GPT-Based Textual Description Generation for Data Chart Navigability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating textual descriptions of data charts are limited in accuracy, flexibility, and accessibility, particularly for users with visual impairments, as they often rely on template-based approaches that fail to provide detailed and context-specific insights, and lack standardization in describing complex data visualizations.
Innovation Solution
A computer-implemented method using a generative pre-trained transformer (GPT) language model to compose and generate textual descriptions for data charts, incorporating precalculated statistical summaries and customizable instructions to ensure accuracy and relevance, while enhancing navigability on graphical user interfaces through keyboard or voice sequential navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template-based approaches are used to generate textual descriptions for data charts, then the implementation is simple and fast, but the accuracy and contextual relevance of the descriptions deteriorate
Solution Approach 1:
The patent transforms static template-based descriptions into dynamic, context-aware descriptions by changing the parameters of the generation process. Instead of using fixed templates, the system adjusts description parameters based on chart type, data characteristics, and user preferences, enabling both speed and accuracy through parameterized generation with precalculated statistical summaries
Solution Approach 2:
The patent introduces an intermediary layer between the chart data and the textual description. This intermediary consists of precalculated statistical summaries and structured data representations that bridge the raw data and the final description, allowing accurate yet efficient generation without direct template matching
2Loss of information
If detailed and context-specific insights are provided in chart descriptions, then the information value increases, but the complexity of generating accurate descriptions increases
Solution Approach 1:
The patent segments the description generation process into distinct components: data processing, statistical summary calculation, chart analysis, and description synthesis. This segmentation allows each component to be optimized independently, reducing overall system complexity while maintaining information completeness through modular processing pipelines
Solution Approach 2:
The patent performs preliminary actions by precalculating statistical summaries and organizing data before generating descriptions. This preliminary processing reduces the complexity of the main generation task by having essential information ready in advance, enabling detailed insights without proportionally increasing generation complexity
3Ease of operation
If accessibility features are enhanced for screen reader users, then user accessibility improves, but the interface complexity and navigation overhead increase
Solution Approach 1:
The patent implements dynamic navigation that adapts to user needs and context. The keyboard navigation system dynamically adjusts focus and announces relevant information based on the current chart state and user position, enhancing accessibility without requiring a static complex interface structure. The system responds dynamically to user actions with context-appropriate announcements
4Stability of the object's composition
If standardized methods are used to describe complex data visualizations, then consistency improves, but the flexibility to handle diverse chart types deteriorates
Solution Approach 1:
The patent creates a universal description framework that can handle multiple chart types through a single standardized interface. The system uses parameterized templates and statistical summaries that adapt to different chart types (bar charts, line charts, pie charts, etc.) while maintaining consistent structure and terminology, achieving both consistency and flexibility through multi-functional design
Data Source
AI summary
The present document discloses a computer-implemented method for generating textual descriptions for a data chart, comprising the steps of: composing a text prompt comprising a textual instruction to generate a chart description, a data series description, and data from the data chart converted to text format; sending the composed text prompt to a generative pretrained language model, which in particular implementations comprises a generative pre-trained transformer (GPT) language model; receiving from the language model a generated textual description; wherein the data from the data chart comprises an array or arrays of pairwise keys, and corresponding value datapoints. Also disclosed herein is a respective system and a non-transitory storage media including program instructions for enhancing navigability on a graphical user interface.