Update the dataset card to provide a better description of the dataset.
Browse files
README.md
CHANGED
@@ -4,24 +4,37 @@ configs:
|
|
4 |
- config_name: david
|
5 |
data_files:
|
6 |
- split: train
|
7 |
-
path:
|
8 |
- split: test
|
9 |
-
path:
|
10 |
- config_name: isabel
|
11 |
data_files:
|
12 |
- split: train
|
13 |
-
path:
|
14 |
- split: test
|
15 |
-
path:
|
16 |
- config_name: marcus
|
17 |
data_files:
|
18 |
- split: train
|
19 |
-
path:
|
20 |
- split: test
|
21 |
-
path:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
---
|
23 |
|
24 |
# The Panza Emails dataset
|
25 |
|
26 |
-
This dataset contains
|
27 |
Except for these changes, the language of the emails is genuine. The intention of this dataset is to allow researchers to study strategies for text personalization.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
- config_name: david
|
5 |
data_files:
|
6 |
- split: train
|
7 |
+
path: david/train.jsonl
|
8 |
- split: test
|
9 |
+
path: david/test.jsonl
|
10 |
- config_name: isabel
|
11 |
data_files:
|
12 |
- split: train
|
13 |
+
path: isabel/train.jsonl
|
14 |
- split: test
|
15 |
+
path: isabel/test.jsonl
|
16 |
- config_name: marcus
|
17 |
data_files:
|
18 |
- split: train
|
19 |
+
path: marcus/train.jsonl
|
20 |
- split: test
|
21 |
+
path: marcus/test.jsonl
|
22 |
+
task_categories:
|
23 |
+
- text2text-generation
|
24 |
+
language:
|
25 |
+
- en
|
26 |
+
pretty_name: Panza email
|
27 |
+
size_categories:
|
28 |
+
- n<1K
|
29 |
---
|
30 |
|
31 |
# The Panza Emails dataset
|
32 |
|
33 |
+
This dataset contains collections of emails of three authentic users (david, isabel, and marcus), with personal information (names, places, etc.) replaced by other ones for donor privacy.
|
34 |
Except for these changes, the language of the emails is genuine. The intention of this dataset is to allow researchers to study strategies for text personalization.
|
35 |
+
The data was donated explicitly for this purpose. This dataset is ethically collected and fully licensed for research use. The dataset
|
36 |
+
is entirely in English, and all authors are highly proficient in the language; all messages are authored entirely by the donor, without any
|
37 |
+
collaborative writing or LLM use.
|
38 |
+
|
39 |
+
This data was used in the paper [**Panza: Design and Analysis of a Fully-Local Personalized Text Writing Assistant**](https://arxiv.org/abs/2407.10994) to fine-tune open-source models to impersonate a specific author.
|
40 |
+
Code to do so is available on [GitHub](https://github.com/IST-DASLab/PanzaMail).
|