Chart Generation for Contextualized Responses to NL COVID-19 - - PowerPoint PPT Presentation
Chart Generation for Contextualized Responses to NL COVID-19 Queries Hannah DeBalsi, Bianca Yu CS 294W Final Presentation Outline Motivation User Research Chart Generation Future Work Quick Recap Quick Recap Final Results Final Results
CS 294W Final Presentation
Motivation User Research Chart Generation Future Work Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned
User Research Chart Generation Future Work Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned Motivation
Alexa, how many COVID-19 cases are there currently worldwide? Ok. There are 6.8 million confirmed cases
Alexa, how many COVID-19 cases are there currently worldwide? Ok. There are 6.8 million confirmed cases
Alexa, how many COVID-19 cases are there currently worldwide? Interesting ... looks like cases are increasing. There are 6.8 million confirmed cases of COVID-19 worldwide, as of June 7.
Alexa, how many COVID-19 cases are there currently worldwide? Interesting ... looks like cases are increasing. There are 6.8 million confirmed cases of COVID-19 worldwide, as of June 7.
User Research Chart Generation Future Work Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned Motivation
User Research Quick Recap Final Results Lessons Learned Final Results Lessons Learned Chart Generation Future Work Quick Recap Motivation
What do users want to know about COVID-19?
+ What are the most common quantitative questions?
Survey #1
What do users want to know about COVID-19?
+ What are the most common quantitative questions?
Survey #1 Survey #2 What kinds of COVID-19 questions are best answered visually?
+ What is the most effective graph for these questions?
What do users want to know about COVID-19?
+ What are the most common quantitative questions?
Survey #1 Survey #2 What kinds of COVID-19 questions are best answered visually?
+ What is the most effective graph for these questions?
Preliminary Semantic Parser
User Research Quick Recap Final Results Lessons Learned Final Results Lessons Learned Chart Generation Future Work Quick Recap Motivation
User Research Quick Recap Final Results Lessons Learned Lessons Learned Chart Generation Future Work Quick Recap Final Results Motivation
Survey #1
Survey #1
Survey #2
Survey #2
Significant agreement: the percentage of participants who assigned the response a particular rank is ≥ 50%
Survey #2
Significant agreement: the percentage of participants who assigned the response a particular rank is ≥ 50% Qs w/ significant agreement on first choice
BAR CHART LINE CHART
Survey #2
Significant agreement: the percentage of participants who assigned the response a particular rank is ≥ 50% Qs w/ significant agreement on first choice Qs w/ significant agreement on last choice
BAR CHART LINE CHART TEXT ONLY
Preliminary Semantic Parser
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
“How many” “cases” “Georgia” “new” “today”
Interrogative phrase Answer type (quantity) Answer type (noun) Location Filter (quantitative) Filter (date/time)
L A B E L S
“How many” “cases” “Georgia” “new” “today” → Chart Type: Line Graph → Column Name: “Positive” → Row: State == “Georgia” → Report single cell value → Date: 06/08/2020
User Research Quick Recap Final Results Lessons Learned Lessons Learned Chart Generation Future Work Quick Recap Final Results Motivation
User Research Quick Recap Final Results Lessons Learned Chart Generation Future Work Quick Recap Final Results Lessons Learned Motivation
A text + graph response is more helpful than just text.
Users are mostly interested in change over time.
For quantitative data, text + graph response seems to always be more helpful.A text + graph response is more helpful than just text.
Users are mostly interested in change over time.
For quantitative data, text + graph response seems to always be more helpful.A text + graph response is more helpful than just text. The graph content itself is critical.
User Research Quick Recap Final Results Lessons Learned Chart Generation Future Work Quick Recap Final Results Lessons Learned Motivation
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned Future Work Motivation
○ Generating plots for questions already supported by ThingTalk (ie weather) ○ Generating plots in Python and then displaying an image of the plot to the user
○ Generating plots for questions about coronavirus ○ Generating plots in Javascript using Chart.js
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned Future Work Motivation
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Lessons Learned Future Work Final Results Motivation
\t now => @com.covidtracking.state(state='ca') => notify;
\t now => @com.covidtracking.us() => notify;
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Lessons Learned Future Work Final Results Motivation
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Future Work Final Results Lessons Learned Motivation
JavaScript High level ideas → actual implementation People
User Research Chart Generation Quick Recap Final Results Lessons Learned Quick Recap Future Work Final Results Lessons Learned Motivation
User Research Chart Generation Future Work Quick Recap Final Results Lessons Learned Quick Recap Final Results Lessons Learned Motivation
Connect the pipeline More types of charts More robust mappings