AutoComPaste: Auto-Completing Text as an Alternative to Copy-Paste
1. AutoComPaste
Auto-Completing Text as an
Alternative to Copy-Paste
Shengdong (Shen) Zhao 1
Fanny Cheviler 2
Wei Tsang Ooi 1
Chee Yuan Lee 1
Arpit Agarwal 1,3
1
3. Background & Motivation
Current copy-paste techniques:
3
Ctrl-C, Ctrl-V Menu selection
Drag & drop X-Win
Chapuis and Roussel. Copy-and-paste between overlapping windows. CHI ’07
13. 13
+ Text Unit Adjustments
Auto-Completing Text as an
Alternative to Copy-Paste
14. 14
+ Text Unit Adjustments
Window management is
common and tedious
Copy-paste often
Interleaves typing
Copy-paste different sizes of text
is common
15. Logger Study
• Logger that logs copy-paste event
– Automatically turned on, data send to a central server
– For each copy-paste event, we record
• Type (copy | paste)
• Number of windows open, host window, and application
name
• Timestamp
• Nearest typing event in terms of time
• Content copied
– “joe12@gmail.com” is stored as “xxx00@xxxxx.xxx”
• Participants
– 22 students (9 female, 13 male, 21-27, M 23.14)
• Duration
– 2 weeks
15
16. Logger Study - Result
• Data collected
– 34.1 MB of text data, 8168 events with 3481 (43%)
copy and 4687 (57%) paste.
• Windows opened
– 83% of the time, users have 6-20 concurrently
opened windows (average 12) when performing CP
• Type of copy-paste
– 57% (2672) cross-document CP
– 43% (2015) within-document CP
• Interleaving with typing
– 42% of copy events were performed after typing, and
54% of paste events were followed by typing
• Text size
– Phrases (39%), Sentences (33%), Paragraphs (28%)
16
17. 17
+ Text Unit Adjustments
Window management is
common and tedious
Copy-paste often
Interleaves typing
Copy-paste different sizes of text
is common
19. How does AutoComPaste
Compare with Traditional
Copy-Paste Techniques?
19
Ctrl-C, Ctrl-V Menu selection
Drag & drop X-Win
Chapuis and Roussel. Copy-and-paste
between overlapping windows. CHI ’07
44. Controlled Experiment
12 university participants
X 2 techniques (XWin, ACP)
X 2 content knowledge type (known, unknown)
X 2 location knowledge type (known, unknown)
X 2 visibility type (visible, invisible)
X 2 pre-copy activity type (isolated, typing)
X 6 trials of 3 different units of text (2 phrases + 2
sentences + 2 paragraphs)
= 2304 trials total
44
46. 46
ACP has 29%
performance
benefit
XWin has 29%
performance
benefit
ACP has 140%
performance
benefit
XWin has 31%
performance
benefit
C(+) L(+)
C(-) L(+)
C(+) L(-)
C(-) L(-)
47. Qualitative Study
• 6 participants (3 female, 3 male; aged 22-25,
mean 23.8)
• Realistic trip planning task
– plan a 5-day trip to Santa Barbara by gathering
relevant information from 10 given webpages
– asked to include at least one outdoor activity, one
indoor activity, and one restaurant for each day of the
trip
• Can use either AutoComPaste and other copy-
paste techniques
47
48. Results
AutoComPaste is heavily used and highly rated by
5/6 participants
However, one rated AutoComPaste negatively
• He is a non-native English speaker participant
48
49. Conclusion
• AutoComPaste nicely complements the
traditional copy-paste techniques
– AutoComPaste has advantage when the
keyword/prefix is known
– When keywords/prefix is known and location is
unknown, AutoComPaste will have the most
advantage
– XWin has advantage when the keyword/prefix is
unknown
• Performance of AutoComPaste is subject to
typing and spelling skills
49
50. Acknowledgment
• Shi Xiaoming for programming the logger
• Guia Gali and Symon Oliver for video editing
• Study participants
• Members in the NUS-HCI lab
• This research is supported by National
University of Singapore Academic Research
Fund R-252-000-464-112
51. Q & A
51
Vignette (CHI ‘12)
You may want to check out
these other projects from
SandCanvas (CHI ‘11) MOGCLASS (CHI ‘11) Magic Cards (CHI ‘09)
earPod (CHI ‘07) Zone & Polygon Menu
(CHI ‘06)
Elastic Hierarchy
(InfoVis ‘05)
Simple Marking Menu
(UIST ‘04)
Editor's Notes
Good afternoon, my name is Shengdong Zhao. You can call me Shen. I am an assistant professor at the National University of Singapore, which is in an island in the south east of Asia. I started and manage the NUS-HCI Lab. This work is done in collaboration with Fanny Cheviler, who is currently a Post-doc researcher at OCAD University in Canada, and Ooi Wei Tsang, who is my colleague in Singapore, and Chee Yuan and Arpit, who are former students and interns in the lab.
So what is this work about:
(add affiliations for Fanny and Arpit)
Presentation flow:
1) Title slide: (using NUS-HCI as the logo)
2) Copy-paste is a common activity:
3) However, it is not necessarily optimized (show the current work flow of it)
(Can we perform copy-paste without the tedium of window switching and highlighting?)
4) What about using AutoCompletion for copy-paste purposes?
5) but it has some challenges (one dictionary size, two, unit of text)
6) We present AutoComPaste
AutoCompaste is built on several assumptions (are these assumptions true)?
How does AutoComPaste compare with traditional copy paste in different scenarios?
To answer the first question, we carry out a Field study.
to answer the second question, we carry out a controlled experiment and a qualitative study.
