Visual Analysis For Opinion Data Of Government Work Based On Text Mining

With the continuous development of construction of service-oriented government,the Chinese government attach increasing importance to people’s opinion on government work,and actively collects people’s opinion on government work through various democratic channels to listen to public opinion and make scientific decisions.These opinions are often targeted to different departments or different region of governments,with complex association features about department and significant association features on geospatial space.Exploring the complex association features between topic and department is helpful for government to accurately grasp the association between the topic and the corresponding department,so as to make targeted work arrangements in future,and exploring association features between topic and geospatial space is beneficial for government agencies to understand the differences of hot topics in different regions,so as to reasonable allocate of resources.However,with the advent of the era of big data,massive amounts of data make it difficult for government to quickly perceive public opinion,and they have to spend a lot of manpower to count data features.Therefore,it’s urgently needed to develop a tool that integrates data,algorithms,and front-end interactive visual display to facilitate government to quickly and accurately perceive public opinion.Opinion data about government work are often textual,text mining and visualization technologies are effective methods to analyze text.Text mining can effectively extract the hidden features of opinion data,visualization technology can express data in a graphical manner to improve transmission efficiency of information,and utilize user interface of human-computer interaction to assist users to explore large-scale data interactively and visually.However,from the perspective of direction,most of the current researches on the analysis of opinions data on government work have ignored the association features between topic and department,as well as the topic and geospatial space.From the perspective of method,more statistical methods are used,and the application of text mining and visualization technique in the field is still relatively rare.Therefore,based on text mining technology,this paper has great significance to visually analyze the association features between topic and department,as well as topic and geospatial space.The research content of this paper is mainly divided into two parts:(1)Visual analysis is carried out on association features between topic and department based on Latent Dirichlet Allocation(LDA).First,we build topic enhanced mining model by the integration of LDA and TF-IDF algorithms,based on which the association between topic and department is explored using information entropy.Second,the visual analysis system with convenient user interaction modes is designed,which adds effective means for users to deeply explore the association features between topics of public concern and government department;The system includes word cloud view to display enhanced keywords,the interactive Sankey diagram to express the associations of topic and department,and the matrix diagram to visually present opinion data distribution characteristics between topic and the department.All of these views realize the comprehensive judgment and tracking analysis of opinion data.Further,we design topic temporal evolution and emotional association visualization methods,and support the user interactively explore the relevance between topic emotional change and department.Finally,we use real data sets to analyze and demonstrate the effectiveness and practicality of the algorithm and system in this section.(2)The association between topic and geospatial space is visually analyzed based on representation learning.First,original text data are represented as high dimensional word vectors using Word2 Vec,which are mapped to two-dimensional semantic space by t-SNE algorithm.Then we use DBSCAN method to divide the semantic space into different topic semantic clusters,and the Poisson disk based adaptive blue noise sampling algorithm is utilized to extract topic keywords uniformly.The semantic similarity calculation is then used to establish the relationship between topic and geospatial space.Second,we design the visual system for topic and geospatial association analysis.This system provides projection view that shows the structure features of opinion in the semantic space,the word cloud view displaying hot topic semantic features in different districts,and the map view with the distribution of topic in geospatial space.At last,the validity and practical application value of the algorithm model and visual analysis system in this chapter are evaluated by analyzing real data set cases.The main contributions of this paper include research perspectives and research methods(1)From the view of research perspectives,this paper conducts visual analysis for association features between topic and department,as well as the topic and geospatial information,whereas other related researches ignore this to some extent.(2)From the view of research methods,this paper uses text mining and visualization techniques to analyze opinion of data government work.However,related research mainly uses traditional statistical methods.In conclusion,this paper is innovative in research perspectives and research methods.