摘要
为了解决传统办案方式可能引起的"类案不同判"等问题,以及满足当事人查找预览与自身情况相似裁判文书的需求,本文提出了一种基于多模态特征融合的裁判文书推荐方法,学习高层次的裁判文书多模态融合特征表示,进而实现相似裁判文书推荐.该方法主要包括预处理、特征提取、特征融合和文书推荐四个阶段.实验结果表明,与只利用单一模态特征以及简单串联多模态特征的方法相比,利用本文方法学习到的多模态融合特征进行裁判文书推荐,推荐结果的准确率、召回率和综合评价指标(F1值)均有显著提高.说明本文提出的多模态特征融合方法对于推荐任务更加有效,体现了算法的优越性.
In order to address the problems such as"different judgments in similar cases"caused by the traditional way of handling cases,and to satisfy the parties’needs to search and preview the judgments which are similar to their own situation,this paper proposes a judgments recommendation method based on multi-modal feature fusion,which can learn high-level fusion feature representation from multiple modalities for the judgments.Then the judgments recommendation can be performed using the learned fusion feature.The proposed method consists of four stages,i.e.,data preprocessing,feature extraction,mult-modal feature fusion,and judgments recommendation.Extensive experiments demonstrate that,compared to the methods using only single-modal features and simple concatenation of multi-modal features,the multi-modal fusion features learned by our method achieves significant improvement in precision,recall,and F1 Value for judgments recommendation.It shows that the multi-modal feature fusion method proposed in this paper is effective for judgments recommendation.
作者
原旭
韩雪姣
陈志奎
钟芳明
赵亮
YUAN Xu;HAN Xuejiao;CHEN Zhi-kui;ZHONG Fang-ming;ZHAO Liang(School of Software Technology,Dalian University of Technology,Dalian 116620,China;Key Laboratory for Ubiquitous Network and Service Software of Liaoning Province,Dalian 116620,China)
出处
《微电子学与计算机》
北大核心
2020年第12期42-47,共6页
Microelectronics & Computer
基金
国家重点研发计划资助(2018YFC0830200,2018YFC0830203)。
关键词
裁判文书推荐
多模态
特征融合
深度神经网络
judgments recommendation
mult-modal
feature fusion
deep neural network
