基于CNN和深层语义匹配的中文实体链接模型

实体链接是知识图谱领域的重要研究内容，现有的实体链接模型研究大多集中在对手工特征的选择上，不能很好地利用实体间的语义信息来实现更高效的实体链接效果。故提出一个基于深度语义匹配模型和卷积神经网络的实体链接模型，候选实体生成阶段采用构造同名字典，并基于上下文进行字典扩充，通过匹配来选择候选实体集。通过卷积神经网络来捕获深层语义信息，进行特征提取，并将其作为语义匹配模型的输入，通过模型训练学习选择出最佳参数，并输出语义相似度最高的候选实体作为实体链接的结果。在NLP & CC2014_ERL 数据集上较Ranking SVM模型准确率提升了3.9%，达到86.7%。实验结果表明了提出的新模型性能优于当前的主流模型。

A Chinese entity linking model based on CNN and deep structured semantic model

Entity Linking is an important research content in the field of Knowledge Graph. Most of the existing entity linking models focus on the selection of manual features, which cannot make good use of the semantic information between entities to achieve better efficient entity linking effect. Therefore, an improved entity linking model based on deep structured semantic model and convolutional neural network is proposed. It captures deep semantic information and extracts features through CNN, and uses them as input of the deep structured semantic model. It selects the best parameter through model training, and outputs candidate entities with the highest semantic similarity as the result of entity linking. Compared with the ranking SVM model, the proposed model improves the accuracy by 3.9% to 86.7% on the NLP & CC2014_ERL dataset. The experimental results show that the proposed model is effective and superior to the current mainstream model in entity linking tasks.