K. Kummerfeld
|
186 |
|
187 |
### Licensing Information
|
188 |
|
@@ -190,21 +231,40 @@ Creative Commons Attribution 4.0
|
|
190 |
|
191 |
### Citation Information
|
192 |
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
206 |
}
|
|
|
207 |
|
208 |
### Contributions
|
209 |
|
210 |
-
Thanks to [@dhruvjoshi1998](https://github.com/dhruvjoshi1998) for adding this dataset.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
annotations_creators:
|
3 |
- expert-generated
|
4 |
language_creators:
|
5 |
+
- found
|
6 |
languages:
|
7 |
- en
|
8 |
licenses:
|
|
|
10 |
multilinguality:
|
11 |
- monolingual
|
12 |
size_categories:
|
13 |
+
- 10K<n<100K
|
14 |
source_datasets:
|
15 |
- original
|
16 |
task_categories:
|
|
|
17 |
- structure-prediction
|
18 |
task_ids:
|
19 |
+
- structure-prediction-other-conversation-disentanglement
|
|
|
20 |
paperswithcode_id: irc-disentanglement
|
21 |
+
pretty_name: IRC Disentanglement
|
22 |
---
|
23 |
|
24 |
+
|
25 |
# Dataset Card for IRC Disentanglement
|
26 |
|
27 |
## Table of Contents
|
|
|
47 |
- [Licensing Information](#licensing-information)
|
48 |
- [Citation Information](#citation-information)
|
49 |
- [Contributions](#contributions)
|
50 |
+
- [Acknowledgments](#acknowledgments)
|
51 |
|
52 |
## Dataset Description
|
53 |
|
54 |
- **Homepage:** https://jkk.name/irc-disentanglement/
|
55 |
- **Repository:** https://github.com/jkkummerfeld/irc-disentanglement/tree/master/data
|
56 |
+
- **Paper:** https://aclanthology.org/P19-1374/
|
57 |
- **Leaderboard:** NA
|
58 |
- **Point of Contact:** jkummerf@umich.edu
|
59 |
|
|
|
61 |
|
62 |
Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. This new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. The dataset is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context.
|
63 |
|
64 |
+
Note, the Github repository for the dataset also contains several useful tools for:
|
65 |
+
|
66 |
+
- Conversion (e.g. extracting conversations from graphs)
|
67 |
+
- Evaluation
|
68 |
+
- Preprocessing
|
69 |
+
- Word embeddings trained on the full Ubuntu logs in 2018
|
70 |
+
|
71 |
### Supported Tasks and Leaderboards
|
72 |
|
73 |
+
Conversational Disentanglement
|
74 |
|
75 |
### Languages
|
76 |
|
|
|
138 |
|
139 |
### Data Splits
|
140 |
|
|
|
141 |
|
142 |
+
The dataset has 4 parts:
|
143 |
+
|
144 |
+
| Part | Number of Annotated Messages |
|
145 |
+
| ------------- | ------------------------------------------- |
|
146 |
+
| Train | 67,463 |
|
147 |
+
| Dev | 2,500 |
|
148 |
+
| Test | 5,000 |
|
149 |
+
| Channel 2 | 2,600 |
|
150 |
+
|
151 |
|
152 |
## Dataset Creation
|
153 |
|
154 |
### Curation Rationale
|
155 |
|
156 |
+
IRC is a synchronous chat setting with a long history of use.
|
157 |
+
Several channels log all messages and make them publicly available.
|
158 |
+
The Ubuntu channel is particularly heavily used and has been the subject of several academic studies.
|
159 |
+
|
160 |
+
Data was selected from the channel in order to capture the diversity of situations in the channel (e.g. when there are many users or very few users).
|
161 |
+
For full details, see the [annotation information page](https://github.com/jkkummerfeld/irc-disentanglement/blob/master/data/READ.history.md).
