摘要
文章采用深度学习技术探讨了藏文文本中的人名识别方法。首先通过word2vec训练出藏文词向量,然后在该词向量的基础上利用神经网络探讨了藏文人名识别技术,通过实验证明识别效果F1能够达到94%以上。训练出了比较好的藏文词向量,且结合藏文特点设计了检测藏文词向量好坏的方法,并采用了前向传播和随机梯度下降算法,经过多组实验验证了藏文人名识别效果。
In this paper, we discussed a method which can be used to recognize Tibetan Personal Name by using deep learning technique. First, Word2vec was used to train the Tibetan word embedding, and then neural net-work was used to distinguish Tibetan Personal Name based on the trained word embedding. Our tests proved that the efficiency of recognition (i.e. F1) can be better than 94%. In this paper, a good Tibetan word embedding is trained, and combining the characteristics of Tibetan word, we discussed a method which can be used to evaluate Tibetan word embedding. Furthermore, the efficiency of recognition of Tibetan Personal Name were tested using forward propagation and the stochastic gradient descent algorithm.
出处
《高原科学研究》
2017年第1期112-124,共13页
Plateau Science Research
基金
国家自然科学基金项目(61262058)
西藏大学珠峰学者人才发展支持计划(藏大字[2016]141号)
关键词
神经网络
藏文人名
词向量
neural network
Tibetan Personal Name
word embedding