ChartDetective: Easy and Accurate Interactive Data Extraction from Complex Vector Charts

Abstract

elements partially or completely occluded. Compared to other approaches relying on raster images, our tool successfully recovered all data, even when hidden, with a 78% lower relative error. CCS CONCEPTS• Human-centered computing → Interactive systems and tools.

Statslator: Interactive Translation of NHST and Estimation Statistics Reporting Styles in Scientific Documents

“…For the remaining 2,642 papers, additional values will be needed. Overall, a large proportion of CHI papers have useful statistical reports, and recall that our conservative analysis likely underestimate actual occurrences and it is possible that values like means can be estimated from figures [58,73] 3.2.2 Number of Decimals. Consistent with APA recommendations [5, sec 6.36], standard deviations, means, F-values, t-scores, and CIs were reported with a median of two decimals.…”

“…However, the SplitBoard paper does not report aggregated means with standard deviations. Sam decides to retrieve these values from the line chart of the WPM using an accurate chart data extraction tool [58,73]. For SplitBoard, the t-score was directly obtained enabling the calculation of Cohen's d. Once the missing values are added in the table, the other values are calculated and Sam can review the effect sizes: Cohen's 𝑑 = 0.64 for SwipeBoard, and 𝑑 = 7.27 for SplitBoard.…”

Charagraph: Interactive Generation of Charts for Realtime Annotation of Data-Rich Paragraphs

“…Some systems help readers interpret and manipulate the results by enhancing the visualizations already present in the document. For example, redesigning charts to be more useful [128], with added overlays [73], and interactive features [88,95]. Or the text can be leveraged to automatically annotate existing charts [62,76] and interactively connect text and charts [113].…”

DIVI: Dynamically Interactive Visualization

“…ReVision [32], Graphical Overlays [12], REV (Reverse-Engineering Visualizations) [23,24], ChartSense [11], and approaches surveyed with Chart Mining [7] use computer vision techniques to automatically extract marks and infer visual encoding channels and data attributes from bitmap images. However, these techniques suffer from a variety of issues that affect inference precision, such as pixel resolution, shape ambiguity (e.g., blurry edges), and mark occlusion [17]. While our work targets structured SVG images, these techniques might enable DIVI to support a broader set of visualization media types in the future.…”

“…al [13] contribute a spatial-constraint model to enable interaction with static visualizations, although its deconstruction output is limited to evenly spaced axes (prohibiting use of log scales) and marks, without support for linked interaction and other non-spatial exploratory interactions (e.g., tooltips and selection / brushing). ChartDetective [17] provides semi-automated data extraction methods for vector charts in PDFs, but lacks support for chart metadata (such as axes) needed for interaction and linking. In addition, semi-automated approaches requiring extensive user input are not as scalable.…”