Future work and limitation:
----
Ok, the overll flow is ok
Need to promote the
as the title implies, it’s about some innovation we have done related to the copy-paste operation, which is very common computing operation that all of us probably perform multiple times everyday. Well, if you do if very often, you might get lucky, like the parents of this beautiful twins.
When we perform copy-paste, we often done it across different documents, in which we open many documents, and copying the various pieces of content from other documents to a working document.
Many of the copy-paste operations are done across multiple documents, in which we replicate a piece of content from one document to another.
There are a variety of ways the copy-paste operation can be carried out, in particular,
I know you can’t read it from the back, so I will try to explain the 6 steps one-by-one.
The first step, which is often a pre-step for copy and paste, is editing some document, in which you are typing some text. Then, you realize that you need to copy some content from another document
In this case, you need to switch your context, and navigate and find the window or document you believe the copied content is located,
Then, you have to perform a visual search to find the exact content you want to copy.
Then you have to use your mouse or some kind of pointing device to acquire and highlight the text, then perform the copy operation (whatever it is)
Then navigate back to the original editing document or window
Then perform the paste operation. This workflow applies to all the 4 copy-paste operation we have mentioned earlier.
However, when we exam these operations, we found it involves quite a bit of windows management, which can be unnecessary, also, step 3 and 4, visual search, acquire target and highlighting, might not be required either. Is there any ways that I can simplify this process, at least for some scenarios?
Instead of go to the source document to acquire the text, what if we build an index of currently opened documents, and use AutoCompletion to fetch the desirable content to copy directly from the editing window or document?
So we replace the middle 4 steps with typing some prefix of keyword of the text we want to copy, and perform selection from a list. But we have to be cautious here, since we don’t know how much text the user wants to copy, so we present the user with a default option, such as the most likely phrase or sentence, and provide ability for users to adjust it after the selection is done. So this is the basic idea behind AutoComPaste. This may sound like a good idea, but there are a number of assumptions we need to check increase our confidence of it.
Since AutoComPaste mainly saves on the time for windows management and visual search, we need to ensure that such type of copy-paste is common, and hopefully it is somewhat tedious and problematic currently.
Second AutoComPaste is keyboard centric technique, it will makes more sense if users often interleaves typing and copy-paste operations, so we want to find out if copy-paste does interleave with typing often. Third, what is the typical content size users copy and paste from? How much do we need to support autocompletion of different unit of text? These assumptions or questions will affect the viability or design of AutoComPaste, so we decide to carry out a log study to find out more.
We developed a logger that can be automatically turned on.
22 participants (9 female, 13 male, aged 21-27, mean 23.14) took
part in the 2-week study. All are university students in Computer
Science or Computer Engineering. Each participant was rewarded
1% course credit after completing the study.
We developed a logging mechanism that collects CP activities
running on the Windows XP/Vista/7 OS. Participants were asked
to install the logger on their primary computer for a period of 14
days. The logger was automatically turned on without any extra
operation from the user, and therefore was constantly running on
the background. Logs were periodically sent to our server.
For each CP event, the logger logs its type (copy or paste), the
host window and application, the timestamp, and the content copied.
We also record the time difference to the nearest typing event when
it applies (duration between a CP event and the latest typing event
performed before, and the earliest typing event performed after).
For each text object copied, we log its content by masking alphabetic
characters and numerical digits to protect the user’s privacy
(e.g. “joe12@gmail.com” is stored as “xxx00@xxxxx.xxx”).
Punctuation and whitespace are preserved to retain structural information
such as the number of words, sentences, and paragraphs.
A total of 34.1 MB of text logs were collected. Among the 8168
events, 3481 (43%) were copies and 4687 (57%) were pastes. A
similar distribution was observed in [15].
Windows management. We found that 83% of the time, users
have 6-20 concurrently opened windows (average 12) when performing
CP. Moreover, among all the 4687 pastes, cross-document
CP happened more often (2672 times, 57%) than within-document
CP (2015 times, 43%). This finding concurs with previous work
(only 35% of the CP events were within-document in [15]), making
a strong case for the importance of cross-document CP techniques.
Units of text copied. Understanding the granularity and amount
of text copied is important for designing AutoComPaste. Such information,
however, has not been reported in literature. We empirically
categorized the copy events into phrases (groups of 8 or less
words), single sentences (groups of 8 or more words ending with
a period), multiple sentences (at least one sentence without a newline),
and paragraphs (one or more paragraphs, each ending with a
newline).
Surprisingly, while CP of phrases is common (39%), CP of one
or more sentences (33%) and paragraphs (28%) are also frequent.
This finding suggests that a CP technique based on AC should support
different granularity of text.
Working context. Stolee et al. [15] found that word processors
were the most popular type of application while performing CP. We
extended the analysis a step further by analyzing the time interval
between CP events and typing in order to identify if CP occurs with
text editing. Empirically taking 30s as a threshold, we found that
42% of all copy events were performed after a typing event, and
54% of all paste events were followed by a typing event. These
results show that CP often occurs together with text editing.
Show AutoComPaste Video
In other words, what are the scenarios in which copy-paste is performed that will affect the performance of AutoComPaste and traditional copy-paste techniques? To find it out, we performed a bunch of pilot studies, and identified a number of factors.
I am going to use one of my own paper here so that I don’t need to get permission from someone else.
I am going to use one of my own paper here so that I don’t need to get permission from someone else.
Show a video of performance advantages.
This is demonstrated by the video.
We are young lab, so I thought some advertisement doesn’t hurt. If you like AutoComPaste, you may want to check out some of the other projects done by my lab and myself. Thank you and now I can take your questions.