yahshibu / nested-ner-tacl2020 Goto Github PK

View Code? Open in Web Editor NEW

37.0 3.0 5.0 75 KB

Implementation of Nested Named Entity Recognition

License: GNU General Public License v3.0

Python 100.00%

named-entity-recognition information-extraction natural-language-processing nested-named-entity-recognition

nested-ner-tacl2020's People

Contributors

Stargazers

Watchers

Forkers

monireh2 speedcell4 fulstock jackypersonal wangwang110

nested-ner-tacl2020's Issues

A typo

Hello,

nested-ner-tacl2020/module/crf.py

Line 53 in 7698f20

self.state_nn.bias.data[self.index_eos:] = minus_inf

I guess this should be negative_inf instead of minus_inf

about Pubmed-shuffle-win-30. bin

Hello, thank you for your open source code, but the file Pubmed-shuffle-win-30. bin cannot be downloaded now, could you please provide a copy?

Deep appreciation for how clear you present your research code

Dear authors,

I was planning on replicating nested NER results from recent literature and when I compare your repository to all others, yours is clearly structured, nicely presented, contains all the necessary instructions, and almost works out of the box.

Seriously, in Computer Science research you would expect to find more of these, yet sadly most repositories don't share these qualities.

As such, this message is not an issue but really just a "good job!" from my side to you :)

Looking forward to seeing more research (and code ;)) from your collaboration.

Cheers,

Jordy

A question about preprocessing and evaluation

Hello,

I ran your parse_ace2005.py to preprocess ACE2005 dataset, and I noticed there are some duplicated entities in the obtained files. For instance, in the train split, I found a sentence like below,

We 'll have a couple of experts come out , so I 'll withhold my comments until then .
0,1 ORG|3,7 PER|3,7 PER|11,12 PER|14,15 PER

where the 3,7 PER appears twice. Is this correct?

And then I checked your evaluation function,

nested-ner-tacl2020/util/evaluate.py

Lines 5 to 20 in 7698f20

 def evaluate(gold_entities: List[List[Tuple[int, int, int]]], pred_entities: List[List[Tuple[int, int, int]]]) \ 

 -> Tuple[int, int, int, int]: 

 prec_all_num, prec_num, recall_all_num, recall_num = 0, 0, 0, 0 

 for g_ets, p_ets in zip(gold_entities, pred_entities): 

 recall_all_num += len(g_ets) 

 prec_all_num += len(p_ets) 

 for et in g_ets: 

 if et in p_ets: 

 recall_num += 1 

 for et in p_ets: 

 if et in g_ets: 

 prec_num += 1 

 return prec_all_num, prec_num, recall_all_num, recall_num

seems you do not remove the duplicated entities, so does this have any influence on the final evaluation results?

yahshibu / nested-ner-tacl2020 Goto Github PK

nested-ner-tacl2020's People

Contributors

Stargazers

Watchers

Forkers

nested-ner-tacl2020's Issues

A typo

about Pubmed-shuffle-win-30. bin

Deep appreciation for how clear you present your research code

A question about preprocessing and evaluation

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	def evaluate(gold_entities: List[List[Tuple[int, int, int]]], pred_entities: List[List[Tuple[int, int, int]]]) \
	-> Tuple[int, int, int, int]:
	prec_all_num, prec_num, recall_all_num, recall_num = 0, 0, 0, 0
	for g_ets, p_ets in zip(gold_entities, pred_entities):
	recall_all_num += len(g_ets)
	prec_all_num += len(p_ets)

	for et in g_ets:
	if et in p_ets:
	recall_num += 1

	for et in p_ets:
	if et in g_ets:
	prec_num += 1

	return prec_all_num, prec_num, recall_all_num, recall_num