Stanford CS448B 16 VisNLP
TLDR
This article contains my notes from Stanford's CS448B (Data Visualization) course, specifically focusing on the sixteenth lecture about visualization and natural language processing. I'll discuss the importance of linking text with visualization, understanding how readers integrate charts and captions, visual question and answering, and natural language interfaces.
Original
Notes
GVC trade grew rapidly in the 1990s but stagnated after the 2008 global financial crisis
Question: How much did the GVC share rise between 1990 and 2008?
“Words and pictures belong together.”
--Tufte, 1983
Linking text with visualization
Facilitating Document Reading by Linking Text and Tables
--Kim et al. 2018
Stage 1: Table Structure Extraction
Stage 2: Match Sentence Text to Table Cells
Stage 3: Rule-based Refinement of Matches
Pipeline Evaluation
Corpus
- Pew Research Reports
- ACL and CVPR papers
- Kong et. al (2014)
User Study
Hypothesis:Our interface facilitates reading documents with tables
Within-subject study
14 adult volunteers, all fluent in English
Task: Annotate references with/without our interface
Understanding how readers integrate charts and captions
Do readers rely more on the chart or captions for their takeaways?
When text and visualization emphasis mismatch, readers rely more on the chart and can miss information in the caption
Visual question and answering
Formative Study
Explanations describe procedure for computing answer
Half of the explanations referred to visual features of chart
Chart QA Pipeline and Generating Explanations
User Study
- 16 participants
- 20 Chart-Question-Answer-Explanation tuples (5 per explanation type)