Synopsis of the work done in my Masters. It describes: 1. SahajBrowser: The benefits and comparative studies 2. Living With Trees: The cognitive experiment and results 3. The Gardener: The intellligent file browser assistant. Architecture and Results

1. Design and evaluation of
a new file browser interface
in Linux environment
Debmalya Sinha
Synopsis Seminar
January 2013

2. Objective
Provide file browsing convenience to new users
● Filesystem Visualization
● File Arrangement
● Finding Files

3. Objective
Provide file browsing convenience to new users
● Filesystem Visualization
● File Arrangement
● Finding Files
File copyDownloads
Desktop
Home
Pictures

4. Objective
Provide file browsing convenience to new users
● Filesystem Visualization
● File Arrangement
● Finding Files

5. Development
SahajBrowser
A novel file Browser
Gardener
A file browser assistant
Helps in Visualization and file
arrangement
Helps saving files into its
correct contextual place

6. Development
SahajBrowser
A novel file Browser
Gardener
A file browser assistant
Sahaj Linux

7. Work Done
1.Design and evaluation of SahajBrowser Visualization
2.Design and evaluation of SahajBrowser File Arrangement
3.Living with Trees: a cognitive study to find the nature and extent
of categorization and organization practices of target users
4.Design and evaluation of Gardener
5.Design and evaluation of Sahaj Linux

8. 1. Filesystem Visualization
● Helps users understand Folder Hierarchy – the parent-child
relationship between folders
● Reduces time for browsing through the filesystem
Good Visualization
Faster File Browsing
Clearly distinguishable
folder relationship

9. 1. Filesystem Visualization
● Placement of the folders
● Length of the representation
Good Visualization
Faster File Browsing
Clearly distinguishable
folder relationship

10. Existing file browsers
● Narrow tree-view
● Congested
● No file list in tree-view
● Length of tree depends on
number of child folders

11. SahajBrowser Visualization

12. SahajBrowser Visualization
● Constant distance
● Length does not depend on
number of child folders
● Better placement
● Intuitive parent-child relationship
● File list in tree view
● Comparing content easier

13. SahajBrowser Visualization
● Constant distance
● Length does not depend on
number of child folders
● Better placement
● Intuitive parent-child relationship
● File list in tree view
● Comparing content easier
between a parent and its child folder

14. SahajBrowser Visualization
● Constant distance
● Length does not depend on
number of child folders
● Better placement
● Intuitive parent-child relationship
● File list in tree view
● Comparing content easier

15. SahajBrowser Visualization
● Constant distance
● Length does not depend on
number of child folders
● Better placement
● Intuitive parent-child relationship
● File list in tree view
● Comparing content easier

16. SahajBrowser Visualization
● Constant distance
● Length does not depend on
number of child folders
● Better placement
● Intuitive parent-child relationship
● File list in tree view
● Comparing content easier

17. Issues addressed
Helps new users understand
the Filesystem Hierarchy
intuitively
Constant distance reduces
File browsing time
User Interaction Survey
Modelling by Fitts' Law

18. The browsing task
Task: To go from one folder in the filesystem to
another folder in an expanded treeview
1) source and destination are L levels away
2) all the folders in between source and destination has
N number of child folders in an average.
/home/ecntrk/Documents/research-writing/Thesis/Chapters

19. The browsing task
Task: To go from one folder in the filesystem to
another folder in an expanded treeview
1) source and destination are L levels away
2) all the folders in between source and destination has
N number of child folders in an average.
/home/ecntrk/Documents/research-writing/Thesis/Chapters
In this example, the level difference
L = 5
source destination

20. The task
● Browse a hierarchy of (N, L):
– N number of average child folders for each folder
– L numbers of level of the hierarchy
● We vary N from 3 to 20 and L from 4 to 8
Level 1
Level 2
Level 3
Level 4
Level 5
} L
N

21. Fitts' Law modeling
Calculates the Index of difficulty(ID)
Destination
source
B
H
D
ID = log2
(1+D/W) Where W = min(B,H)

22. Windows Explorer
● For Explorer, the distance from one source folder to a
destination folder depends on both the number of
average child folders (N) and the level difference (L)
D = (N * L * 17) px
W = 17 px
ID = log2
( 1 + N * L )