|
162 |
|
163 |
### Source Data
|
164 |
|
165 |
#### Initial Data Collection and Normalization
|
166 |
|
167 |
+
Data was collected from the Ubuntu IRC channel logs, which are publicly available at [https://irclogs.ubuntu.com/](https://irclogs.ubuntu.com/).
|
168 |
+
The raw files are included, as well as two other versions:
|
169 |
+
|
170 |
+
- ASCII, converted using the script [make_txt.py](https://github.com/jkkummerfeld/irc-disentanglement/blob/master/tools/preprocessing/make-txt.py)
|
171 |
+
- Tok, tokenised text with rare words replaced by UNK using the script [dstc8-tokenise.py](https://github.com/jkkummerfeld/irc-disentanglement/blob/master/tools/preprocessing/dstc8-tokenise.py)
|
172 |
+
|
173 |
+
The raw channel two data is from prior work [(Elsner and Charniak, 2008)](https://www.aclweb.org/anthology/P08-1095.pdf)].
|
174 |
|
175 |
#### Who are the source language producers?
|
176 |
|
177 |
+
The text is from a large group of internet users asking questions and providing answers related to Ubuntu.
|
178 |
|
179 |
### Annotations
|
180 |
|
181 |
#### Annotation process
|
182 |
|
183 |
+
The data is expert annotated with:
|
184 |
+
|
185 |
+
- Training, one annotation per line in general, a small portion is double-annotated and adjudicated
|
186 |
+
- Dev, Channel 2, double annotated and adjudicated
|
187 |
+
- Test, triple annotated and adjudicated
|
188 |
+
|
189 |
+
| Part | Annotators | Adjudication? |
|
190 |
+
| ------------- | --------------- | ------------------------------------- |
|
191 |
+
| Train | 1 or 2 per file | For files with 2 annotators (only 10) |
|
192 |
+
| Dev | 2 | Yes |
|
193 |
+
| Test | 3 | Yes |
|
194 |
+
| Channel 2 | 2 | Yes |
|
195 |
|
196 |
#### Who are the annotators?
|
197 |
|
198 |
+
Students and a postdoc at the University of Michigan.
|
199 |
+
Everyone involved went through a training process with feedback to learn the annotation guidelines.
|
200 |
|
201 |
### Personal and Sensitive Information
|
202 |
|
203 |
+
No content is removed or obfuscated.
|
204 |
+
There is probably personal information in the dataset from users.
|
205 |
|
206 |
## Considerations for Using the Data
|
207 |
|
208 |
### Social Impact of Dataset
|
209 |
|
210 |
+
The raw data is already available online and the annotations do not significantly provide additional information that could have a direct social impact.
|
211 |
|
212 |
### Discussion of Biases
|
213 |
|
214 |
+
The data is mainly from a single technical domain (Ubuntu tech support) that probably has a demographic skew of some sort.
|
215 |
+
Given that users are only identified by their self-selected usernames, it is difficult to know more about the authors.
|
216 |
|
217 |
### Other Known Limitations
|
218 |
|
219 |
+
Being focused on a single language and a single channel means that the data is likely capturing a particular set of conventions in communication.
|
220 |
+
Those conventions may not apply to other channels, or beyond IRC.
