基于卷积神经网络的FrameNet框架消歧研究

Research on Disambiguation of FrameNet Framework Based on Convolutional Neural Network

针对当前FrameNet框架下的词义消歧准确率较低的问题,采用卷积神经网络应用于FrameNet框架进行框架消歧研究.该模型依托依存句法分析树排序选出待消歧词的6个邻接单词节点,并选择单词词义、父节点词义、单词词性、单词依存分析类型作为消歧特征,使用Softmax函数作为全连接层分类器,通过输出待消歧词可激活的各框架概率选出概率值最高的作为激活框架,从而判定待消歧词词义.实验结果表明,该模型在FrameNet框架的消歧准确率较高于条件随机场等其他普遍算法,各目标词的准确率较为稳定,通过该模型切实提升了FrameNet框架消歧的准确率.

Aiming at the problem of low accuracy of word sense disambiguation under the current FrameNet framework,the convolutional neural network was applied to the FrameNet framework for frame disambiguation research.Adjacency words which need be disambiguated could be selected by dependency syntax tree.Softmax function was used as the full connect layers(FC)classifier,and the word sense,parent node word sense,word part of speech and word dependency analysis type were selected as the disambiguation features.By outputting the frame probabilities that the word to be disambiguated could be activated,the one with the highest probability value was selected as the activation frame to determine the meaning of the word to be disambiguated.The experiment results show that the disambiguation accuracy of the model in the FrameNet framework is higher than other general algorithms such as conditional random fields,and the accuracy of each target word is relatively stable.The accuracy of the disambiguation of the FrameNet framework is effectively improved by the model.