23. SahajBrowser
● For SahajBrowser, the distance between parent and
child does not depend on number of child folders.
● Thus, distance from one source folder to one
destination folder depends only on the level L
D = (L * 238.64) px
W = 170 px
ID = log2
( 1 + (1.40 L) )∗

24. Result
Windows
Explorer
SahajBrowser
Average number of child folders (N)
IndexofDifficulty
Level
8
6
4

25. User Interaction Survey
● Finds out how easily new users understand filesystem hierarchy
(with two different visualizations)
● Comparison of Explorer and SahajBrowser
● Users: 21 new users
– Computer inexperienced
– Digital experience: mobile phone
– Age: 32-40
● Were given a brief introductory concept on folder hierarchy

26. Method
● Set up and expanded a hierarchy
of level 3 with 12 child folders at
each level
● The folders has same names
(alphabets A – L ) in each level
● 3 Questions asked :
– Find parent
– Find Grandparent
– Find 2 siblings of parent
named “B” and “J”
A
B
C
D
E
F
G
H
I
J
K
L
A
B
C
D
E
F
G
H
I
J
K
L
A
B
C
D
E
F
G
H
I
J
K
L

27. Result
Q1 Q2 Q3
0
5
10
15
20
25
SahajBrowser
Explorer
Questions
ResponseTimeinseconds

28. 2. File Arrangement
One Source Folder One Destination Folder
File copy
Existing file browsers only provide

29. Multi Folder File Arrangement
Pictures
File copyDownloads
Desktop
Home

30. Multi Folder File Arrangement
Pictures
File copyDownloads
Desktop
Home
Conventional method: one-to-one copy for each source folders

31. Accross the room
Jars of candy !
Empty jar to put some candies

33. Empty Bowl
Empty jar to put some candies

34. File arrangement task
/home
user1 user2 user3
music songs Downloads
media
files
files
Level 1
Level 2
Level 3
Level 4copy

35. SahajBrowser File Arrangement
Select item A1, A2
from folder A
Select item B1, B2, B3
from folder B
Start
End
Selection Queue
Existing Browsers SahajBrowser
A1, A2
B1, B2, B3 A1, A2, B1, B2, B3
A1, A2
A1, A2, B1, B2, B3B1, B2, B3
Nil Nil
Job:

36. SahajBrowser Multi-folder selection

37. KLM-GOMS model analysis
Algorithm for Nautilus
Algorithm for Windows Explorer
Algorithm for SahajBrowser
with Multiple Selection

38. KLM-GOMS
Kieras(2003) has analytically and experimentally defined average completion
times for the primitive operators in Keystroke Level Modeling(KLM)

39. Predicted time in seconds
Assuming Level = L for every folder to operate on. We also assume
the average time to select files from one source folder is C seconds.
tnaut
= (n (5 L + C + 0.1)) sec∗ ∗
texpl
= (2.5 L + n (2.5 L + C + 1.3)) sec∗ ∗ ∗
tsahaj
= (n (2.5 L + C) + 2.5 L + 0.1) sec∗ ∗ ∗
n = total number of source folders
L = average level of every source and destination folders
C = time to select the files in a single source folder

40. 2 3 4 5 6 7 8 9 10
0
50
100
150
200
250
Nautilus
Explorer
SahajBrowser
Number of source folders
Timeinseconds
Result
Comparison of SahajBrowser with two existing file browsers;
Nautilus and Windows Explorer
Here we vary the number of source
folders “n” and have the graph for 3
different browsers.
Assuming: Level L = 4

41. 3. Finding files in a filesystem
Maa,
Have you seen
My thesis draft?

42. 3. Finding files in a filesystem
● It is pretty hard to find from an unorganized pile
● Users hardly use “find” utilities (Sasse 2003, Nardi 2001)
● People “Browse” contextually to find files in filesystem

43. Our approach
● Browsing by context needs a semantic hierarchy
● Such hierarchy needs to be maintained by users.
● We provide necessary assistance to help users maintaining
semantic folder hierarchy by putting the files into their correct
contextual place while saving
Maa,
Have you seen
My thesis draft?
PUT IT WHERE IT
BELONGS
IN THE FIRST PLACE

44. Semantic Folder Hierarchy
● It is very easy to find items while browsing in a Semantic
Folder Hierarchy
● Everytime a new file is added, it has to be put in its correct
contextual place.
● But users are reluctant to maintain it this way
● We help them maintaining the semantic hierarchy

45. Our work
● Living with Trees: a cognitive experiment that
finds out the nature of categorization and
organization practices of the target users
● Gardener: A filebrowser assistant that helps
users maintaining semantic filesystem hierarchy.

