基于字序列的非结构化简历信息解析方法

为有效解决传统简历解析方法效率低、成本高、泛化能力差的问题,提出一种基于字序列的非结构化文本简历解析方法。利用BLSTM对字序列进行建模,获得一个包含字序列信息的词表示;由BLSTM神经网络强大的学习能力对特征进行学习,获得相应的特征;根据前后标签的约束,使用CRF获得最优标签序列(CBLSTM-CRF);利用梯度下降算法训练神经网络,使用预训练字向量、Dropout优化神经网络,完成对中文简历的解析工作。实验结果表明,CBLSTM-CRF方法对简历解析的效果优于传统方案,利用BLSTM对字序列进行建模的方法在其它模型上也取得了较好的效果。

Analysis method of unstructured resume information based on character sequence

To solve the problem of low efficiency, high cost and poor generalization ability of traditional resume analysis methods effectively, an unstructured text resume analysis method based on character sequence model was proposed. A BLSTM neural network was employed to model character sequences and obtain the corresponding internal features of words. The strong learning ability of BLSTM was used to learn the features and the corresponding features were extracted. According to the constraints of the front and rear labels, the CRF was utilized to obtain the optimal labeling sequence (CBLSTM-CRF). All of the neural networks were trained using the gradient descent algorithm and optimized using the pretrained character embeddings and Dropout. Experimental results show that CBLSTM-CRF method is superior to the traditional schemes. And employing the BLSTM neural network to model character sequences achieves better results in other models.