1
00:00:00,400 --> 00:00:02,800
Hello, I am excited to present DashGuide,  

2
00:00:02,800 --> 00:00:07,760
a system for authoring dashboard 
tours that can guide dashboard users.

3
00:00:07,760 --> 00:00:14,280
Dashguide is a web-based system and works with any 
dashboard hosted in Tableau Public’s service. The  

4
00:00:14,280 --> 00:00:20,400
dashboard in this example visualizes different 
metrics from web traffic data for a website.

5
00:00:20,400 --> 00:00:25,520
Consider the scenario that the author of this 
dashboard wants to create a guide that can  

6
00:00:25,520 --> 00:00:31,000
explain the charts in the dashboard. To create 
that guide, the author can click on the record  

7
00:00:31,000 --> 00:00:36,480
button which opens a modal. The author can 
choose a communication goal and optionally  

8
00:00:36,480 --> 00:00:42,760
provide special instructions and a title for the 
tour. For this example, let’s choose “dashboard  

9
00:00:42,760 --> 00:00:47,920
composition and semantics” as the goal. 
After hitting the start recording button,  

10
00:00:47,920 --> 00:00:53,000
we can start interacting with the dashboard 
to create the tour. Each interaction in the  

11
00:00:53,000 --> 00:00:58,280
dashboard creates a placeholder step on the 
left-hand side. The author can continue to  

12
00:00:58,280 --> 00:01:05,720
interact with the dashboard to add new 
steps. After completing the interactions,  

13
00:01:05,720 --> 00:01:13,000
the author can click on this button again to stop 
recording. The system prompts GPT-4 to generate  

14
00:01:13,000 --> 00:01:19,920
titles and descriptions for the steps. We can now 
see the completed guide on the left-hand side.

15
00:01:19,920 --> 00:01:24,640
Once we have the initial guide, we can play back 
the original interactions by clicking on the play  

16
00:01:24,640 --> 00:01:31,000
buttons placed on the left-hand side of each 
step. This plays back the original interaction  

17
00:01:31,000 --> 00:01:35,440
while placing the title and description for 
the step as an overlay near the location of  

18
00:01:35,440 --> 00:01:41,840
the interaction. The author can move forward 
or backward in the dashboard tour and adjust  

19
00:01:41,840 --> 00:01:51,000
the position of the overlay. The author can 
refine this initial tour in multiple ways.

20
00:01:51,000 --> 00:01:53,520
For example, the author can provide an instruction  

21
00:01:53,520 --> 00:01:59,840
to shorten the descriptions in the 
tour and ask the LLM to regenerate.

22
00:02:02,600 --> 00:02:06,040
After regenerating, notice that 
the descriptions are now much  

23
00:02:06,040 --> 00:02:10,160
shorter than previous descriptions. 
Satisfied with the updated version,  

24
00:02:10,160 --> 00:02:15,720
the author uses the play button at the top to 
see the tour in consumer view. This feature  

25
00:02:15,720 --> 00:02:25,040
will hide everything except the dashboard, 
and start the guide from the beginning.

26
00:02:25,040 --> 00:02:30,200
There are several other features available in 
DashGuide that authors can use to create tours  

27
00:02:30,200 --> 00:02:32,160
quickly while exercising their creative  

28
00:02:32,160 --> 00:02:35,960
freedom. Let’s look at another 
example to see other features.

29
00:02:35,960 --> 00:02:41,240
The dashboard in this example analyzes 
flavors such as strawberry, orange, peach,  

30
00:02:41,240 --> 00:02:48,760
etc. in fruit snacks. Consider that the dashboard 
creator wants to author a dashboard tour showing  

31
00:02:48,760 --> 00:02:52,720
some insights. Contrary to the first example, 
the author selects “data facts” as goal this  

32
00:02:52,720 --> 00:02:59,840
time. The author then interacts 
with the charts to create the tour.  

33
00:03:03,520 --> 00:03:09,400
Notice that the created tour focuses 
on explaining the numbers and facts.

34
00:03:09,400 --> 00:03:14,280
Suppose, the author has decided to explain 
the jitter plot in the middle as people may  

35
00:03:14,280 --> 00:03:20,040
not know what a jitter plot is. The 
author opens the setting for the step,  

36
00:03:20,040 --> 00:03:25,160
changes the goal to “dashboard composition and 
semantics”, and asks the LLM to regenerate the  

37
00:03:25,160 --> 00:03:29,760
step. The revised step explains 
the jitter plot and its encoding.

38
00:03:33,800 --> 00:03:40,240
The author can add new steps by clicking on the 
record icons in between the steps. For example,  

39
00:03:40,240 --> 00:03:46,120
the author can click on the record button at 
the end to insert new steps in this position.  

40
00:03:46,120 --> 00:03:54,440
The author can then interact with the dashboard 
to add the new steps. The author can also add a  

41
00:03:54,440 --> 00:04:00,760
step without any interactions, for example by 
prompting the LLM to introduce the dashboard's  

42
00:04:00,760 --> 00:04:07,280
purpose. The steps without interactions 
appear differently than the steps with  

43
00:04:07,280 --> 00:04:14,400
interactions in the tour. Finally, it 
is possible to manually edit or write  

44
00:04:14,400 --> 00:04:20,160
the steps. The author just needs to open 
the setting for a step and edit the text.

45
00:04:20,160 --> 00:04:25,280
Following the similar process, DashGuide 
can be used to author dashboard tours for  

46
00:04:25,280 --> 00:04:31,280
a wide variety of dashboards. Please see 
our paper for more details. Thank you!