46. Our work
● Living with Trees: a cognitive experiment that
finds out the nature of categorization and
organization practices of the target users
● Gardener: A filebrowser assistant that helps
users maintaining semantic filesystem hierarchy.

47. Living with Trees
● Categorization: Everyone can .. but people
are reluctant to actually do so
● Do people organize the categories
hierarchically?
● upto how many levels target users do
organize hierarchically?
● What is the effect of external stimulation
on their performance?
● The findings are crucial for implementing
semantic folder hierarchy.

48. The method
● An unorganized pile of 82 cards (bearing pictures of common
objects and personalities) is given to the participants
● They were asked to categorize the cards on the floor
● Unlimited space given for placing
● We noted the if they are making smaller concepts from more
general concepts
● After they finishes, we asked them if they can do better
● The final performance after our stimulation, is noted

49. Sample placement

50. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
2
4
6
8
10
12
14
Participants
HighestLevelofHierarchy
Results - (top-down)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
5
10
15
20
25
30
Participants
NumberofCategories
After Stimulation
Before Stimulation
After Stimulation
Before Stimulation

51. Results - (bottom-up)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
1
2
3
4
5
6
7
Participants
LevelofBottom-upHierarchies
After
Before

52. TopDown and BottomUp comparison
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
1
2
3
4
5
6
Top Down
Bottom Up
Participants
HighestLevelofHierarchy
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0
1
2
3
4
5
6
7
8
9
Participants
HighestLevelofCategories
Top Down
Bottom Up
Before
Stimulation
After
Stimulation

53. Result
In top down approach:
1. Average Level of hierarchy during categorization is 3.
2. Average Number of Categories are 6.43
3. Both the numbers shoot up to 7 and 16.29 after some help from the experimenter
In bottom up approach:
1. Participants are much less able to categorize like this
2. Help from experimenter only increases the number to very tiny portion.

54. Conclusion
● The initial hierarchical organization is, in average,
moderate (3 levels)
● External assistance has a huge impact on the
performance (135%)
● People can organize much better if an assistant is
guiding them.

55. Challenges
● The main challenge for users is:
Maintaining the Semantic Hierarchy each time a file is saved
● Maintenance is a major problem for two main reasons:
1. Users have to “remember” all concepts in filesystem in
order to find a suitable place for the new item
2. Filesystem is non associative. Users can't directly access
files. Have to browse by the hierarchy

56. The Gardener
A filebrowser assistant to help users
create and maintain semantic folder hierarchy tree

57. How Gardener works
● Whenever a user tries to create a file, Gardener takes the input
filename/foldername and suggests a location
● It generates a set of similar keywords from the filename and then
searches the filesystem for a contextual match.
● User can store the file/folder in a single click from the Gardener
interface.
Let's see a little demo

58. Contextual Suggestions
Does not suggest only by filenames, but its context
Richard_Feynman_smiling_awkwardly.png
Save this picture..
but where?

59. Contextual Suggestions
Richard_Feynman_smiling_awkwardly.png
Gotcha..
this fellow is a
Nuclear
Physicist
Does not suggest only by filenames, but its context
~/Pictures/people/physicists
So let's save it in:

60. Architecture
Gardener is a filebrowser plugin. It has two main parts in its execution. The back-end part
has 3 blocks. The output suggestions of the backend are shown by a front end.
1. Input sanitizer: The input keyword is generated from the raw input.
2. Keyword generator: Takes sanitized input keyword and generates Hypernym and
Synonym keywords.
3. Filesystem Search: Takes this list of the keywords list searches them in the filesystem
to give parent and the peer suggestions fro the front end.
Input Filename
Input
Sanitizer
Gardener
Interface
Keyword
Generator
Filesystem
Search
Output
Suggestions
Input
keyword
Synonyms
Hypernyms
RegExp
matches