|
221 |
|
222 |
## Additional Information
|
223 |
|
224 |
### Dataset Curators
|
225 |
|
226 |
+
Jonathan K. Kummerfeld
|
227 |
|
228 |
### Licensing Information
|
229 |
|
|
|
231 |
|
232 |
### Citation Information
|
233 |
|
234 |
+
```
|
235 |
+
@inproceedings{kummerfeld-etal-2019-large,
|
236 |
+
title = "A Large-Scale Corpus for Conversation Disentanglement",
|
237 |
+
author = "Kummerfeld, Jonathan K. and
|
238 |
+
Gouravajhala, Sai R. and
|
239 |
+
Peper, Joseph J. and
|
240 |
+
Athreya, Vignesh and
|
241 |
+
Gunasekara, Chulaka and
|
242 |
+
Ganhotra, Jatin and
|
243 |
+
Patel, Siva Sankalp and
|
244 |
+
Polymenakos, Lazaros C and
|
245 |
+
Lasecki, Walter",
|
246 |
+
booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
|
247 |
+
month = jul,
|
248 |
+
year = "2019",
|
249 |
+
address = "Florence, Italy",
|
250 |
+
publisher = "Association for Computational Linguistics",
|
251 |
+
url = "https://aclanthology.org/P19-1374",
|
252 |
+
doi = "10.18653/v1/P19-1374",
|
253 |
+
pages = "3846--3856",
|
254 |
+
arxiv = "https://arxiv.org/abs/1810.11118",
|
255 |
+
software = "https://jkk.name/irc-disentanglement",
|
256 |
+
data = "https://jkk.name/irc-disentanglement",
|
257 |
+
abstract = "Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89{\%} of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.",
|
258 |
}
|
259 |
+
```
|
260 |
|
261 |
### Contributions
|
262 |
|
263 |
+
Thanks to [@dhruvjoshi1998](https://github.com/dhruvjoshi1998) for adding this dataset.
|
264 |
+
|
265 |
+
Thanks to [@jkkummerfeld](https://github.com/jkkummerfeld) for improvements to the documentation.
|
266 |
+
|
267 |
+
|
268 |
+
### Acknowledgments
|
269 |
+
|
270 |
+
This material is based in part upon work supported by IBM under contract 4915012629. Any opinions, findings, conclusions or recommendations expressed are those of the authors and do not necessarily reflect the views of IBM.
|
dataset_infos.json
CHANGED
@@ -1 +1 @@
|
|
1 |
-
{"ubuntu": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@
|
|
|
1 |
+
{"ubuntu": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large, \n title = \"A Large-Scale Corpus for Conversation Disentanglement\", \n author = \"Kummerfeld, Jonathan K. and \n Gouravajhala, Sai R. and \n Peper, Joseph J. and \n Athreya, Vignesh and \n Gunasekara, Chulaka and \n Ganhotra, Jatin and \n Patel, Siva Sankalp and \n Polymenakos, Lazaros C and \n Lasecki, Walter\", \n booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\", \n month = jul, \n year = \"2019\", \n address = \"Florence, Italy\", \n publisher = \"Association for Computational Linguistics\", \n url = \"https://aclanthology.org/P19-1374\", \n doi = \"10.18653/v1/P19-1374\", \n pages = \"3846--3856\", \n arxiv = \"https://arxiv.org/abs/1810.11118\", \n software = \"https://jkk.name/irc-disentanglement\", \n data = \"https://jkk.name/irc-disentanglement\", \n abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\", \n}", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "date": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "irc_disentangle", "config_name": "ubuntu", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 55970472, "num_examples": 220616, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 3916881, "num_examples": 15010, "dataset_name": "irc_disentangle"}, "validation": {"name": "validation", "num_bytes": 3079360, "num_examples": 12510, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470349, "checksum": "9e2c98a15191d729c0dbe10f309d836bd1ab32c0d03ffb2e0e4205287405fc4d"}}, "download_size": 118470349, "post_processing_size": null, "dataset_size": 62966713, "size_in_bytes": 181437062}, "channel_two": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large, \n title = \"A Large-Scale Corpus for