基于知识表示增强的类案推荐模型

Similar case recommendation model based on knowledge representation enhancement

为解决推荐模型面对长文本推荐任务存在推荐信息获取不准确的问题,提出一种基于知识表示增强的类案推荐模型。在该模型中,将案件案情信息构建为案件知识图谱后进行图谱嵌入学习,缓解知识嵌入中向量稀疏问题。为提升类案推荐结果的准确性和全面性,基于知识图谱和注意力机制增强查询案件和候选案件的语义向量,使用文本卷积神经网络对增强知识向量进行特征提取,使模型具备更深层次的语义理解能力,提升模型对案件知识的表达能力,提高类案推荐模型的准确性。在公开数据集LeCaRD上的实验结果表明,该模型能够有效提升推荐结果的准确性和全面性。

To solve the problem of inaccurate acquisition of recommendation information in the recommendation model for long text recommendation tasks,a similar case recommendation model based on knowledge representation enhancement was proposed.In this model,the case information was constructed as a case knowledge graph and then the graph embedding learning was performed to alleviate the vector sparsity problem in knowledge embedding.To improve the accuracy and comprehensiveness of similar case recommendation results,the semantic vectors of query cases and candidate cases were enhanced based on knowledge graphs and attention mechanisms.The text convolutional neural network was used to extract features of the enhanced knowledge vector,so that the model had a deeper semantic understanding ability,thereby improving the model’s ability to express case knowledge and improve the accuracy of similar case recommendation models.Experimental results on the public dataset LeCaRD show that the model can effectively improve the accuracy and comprehensiveness of the recommendation results.