Commit
·
446790a
1
Parent(s):
af958c9
Upload README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,83 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# <a href="https://arxiv.org/abs/2309.09800">AMuRD</a>: Annotated Multilingual Receipts Dataset for Cross-lingual Key Information Extraction and Classification
|
2 |
+
|
3 |
+
by
|
4 |
+
Abdelrahman Abdallah,
|
5 |
+
Mahmoud Abdalla,
|
6 |
+
Mohamed Elkasaby,
|
7 |
+
Yasser Elbendary,
|
8 |
+
Adam Jatowt
|
9 |
+
|
10 |
+
|
11 |
+
|
12 |
+
|
13 |
+

|
14 |
+
|
15 |
+
|
16 |
+
|
17 |
+
|
18 |
+
## Abstract
|
19 |
+
|
20 |
+
> Key information extraction involves recognizing and extracting text from scanned receipts,
|
21 |
+
enabling retrieval of essential content, and organizing it into structured documents.
|
22 |
+
This paper presents a novel multilingual dataset for receipt extraction, addressing key challenges in information extraction and item classification.
|
23 |
+
The dataset comprises $47,720$ samples, including annotations for item names, attributes like (price, brand, etc.), and classification into $44$ product categories.
|
24 |
+
We introduce the InstructLLaMA approach, achieving an F1 score of $0.76$ and an accuracy of $0.68$ for key information extraction and item classification.
|
25 |
+
|
26 |
+
|
27 |
+
## Demo for our Instruct LLama
|
28 |
+
|
29 |
+
|
30 |
+
Explore our Instruct LLama system through our live demo:
|
31 |
+
|
32 |
+
[**Demo for our Instruct LLama**](http://18.188.209.98:5052/)
|
33 |
+
|
34 |
+
|
35 |
+
## Examples
|
36 |
+
|
37 |
+
| Example | Input | Class | Brand | Weight | Number of units | Size of units | Price | T.Price | Pack | Unit |
|
38 |
+
| ------- | ------------------------------------ | ---------------------- | -------------| --------- | ---------------- | --------------- | ------- | ------- | ------ | ----- |
|
39 |
+
| Example 1| `40.99 20.99 2 chunks sunshine` | Tins, Jars & Packets | sunshine | No Weight | 2 | No Size of units| 20.99 | 40.99 | علبة | No Unit |
|
40 |
+
| Example 2| `برسيل اتوماتيك جل روز 2.6` | Cleaning Supplies | برسيل | 2.6ل | 1 | No Size of units| No Price| No T.Price | عبوة | ل |
|
41 |
+
| Example 3| `regina Pasta penne 400g` | Rice, Pasta & Pulses | regina | 400g | 1 | No Size of units| No Price| No T.Price | كيس | g |
|
42 |
+
| Example 4| `10.00 400g Penne Pasta ElMaleka` | Rice, Pasta & Pulses | ElMaleka | 400g | 1 | No Size of units| 10 | 10 | كيس | g |
|
43 |
+
|
44 |
+
|
45 |
+
## Getting the code
|
46 |
+
|
47 |
+
To get started with the code and utilize the AMuRD dataset for your research or projects, you can clone this repository:
|
48 |
+
|
49 |
+
```bash
|
50 |
+
git clone https://github.com/yourusername/AMuRD.git
|
51 |
+
```
|
52 |
+
|
53 |
+
|
54 |
+
|
55 |
+
|
56 |
+
## Dependencies
|
57 |
+
|
58 |
+
## Reproducing the results
|
59 |
+
|
60 |
+
|
61 |
+
|
62 |
+
## Citation
|
63 |
+
Please consider to cite our paper:
|
64 |
+
```
|
65 |
+
@misc{abdallah2023amurd,
|
66 |
+
title={AMuRD: Annotated Multilingual Receipts Dataset for Cross-lingual Key Information Extraction and Classification},
|
67 |
+
author={Abdelrahman Abdallah and Mahmoud Abdalla and Mohamed Elkasaby and Yasser Elbendary and Adam Jatowt},
|
68 |
+
year={2023},
|
69 |
+
eprint={2309.09800},
|
70 |
+
archivePrefix={arXiv},
|
71 |
+
primaryClass={cs.CL}
|
72 |
+
}
|
73 |
+
```
|
74 |
+
|
75 |
+
## License
|
76 |
+
|
77 |
+
|
78 |
+
Note: The AMuRD Dataset can only be used for non-commercial research purposes.
|
79 |
+
For researchers who want to use the AMuRD database, please first fill
|
80 |
+
in this [Application Form](Application_Form/Application_Form_for_AMuRD.doc)
|
81 |
+
and send it via email to us ([[email protected]](mailto:[email protected]), [[email protected]](mailto:[email protected]), [[email protected]](mailto:[email protected])).
|
82 |
+
|
83 |
+
|