Research Diary


Oct 20, 2018

Reading the core code, and now I’m still debuging.

Oct 19, 2018

Installing the environment of NEL, which is so complicate. and run some examples.

Oct 16, 2018

Discussing some issues of NEL and NER with teamates.

Oct 15, 2018

Read the paper of NEL.

Oct 13, 2018

Aiming at enhancing the NER performance on wiki-style passage, I read the paper Learning multilingual named entity recognition from Wikipedia, which proves that differences between CoNLL and Wikipedia, traning on the CoNLL will lead to bad performance on Wikigold. And we can create a wiki-style training dataset, which can enhancing the performance of NER on Wikipedia. And I think after finishing the pipeline of NER and NEL, I can try it.

Oct 10, 2018

Ran the old NER model and code, finding that new code runs faster and the f1 score is a little bit higher than the old.

Oct 9, 2018

Tested the model on WikiGold dataset, which is manually annotated, only to found the fi score is 65. And I changed the IO annotation to BIO annotation, finding that BIO is harder for NER model.

Oct 8, 2018

After finishing modifying the NER model and code, I ran the code of NER, training on the CoNLL dataset, the F1 score on testb can reach 90.6.

Sep 29, 2018

Got the NER dataset of WekiPedia, wikigold.conll. I viewed the whole project, and processing teh dataset, now I’m still debugging.

Sep 28, 2018

View the blog given by the author, and read sevaral papers, making myself qualified to writing relating codes.

Sep 27, 2018

Recieve the work of NER(QA4IE), I fund a related work of NER, which is implemented by advanced tensorflow API. The speed and the quality of NER are nice, I plan to redo the work of NER in QA4IE, enhancing the main entity recognizing score from 85% to 95%. In addition, I planed to use some dataset which is related to Weki. I ran several small experiments given by the author, and plan to learn basic knowledge of NER tormorrow.