Chinese Named Entity Recognition with Inception Architecture and Weight Loss
Jun Zhao, Wei Fang · 2018
Named Entity Recognition(NER) is a very important part of many Natural Language Processing(NLP) tasks, but the accuracy rate of NER has not reached our expectation, especially in Chinese.Therefore, we studied NER task in Chinese social media.Compared to the previous papers on this dataset, we propose a new network structure that greatly improve the recognition performance.At the same time, we noticed that the recall rate of their experimental results is much lower than the precision rate, so we also explored a method to mitigate this situation by changing the loss function.Finally, the experimental results of ours with the new loss function obtained not only higher recall rates, but also significant speedup in training phase compared to the state-of-the-art methods.