Employing a Parametric Model for Analytic Provenance

Abstract

We introduce a propagation-based parametric symbolic model approach to supporting analytic provenance. This approach combines a script language to capture and encode the analytic process and a parametrically controlled symbolic model to represent and reuse the logic of the analysis process. Our approach first appeared in a visual analytics system called CZSaw. Using a script to capture the analyst's interactions at a meaningful system action level allows the creation of a parametrically controlled symbolic model in the form of a Directed Acyclic Graph (DAG). Using the DAG allows propagating changes. Graph nodes correspond to variables in CZSaw scripts, which are results (data and data visualizations) generated from user interactions. The user interacts with variables representing entities or relations to create the next step's results. Graph edges represent dependency relationships among nodes. Any change to a variable triggers the propagation mechanism to update downstream dependent variables and in turn updates data views to reflect the change. The analyst can reuse parts of the analysis process by assigning new values to a node in the graph. We evaluated this symbolic model approach by solving three IEEE VAST Challenge contest problems (from IEEE VAST 2008, 2009. In each of these challenges, the analyst first created a symbolic model to explore, understand, analyze, and solve a particular subproblem and then reused the model via its dependency graph propagation mechanism to solve similar subproblems. With the script and model, CZSaw supports the analytic provenance by capturing, encoding, and reusing the analysis process. The analyst can recall the chronological states of the analysis process with the CZSaw script and may interpret the underlying rationale of the analysis with the symbolic model.

Survey on the Analysis of User Interactions and Visualization Provenance

“…For example, Ragan et al [RGT15] performed a controlled study to evaluate how the presence of a visual history aid affects the memory of a user with recalling the details of their exploration process, finding that providing such an aide for a user to explore their past interactions was a substantial benefit. A similar conclusion was reached by Chen et al [CQW*14] in their propagation‐based parametric model approach, also noting that analysts were better able to recall the chronological states of the analysis process with their symbolic model. In terms of future effects, Wrangler [KPHH11] enables analysts to understand the effects of a variety of potential operations in the interaction space.…”

“…Beyond formal grammars, researchers have developed their own domain‐specific languages to encode the user's interactions with their system. In the papers by Kadivar et al [KCD*09] and Chen et al [CQW*14], the authors present the CzSaw system that generates a reusable script based on the user's interactive analysis of graphs. Kandel et al [KPHH11] propose the Wrangler system that helps a user perform data cleaning.…”

“…There is substantial overlap between the first stage of this approach and the discussion of grammar‐based encodings in Section 7.2, as the use of a grammar‐based approach [DC16], or the design of a domain specific language [KPHH11, KCD*09, CQW*14] to encode the provenance data often underlies this approach. These approaches are varied, with analysis taking place on encodings ranging from text‐based approaches such as SQL queries [FPV*13] and regular expressions [MVCJ16] to more complex forms like graphs [KCD*09, SPEB18] and Boolean logic [KBGE09, Wea09].…”

Using Machine Learning to Infer Reasoning Provenance From User Interaction Log Data

“…The initial findings suggest that the approach of capturing insight provenance was promising. Chen, Qian, Woodbury, Dill, & Shaw (2014) used a parametric symbolic model (dependency graph) to represent the provenance of an analysis. As the user interacts with a Visual Analytics tool, the symbolic model is parsed automatically from the interactions.…”

“…Although Chen's system provided automated capture of analytic provenance, the recovery of reasoning provenance was left to the analyst by browsing the dependency graph and visualisation history. In contrast to Gotz and Zhou (2008b) and Chen, Qian, Woodbury, Dill, & Shaw (2014) our aim is to recover reasoning provenance from analytic provenance without the need for manual intervention based on a learned mapping Sensemaking.…”