Constructing a Taxonomy for Sentiment Visualization Analysis Using Visual Metaphors

Abstract

Background Due to the recent development of data mining and Natural Language Processing(NLP) technologies, sentiment analysis targets are more diverse tendencies than just information with affirmative or negative side. Accordingly, there are increasing cases of analyzing sentiment information using high-dimensional visualization technology, and these cases are often difficult to understand from a public point of view. In order to solve these problems, there is an increasing movement to visualize sentiment information using visual metaphors. Therefore, for the purpose of easier to understand related cases, there is a need for a measure to systematically organize information on their research methods, purposes, and visual metaphors. Methods In this study, a taxonomy is proposed that can examine in detail the research process of sentiment analysis visualization cases based on visual metaphors. First, sentiment visualization cases based on visual metaphors are collected and used as data for constructing a taxonomy. Second, selecting the criteria that constitute the taxonomy, and based on the step-by-step analysis work that appears in the metaphor process, the attributes of the criteria are largely divided into five elements (Target, Task-oriented Intermediation, Representation, Visual Variables and Visualization Technique) and detailed sub-elements are selected. Third, classification work is performed using an actual study based on the created taxonomy. Finally, in order to find the utility and improvements of the taxonomy, an qualitative evaluation is conducted for subjects. Results The designed taxonomy in this study provided it easy to understand what kind of sentiment information the visual metaphor from sentiment visualization is based on, what motives or backgrounds the metaphor has progressed, what representation have replaced sentiment information, and how visual variables are performed to add interpretive meaning to the representation. During the verification process, it was estimated that the taxonomy of this study helps to understand inclusively sentiment visualizations using visual metaphors. On the other hand, we also confirmed the need for adding more Subcategory level 1 elements to the representation and visualization techniques. Furthermore, specifying as well as subdividing the definitions of both criteria's element is just as necessary. Conclusions We expect that the taxonomy proposed in this study can be a guideline to inform researchers of the visual techniques and ideas that is required to become a comprehendible sentiment analysis visualization result to users. In future work, this study will improve the taxonomy to easily explain the visual metaphor process that appears in different cases. In addition, we will conduct a quantitative evaluation of several related researchers to verify the effectiveness of the classification system and the understanding of the visual metaphor process. Finally, more diverse visual metaphor cases will be classified and converted into a database, and we will create a system that can help users to freely explore related cases.