61. Architecture
Input Filename
Input
Sanitizer
Gardener
Interface
Keyword
Generator
Filesystem
Search
Output
Suggestions
Input
keyword
Synonyms
Hypernyms
RegExp
matches
falling leaves in
autumn123.jpg
['leaf', 'autumn',
'fall'] ['plant organ',
'leaf', 'season',
'leafage', 'autumn',
'foliage', 'fall', 'time
of year']
1) ~/Pictures/walpaper/seasons
2) ~/Pictures/leafy sky

62. Input Sanitizer
1. Stop Word Removal: Removes words such as “and”, “of”, etc
2. Stemming: Removes unwanted suffixes like numbers and special symbols etc
3. Lemmatization: helps finding the base word from the stemmed word.
4. POS Tagging: chooses only the nouns and verbs from a compound filename
Input Filename
Stopword
Removal
Stemming
LemmatizationKeyword
compound
input
Significant
Words
Unwanted Prefix
and Suffix removal
POS
Tagger
Clean
Words
Extracted
Nouns
Input Sanitizer Module
To be used by
Keyword Generator

63. Keyword Generator
Synonym
Generation
Hypernym
Generation
Input keyword
Synonym
keywords
Hypernym
keywords
WordNet
Database
NLTK DB
(RDF schema)
Instance of Superset
keywords

64. Filesystem Search
● Categorizes the files into 4 parts: Documents, Music & Video,
Pictures and Misc.
● Search the keywords in the appropriate folder according to file type.
● will search .jpg, .png etc in /home/Pictures (changable)
● If any match has not been found, it'll ask the user to create a new
folder with the right context
● User can cancel suggestions and browse to save in any other
folder

65. How good is Gardener
● Usability survey with 12 computer experienced users
● Task is to save 32 files with Gardener
● Categorized Gardener output into 4 parts:
– Actual words, Hypernym, Synonyms and new folder
suggesiton
● Average acceptance rate =
Total number of suggestion clicked
Total number of suggestions generated

66. Average usage of Gardener
suggestions
24.47
9.11
20.83
49.73
4.17
Actual
Synonym
Hypernym
New folder
Cancel
4.17

67. Average acceptance rate
Averageacceptance(inpercentage)
Total number of suggestion clicked
We define the “Average Acceptance rate” =
Total number of suggestions generated

68. System Usability Scale
● Measured usability with 10 questions of SUS
● Each question is a 5 point “Likert scale” type (Strongly agree (5 points) to
Strongly disagree (1 point) )
1. I think that I would like to use this system frequently.
2. I found the system unnecessarily complex.
3. I thought the system was easy to use.
4. I think that I would need the support of a technical person to be able to use this system.
5. I found the various functions in this system were well integrated.
6. I thought there was too much inconsistency in this system.
7. I would imagine that most people would learn to use this system very quickly.
8. I found the system very cumbersome to use.
9. I felt very confident using the system.
10. I needed to learn a lot of things before I could get going with this system.

69. SUS score
The average SUS score of Gardener is very high at 89.42 out of a possible 100.
Questions
Strongly
disagree

70. Future Work
Clarity.mp3
Black Pearl.jpg
User saves
Which is a song by
John Mayer
Pirates of
Caribbean
If the user doesn't provide this, Gardener can't decide the
right folder, Unless there's an entry about it in the RDF
schema, which is very impractical
We will try to include more sparse informations like these in
future from Internet.
Which is a ship in

71. Sahaj Linux

72. UI survey
● Two tasks were given to 21 target users:
– Open a text file
– Play a music file
● How to open something by clicking was explained.
● The response time was noted.

73. Results
Task 1 Task 2
0
5
10
15
20
25
30
35
40
Sahaj Linux
Windows XP
ResponseTimeinseconds

74. Overview
1.Design and evaluation of SahajBrowser Visualization
2.Design and evaluation of SahajBrowser File Arrangement
3.Living with Trees: a cognitive study to find the nature and extent
of categorization and organization practices of target users
4.Design and evaluation of Gardener
5.Design and evaluation of Sahaj Linux