Chart Similarity Visualization for Collaborative Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Exploratory visual analysis becomes challenging with large and complex datasets, requiring multi-user collaboration and effective summarization of data charts to facilitate knowledge building and future data exploration directions, while existing solutions lack informative guidance and efficient information sharing among analysts.
Innovation Solution
A graphical user interface (GUI) tool utilizing deep learning techniques, specifically Grammar Variational Autoencoder (GVAE), to generate and visualize compact vector representations of charts on a 2D canvas, allowing for interactive steering and chart recommendations based on similarity, enabling analysts to identify clusters, trends, and holes in the data space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple analysts perform EVA independently and combine results iteratively, then domain knowledge can be integrated into data analysis, but the exploration space becomes overly large and coordination between analysts becomes difficult
Solution Approach 1:
The patent merges multiple analysts' charts into a unified visual summary that displays all charts simultaneously with semantic relationships. This combines the exploratory results of multiple analysts into a single coordinated view, reducing the complexity of managing separate exploration spaces while preserving domain knowledge from each analyst.
Solution Approach 2:
The system introduces an automated summarization interface as an intermediary between multiple analysts' work. This intermediary automatically generates visual summaries, identifies semantic relationships, and provides navigation assistance, reducing the coordination burden on analysts while maintaining the benefits of collaborative domain knowledge integration.
2Measurement precision
If analysts manually explore data space with numerous visual encoding options, then detailed data analysis can be performed, but the effort and time required for exploration increases significantly
Solution Approach 1:
The system performs preliminary actions by automatically generating visual summaries of all charts and pre-computing semantic relationships before the analyst needs to explore. This preliminary organization of data and visual encodings allows analysts to quickly navigate to relevant insights without manually exploring the entire data space, reducing exploration time while maintaining analysis detail.
Solution Approach 2:
The system provides feedback by analyzing the analyst's current view and automatically suggesting relevant charts, encodings, or insights that may be of interest. This feedback mechanism guides analysts through the data space efficiently, reducing the time required to discover important patterns while maintaining the ability to perform detailed analysis when needed.
3Loss of information
If all charts from multiple analysts are displayed individually, then complete information is available, but obtaining an understanding of past work and coordinating between analysts becomes difficult
Solution Approach 1:
The patent merges multiple individual charts into a unified visual summary that displays all charts simultaneously with their semantic relationships. This merged view maintains complete information from all analysts while making it easy to understand the overall exploration landscape and coordinate between analysts through visual navigation.
Solution Approach 2:
The system adds a new dimension to chart display by organizing charts in a multi-dimensional visual space that preserves semantic relationships. This dimensional organization allows analysts to see all charts and their relationships simultaneously, making it easy to understand past work and coordinate future exploration without losing information completeness.
Data Source
AI summary
Example implementations described herein are directed to a graphical user interface (GUI) tool that provides representations of generated charts on a map, wherein distances between representations are provided based on similarity between charts. Similarity is determined through machine learning techniques that are applied on a vectorized form of charts. Example implementations described herein encode charts into vectors using deep learning techniques, which facilitates machine learning techniques such as nearest neighbor to be utilized to determine similarity between charts based on their corresponding vectors.