基于可解释性方法的网络问政效果研究——以“有话请您说”网络问政平台为例

Research on the Effectiveness of Online Administrative Inquiry Based on Explainable Methods—Taking the"Please Speak Up"Online Administrative Inquiry Platform as an Example

随着数字化政府建设的加速推进,网络问政在中国社会治理中发挥着不可忽视的作用.为了探讨影响网络问政效果的关键因素,该研究以泸州市网络问政平台“有话您请说”中的相关数据为研究对象,采用多种机器学习和深度学习模型结合的文本数据挖掘方法,识别网络问政文本中的特征变量,构建问政满意度分类模型,并结合多种可解释性方法分别从结构特征和语义特征对模型结果进行解释分析.研究发现:问政情感,问政文本长度,诉求类型,回复情感,回应单位类型,回应时间等变量均对问政满意度有不同程度的影响.此外,研究构建的可解释性框架还能够有效识别问政中的时间,地点,组织机构名称等关键内容.

With the accelerated advancement of digital government construction,online administrative inquiry plays an indispensable role in social governance of China.In order to explore the key factors affecting the effectiveness of online administrative inquiry,this study focuses on the relevant data from the Luzhou online administrative inquiry platform"Please Speak Up".This study adopts a text data mining method combining various machine learning and deep learning models to identify characteristic variables in online administrative inquiry texts,construct two public satisfaction classification models.And multiple explainable methods are used to explain the model results from both structural and semantic features.The research finds that variables such as administrative inquiry sentiment,length of administrative inquiry text,type of appeal,response sentiment,type of response agency,length of response time all have varying degrees of influence on public satisfaction.In addition,the explainable framework constructed by this study can also effectively identify key content in online administrative inquiry,such as time,location,and organization names.