[Adapt] Seminar

[王绪凯]

Hi Adapters,
    This weak, I will give u a talk about some method to improve name entity recognition with weakly labeled. Traditional supervised method for NER require a large amount of high-quality corpus (have token-level labels) for model training. So many works tend to use knowledge bases, gazetteers or dictions to generate a large amount of labeled training set from unlabeled text data named weakly labeled data. However, those weakly labeled data have some disadvantage compare to the human annotated ground truth data like: Incomplete annotation and noisy annotation. Those disadvantages an reduce the recall and cause misprediction. 
    I will introduce two papers focus on avoiding the bad effect of weakly labeled data. “Low-Resource Name Tagging Learned with Weakly Labeled Data. EMNLP-2019.” And “BOND: BERT-Assisted Open-Domain Named Entity Recognition with Distant Supervision. KDD2020.” In those two papers, the authors propose some methods to maximize the potential of weakly labeled data. In my next talk, I will show you another two papers on this topic. 
    Hope you are interested in my talk!

Time: Wed 4:00pm
Venue: SEIEE 3-414
Best Wish,
Xukai Wang