Research On Narrative Visualization Techniques For Tabular And Textual Data

In the era of big data,various types of data have emerged in people’s lives,among which tabular and textual data are two common data types.As the volume of data increases,it is often difficult for people to extract important information from massive amounts of data.The emergence of data visualization techniques helps people transform obscure raw data into intuitive graphics and symbols,improving the efficiency of information recognition.In the field of data visualization,narrative visualization helps people understand and remember the content of data more quickly by generating visual stories.The key issue of narrative visualization is how to discover important information from data and present it in the form of a story.As the processing methods of narrative visualization are different for different types of data,this thesis mainly studies the method of generating visual stories for tabular and textual data.The work of this thesis is as follows:(1)For narrative storytelling of tabular data,this thesis proposes a BERT-based method to evaluate the semantic coherence of stories,addressing the problem that current methods for evaluating narrative coherence only consider visual coherence and neglect semantic coherence.Experiments show that the proposed method performs well in evaluating the semantic coherence of stories.For the problem of generating data comics,a user-driven method for data comic generation is proposed,which can automatically generate data comics from tabular data according to several parameters set by users.Based on these techniques,this thesis designs a data comic generation system and demonstrates the effectiveness of the system through experimental cases.(2)For narrative storytelling of textual data,this thesis proposes a keyword placement region constraint based on the medial axis to expand the existing shapebounded word cloud generation algorithm,which can generate shape-bounded storytelling word clouds.This is in response to the current problem that storytelling word cloud generation methods cannot meet specific shape requirements and shapebounded word cloud generation methods cannot incorporate semantic information of keywords.Additionally,this thesis proposes a user-preferred keyword placement direction generation method that generates word cloud layouts with different keyword placement styles.Based on these techniques,this thesis designs a shape-bounded storytelling word cloud generation system and demonstrates through experiments the effectiveness of proposed approach in generating shape-bounded storytelling word clouds and generating keyword placement directions that satisfy user preferences.