基于深度置信网络的语义相关度计算模型

Computational Model of Semantic Relevancy Based on Deep Belif Net

语义相关度是问答系统等领域的关键技术之一,目前的相关度计算方法往往对语义因素考虑不全,造成计算结果的准确度不足。由受限玻尔兹曼机构造的深度置信网是一种深度学习模型,能模仿人类大脑抽象语义特征,由此提出了一种基于深度置信网络的语义相关度计算模型。首先,对组成模型的受限玻尔兹曼机进行介绍;然后,详细阐述了模型的构建及其训练和微调过程;最后,通过对比实验验证了提出的计算模型相对基准方法在评价指标上获得了更好的效果。

Semantic relevancy is one of the key technologies of QA systems etc. The current method to calculate the similarity are not completely in terms of semantics, resulting the lack of accuracy. The deep belief net maded by the restricted Bohzmann machines is a deep learning model. It can imitate the human brain to abstract semantic features. Thus, a computational model of semantic relevancy based on the deep belief net is proposed. First of all, the restricted Boltzmann machines which comprised the model is introduced. Then, the process of how the model is constructed and how to train and fine-tune the model is described in detail. Finally, the contrast experiment shows the proposed algorithm has better results in evaluation indicators compared with the reference method.