Key points are not available for this paper at this time.
We are building a smart visual dialog system that aids users in investigating large and complex data sets. Given a user's data request, we automate the generation of a visual response that is tailored to the user's context. In this paper, we focus on the problem of data transformation, which is the process of preparing the raw data (e.g., cleaning and scaling) for effective visualization. Specifically, we develop an optimization-based approach to data transformation. Compared to existing approaches, which normally focus on specific transformation techniques, our work addresses how to dynamically determine proper data transformations for a wide variety of visualization situations. As a result, our work offers two unique contributions. First, we provide a general computational framework that can dynamically derive a set of data transformations to help optimize the quality of the target visualization. Second, we provide an extensible, feature-based model to uniformly represent various data transformation operations and visualization quality metrics. Our evaluation shows that our work significantly improves visualization quality and helps users to better perform their tasks.
Wen et al. (Sun,) studied this question.