Interactive Tweaking of Text Analytics Dashboards

Abstract

With the increasing importance of text analytics in all disciplines, e.g., science, business, and social media analytics, it has become important to extract actionable insights from text in a timely manner. Insights from text analytics are conventionally presented as visualizations and dashboards to the analyst. While these insights are intended to be set up as a one-time task and observed in a passive manner, most use cases in the real world require constant tweaking of these dashboards in order to adapt to new data analysis settings. Current systems supporting such analysis have grown from simplistic chains of aggregations to complex pipelines with a range of implicit (or latent) and explicit parametric knobs. The re-execution of such pipelines can be computationally expensive, and the increased query-response time at each step may significantly delay the analysis task. Enabling the analyst to interactively tweak and explore the space allows the analyst to get a better hold on the data and insights. We propose a novel interactive framework that allows social media analysts to tweak the text mining dashboards not just during its development stage, but also during the analytics process itself. Our framework leverages opportunities unique to text pipelines to ensure fast response times, allowing for a smooth, rich and usable exploration of an entire analytics space.