使用FUNDS微调模型。
图片:
82250337_0338.png(点击展开)
- 在测试数据的FUNDS数据集中*
--do_predict的结果是:(点击展开)
TO: S-QUESTION
FROM: S-QUESTION
DATE: S-QUESTION
MANUFACTURER: S-QUESTION
BRAND: S-QUESTION
Oct. S-QUESTION
Dec. S-QUESTION
Jan. S-QUESTION
Nov. S-QUESTION
X S-ANSWER
B&W S-QUESTION
82250337 O
COMPETITIVE B-HEADER
PRODUCT I-HEADER
INTRODUCTION I-HEADER
PROGRESS I-HEADER
REPORT E-HEADER
Sam B-ANSWER
Zolot E-ANSWER
Kool B-ANSWER
Waterfall E-ANSWER
All I-ANSWER
Packings E-ANSWER
TYPE B-QUESTION
OF I-QUESTION
PACKINGS: E-QUESTION
D. B-ANSWER
J. I-ANSWER
Landro E-ANSWER
2- B-ANSWER
Dec- I-ANSWER
97 E-ANSWER
REPORTING B-QUESTION
PERIODS: E-QUESTION
TEST B-QUESTION
MARKET I-QUESTION
GEOGRAPHY: E-QUESTION
Divisions B-ANSWER
621 I-ANSWER
and I-ANSWER
627 I-ANSWER
wisconsin E-ANSWER
PRICE B-QUESTION
POINT: E-QUESTION
FULLS I-QUESTION
$ I-QUESTION
P/ I-QUESTION
V I-QUESTION
$ E-QUESTION
(Indicate I-QUESTION
Distributor's I-QUESTION
Cost I-QUESTION
Per I-QUESTION
Carton) E-QUESTION
SALES B-HEADER
FORCE I-QUESTION
INVOLVEMENT: E-HEADER
They B-ANSWER
have I-ANSWER
crew- I-ANSWER
worked I-ANSWER
distribution, I-ANSWER
and I-ANSWER
it I-ANSWER
is I-ANSWER
reported I-ANSWER
that I-ANSWER
they I-ANSWER
may I-ANSWER
crew- I-ANSWER
work I-ANSWER
it I-ANSWER
again I-ANSWER
Sales I-ANSWER
force I-ANSWER
has I-ANSWER
been I-ANSWER
busy E-ANSWER
promoting I-ANSWER
old I-ANSWER
style I-ANSWER
backs I-ANSWER
to I-ANSWER
clean E-ANSWER
如何知道输出标签之间的关系,或者将答案与他的问题联系起来?
例如:
TO: S-QUESTION
关系:
Sam B-ANSWER
Zolot E-ANSWER
格式化的输出可以是:{ "TO": ["Sam", "Zolot"] }
或者可以更复杂:
Oct. S-QUESTION
Dec. S-QUESTION
Jan. S-QUESTION
Nov. S-QUESTION
X S-ANSWER
当X S-ANSWER仅属于Nov. S-QUESTION时,其他结果为空白(形式为空,参考图片)。
有人能帮帮我吗?:)
4条答案
按热度按时间6yjfywim1#
您可以参考FUNSD论文,该论文介绍了一种通过另一种二元分类器连接键值对的方法。
utugiqy62#
Hi,
How to use the pretrained layoutlm model directly to our task such as receipt understanding.
Is the pretrained model able to predict the receipt information.?
Thank you
vktxenjb3#
@wolfshow text_predictions.txt 不是一个链接输出。需要使用另一个分类器来链接它吗?
wgmfuz8q4#
我想知道二元分类器是如何用于链接问题和答案的。你能稍微解释一下吗?