In visual analytics, sensemaking is facilitated through interactive visual exploration of data. Throughout this dynamic process, users combine their domain knowledge with the dataset to create insight. Therefore, visual analytic tools exist that aid sensemaking by providing various interaction techniques that focus on allowing users to change the visual representation through adjusting parameters of the underlying statistical model. However, we postulate that the process of sensemaking is not focused on a series of parameter adjustments, but instead, a series of perceived connections and patterns within the data. Thus, how can models for visual analytic tools be designed, so that users can express their reasoning on observations (the data), instead of directly on the model or tunable parameters? Observation level (and thus "observation") in this paper refers to the data points within a visualization. In this paper, we explore two possible observation-level interactions, namely exploratory and expressive, within the context of three statistical methods, Probabilistic Principal Component Analysis (PPCA), Multidimensional Scaling (MDS), and Generative Topographic Mapping (GTM). We discuss the importance of these two types of observation level interactions, in terms of how they occur within the sensemaking process. Further, we present use cases for GTM, MDS, and PPCA, illustrating how observation level interaction can be incorporated into visual analytic tools.KEYWORDS: observation-level interaction, visual analytics, statistical models. INDEX TERMS: H.5.0 [Human-Computer Interaction] INTRODUCTIONVisual analytics is "the science of analytical reasoning facilitated by interactive visual interfaces" [1]. The goal of visual analytics (VA) is to extract information, perform exploratory analyses, and validate hypotheses through an interactive exploration process known as sensemaking [2]. In this sensemaking loop, users proceed through a complex combination of proposing and evaluating hypotheses and schemas about their data, with the ultimate goal of gaining insight (i.e. "making sense of" the data). A wide variety of statistical models have been specifically designed for visualizations of this purpose. Thus, many visual analytic systems are fundamentally based on interaction with statistical models and algorithms, using visualization as the medium for the communication (i.e. where the interaction occurs). This communication is performed via direct interaction with the parameters of the model. For example, Interactive Principal Component Analysis, iPCA [3], allows the user to change the weight for each dimension in calculating the direction of projection using multiple sliders (one slider per dimension). Also, in an interactive visualization using MDS [4], the user can weight the dissimilarities in the calculation of the stress function through similar visual controls.In both instances, the model is made aware of the user input through a formal and direct modification of a parameter (i.e. parameter level interacti...

Typical data visualizations result from linear pipelines that start by characterizing data using a model or algorithm to reduce the dimension and summarize structure, and end by displaying the data in a reduced dimensional form. Sensemaking may take place at the end of the pipeline when users have an opportunity to observe, digest, and internalize any information displayed. However, some visualizations mask meaningful data structures when model or algorithm constraints (e.g., parameter specifications) contradict information in the data. Yet, due to the linearity of the pipeline, users do not have a natural means to adjust the displays. In this paper, we present a framework for creating dynamic data displays that rely on both mechanistic data summaries and expert judgement. The key is that we develop both the theory and methods of a new human-data interaction to which we refer as “ Visual to Parametric Interaction” (V2PI). With V2PI, the pipeline becomes bi-directional in that users are embedded in the pipeline; users learn from visualizations and the visualizations adjust to expert judgement. We demonstrate the utility of V2PI and a bi-directional pipeline with two examples.

Abstract-When high-dimensional data is visualized in a 2D plane by using parametric projection algorithms, users may wish to manipulate the layout of the data points to better reflect their domain knowledge or to explore alternative structures. However, few users are well-versed in the algorithms behind the visualizations, making parameter tweaking more of a guessing game than a series of decisive interactions. Translating user interactions into algorithmic input is a key component of Visual to Parametric Interaction (V2PI) [13]. Instead of adjusting parameters, users directly move data points on the screen, which then updates the underlying statistical model. However, we have found that some data points that are not moved by the user are just as important in the interactions as the data points that are moved. Users frequently move some data points with respect to some other "unmoved" data points that they consider as spatially contextual. However, in current V2PI interactions, these points are not explicitly identified when directly manipulating the moved points. We design a richer set of interactions that makes this context more explicit, and a new algorithm and sophisticated weighting scheme that incorporates the importance of these unmoved data points into V2PI.

No abstract

No abstract