Conversation Disentanglement\", \n author = \"Kummerfeld, Jonathan K. and \n Gouravajhala, Sai R. and \n Peper, Joseph J. and \n Athreya, Vignesh and \n Gunasekara, Chulaka and \n Ganhotra, Jatin and \n Patel, Siva Sankalp and \n Polymenakos, Lazaros C and \n Lasecki, Walter\", \n booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\", \n month = jul, \n year = \"2019\", \n address = \"Florence, Italy\", \n publisher = \"Association for Computational Linguistics\", \n url = \"https://aclanthology.org/P19-1374\", \n doi = \"10.18653/v1/P19-1374\", \n pages = \"3846--3856\", \n arxiv = \"https://arxiv.org/abs/1810.11118\", \n software = \"https://jkk.name/irc-disentanglement\", \n data = \"https://jkk.name/irc-disentanglement\", \n abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\", \n}", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "irc_disentangle", "config_name": "channel_two", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"dev": {"name": "dev", "num_bytes": 197189, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot": {"name": "pilot", "num_bytes": 92514, "num_examples": 501, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 186494, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot_dev": {"name": "pilot_dev", "num_bytes": 289695, "num_examples": 1501, "dataset_name": "irc_disentangle"}, "all_": {"name": "all_", "num_bytes": 495666, "num_examples": 2602, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470349, "checksum": "9e2c98a15191d729c0dbe10f309d836bd1ab32c0d03ffb2e0e4205287405fc4d"}}, "download_size": 118470349, "post_processing_size": null, "dataset_size": 1261558, "size_in_bytes": 119731907}}
|
irc_disentangle.py
CHANGED
@@ -23,19 +23,29 @@ import datasets
|
|
23 |
|
24 |
|
25 |
_CITATION = """\
|
26 |
-
@
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
}
|
40 |
"""
|
41 |
|
@@ -187,8 +197,8 @@ class IRCDisentangle(datasets.GeneratorBasedBuilder):
|
|
187 |
if self.config.name == "ubuntu":
|
188 |
# run loop for each date
|
189 |
all_files = sorted(glob.glob(os.path.join(filepath, "*.annotation.txt")))
|
190 |
-
all_dates = [Path(
|
191 |
-
all_info = [Path(
|
192 |
|
193 |
elif self.config.name == "channel_two":
|
194 |
# run loop once (there are no dates for this config)
|
|
|
23 |
|
24 |
|
25 |
_CITATION = """\
|
26 |
+
@inproceedings{kummerfeld-etal-2019-large,
|
27 |
+
title = "A Large-Scale Corpus for Conversation Disentanglement",
|
28 |
+
author = "Kummerfeld, Jonathan K. and
|
29 |
+
Gouravajhala, Sai R. and
|
30 |
+
Peper, Joseph J. and
|
31 |
+
Athreya, Vignesh and
|
32 |
+
Gunasekara, Chulaka and
|
33 |
+
Ganhotra, Jatin and
|
34 |
+
Patel, Siva Sankalp and
|
35 |
+
Polymenakos, Lazaros C and
|
36 |
+
Lasecki, Walter",
|
37 |
+
booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
|
38 |
+
month = jul,
|
39 |
+
year = "2019",
|
40 |
+
address = "Florence, Italy",
|
41 |
+
publisher = "Association for Computational Linguistics",
|
42 |
+
url = "https://aclanthology.org/P19-1374",
|
43 |
+
doi = "10.18653/v1/P19-1374",
|
44 |
+
pages = "3846--3856",
|
45 |
+
arxiv = "https://arxiv.org/abs/1810.11118",
|
46 |
+
software = "https://jkk.name/irc-disentanglement",
|
47 |
+
data = "https://jkk.name/irc-disentanglement",
|
48 |
+
abstract = "Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.",
|
49 |
}
|
50 |
"""
|
51 |
|
|
|
197 |
if self.config.name == "ubuntu":
|
198 |
# run loop for each date
|
199 |
all_files = sorted(glob.glob(os.path.join(filepath, "*.annotation.txt")))
|
200 |
+
all_dates = [Path(filename).name[:10] for filename in all_files]
|
201 |
+
all_info = [Path(filename).name[10:-15] for filename in all_files]
|
202 |
|
203 |
elif self.config.name == "channel_two":
|
204 |
# run loop once (there are no dates for this config)
|