Interactive Exploration of Versions across Multiple Documents

Introduction The need to compare two or more documents arises in
a variety of situations. Some instances include detection
of plagiarism in academic settings and comparing
versions of computer programs. Extensive research has
been performed on comparing documents based on their
content (Si et al., 1997; Brin et al., 1995) and there also
exist several tools such as windiff to visually compare a
pair of documents. However, little work has been done
on providing an effective visual interface to facilitate the
comparison of more than two documents simultaneously.
The Versioning Machine (Schreibman et al., 2003) is
a web-based interface that provides the facility to view
multiple versions of a document, along with the changes
across versions. Motivated by the Versioning Machine
(VM), we build a tool MultiVersioner that facilitates
viewing multiple versions of multiple documents at
once, and provides the user with a rich set of information
regarding their comparison. The primary user during the
development of MultiVersioner was Tanya Clement, a
doctoral candidate in English at the University of Maryland,
who researches the works of experimental poets.
Related Work
ScentHighlights (Chi et al., 2005) has demonstrated the
effectiveness of using color-coded highlighting to display
the similarities and differences across documents.
There exist literature (Brin et al., 1995) and tools like
CHECK (Si et al., 1997) and MOSS (MOSS) on plagiarism
and source code comparison, which are relevant to
our work. FeatureLens (Don et al., 2007) facilitates pattern
finding in text collections by providing visualizations
of the results of text mining algorithms.
In Tanya Clement’s research, she compares not only versions
of a single poem, but also multiple versions across
several poems. VM can display the versions of just one
document at a time. To open another document, all versions
of the current document have to be closed first. VM
also does not provide any search capabilities.
Description of the Interface
Background
The two-fold goal of MultiVersioner is to provide an
effective overview of the content and size of all documents,
as well as to provide a detailed display, along
with a variety of search capabilities, in accordance with
the Info-Viz Mantra Overview first, zoom and filter, details
on demand (Shneiderman 1996).
MultiVersioner is implemented in Java 6.0 using the
Swing GUI toolkit. It uses the same input format file as
VM, an XML file, containing information about the various
additions and deletions made across all versions of
a document. MultiVersioner contains a built-in parser to
parse these XML files. Loading an XML file opens all
the versions of a poem in separate version panels and
multiple such documents can be loaded simultaneously.
Version panels are displayed in the central part of the
interface with a tool panel located on the right. The name
of the version appears on top of the respective version
panel. The names of all the versions of a particular document
are displayed in the same color in order to group
them together.
Overview
In the overview, words are denoted by equal sized boxes.
Hovering over a box pops up a tooltip containing the entire
sentence, with the current word being displayed in
bold. In the tooltip, words added in the current version
are shown in blue, and words deleted are shown in red.
Clicking on a box brings up a detail window (Figure 1)
containing the entire sentence. The purpose of the detail window is to display a sentence of interest on the
screen, analogous to a post-it note. The detail windows
can be made either opaque or transparent, and can either
be moved around, or aligned together. A line is drawn
between a detail window and its corresponding location
in the version panel, to keep track of its origin. A detail
window could be closed either by right clicking it
and choosing close, or by simply dragging it out of the
screen. Text View
Word boxes are used primarily to obtain an overview of
all the documents. To explore the versions in detail, a
representation displaying the actual sentences, instead of
word boxes, is preferred (Figure 2).
Search
The basic search feature is the word search (Figure 2).
A search bar is provided where the user can type in a
word or a phrase to be searched across all documents.
A search can be made case-sensitive if desired. Alternatively,
a word can be searched by right-clicking an instance
of it. Inspired by ScentHighlights (Chi 2005 et al.,
2005), Search results are color-coded. The instances of
a searched word in all documents are highlighted using
the same color. A search history as well as the facility to
clear search results is available.
A line search feature is available as well. The similarity
of a pair of lines is computed by taking into account the
number and relative positions of the words common to
them. Right-clicking the anchor-box present at the beginning
of a line triggers a line search and lines similar to
the specified line across all documents are highlighted. Word Frequency Table
MultiVersioner computes a frequency table containing
the number of occurrences of each unique word in all
documents and their versions. When comparing different
versions of a document or comparing different documents
that are related, researchers in literature need to
identify unique and common words and sentences. It has
been shown that an approach as simple as a frequency
table listing is powerful in providing insight by letting
users know which words are common across documents
and which ones are unique to a single document (Filippova,
2007). Other features
There are sliders available to control the version panel
height, width and the sizes of the word boxes. MultiVersioner
also has a scroll lock, used when the documents
are long, to synchronize the scrolling of the documents
with each other.
User Feedback and Future Work
The first three authors of the abstract were the developers
(students in an Information Visualization class). The
last three authors used the prototypes several times during
the development and provided suggestions. Examples
are given below.
Users reported that they could not easily distinguish versions
of a single document from versions of other documents.
We achieved this by using the same color for the
titles of all the version panels associated with a the same
document. The ability to search for words across the
documents was greatly appreciated and the color-coded
highlighting of the results was found helpful, but overall they commented that softer colors would be more pleasing.
Users liked the synchronized scrolling between windows.
Still it was difficult to associate the detail window
to its originating location, so some visual linking was
suggested (e.g. drawing a line or matching color highlighting).
Regarding the choice between Text View and
Overview, users stressed that it is preferable to see the
actual words rather than seeing the abstract word boxes.
Still, they acknowledged the benefit of the overview
when dealing with a large number of documents and versions.
The overview was more helpful for a high-level
view and orientation, the text view was more useful for
analysis.
In summary we believe that our prototype illustrates that
it is possible to facilitate visual comparison. We built
on the Versioning Machine by allowing users to compare
multiple documents, each of which consisting of
multiple versions. We also provide the ability to search
for entities such as words and lines across the documents
and versions and analyze their frequency patterns.
MultiVersioner was designed to compare small poems,
and future work needs to address the problem of longer
documents. Utilizing the entire screen space, by dynamically
resizing all open documents to fit the screen, might
be a promising direction. Further information about this
project including a more detailed report, slides, a demo
video and software can be found at https://wiki.cs.umd.
edu/cmsc734_08/index.php?title=MultiVersioner.
References
Brin, S., Davis J., and Garcia-Molina H. (1995). Copy
detection mechanisms for digital documents, Proc. of the
1995 ACM SIGMOD international conference on Management
of data, 398-409.
Chi, E. H., Hong, L., Gumbrecht, M., and Card S. K.
(2005). ScentHighlights: highlighting conceptuallyrelated
sentences during reading, Proc. of the 10th International
Conference on Intelligent User Interfaces,
272-274.
Don A., Zheleva, E., Gregory, M., Tarkan, S., Auvil, L.,
Clement, T., Shneiderman, B., and Plaisant, C. (2007).
Discovering interesting usage patterns in text collections:
integrating text mining with visualization, Proc.
of the sixteenth ACM Conference on Information and
Knowledge Management, 213-222.
Filippova, D. (2007). BasketLens: interface for document
visualization and exploration, http://www.cs.umd.
edu/hcil/textvis/basketlens/.
MOSS. http://theory.stanford.edu/~aiken/moss, retrieved
04-10-2008
Schreibman, S., Kumar, A., and McDonald, J. (2003).
“The Versioning Machine.” Literary and Linguistic
Computing, 18(1), 101-107 (http://www.v-machine.org)
Shneiderman, B. (1996). “The Eyes Have It: A Task by
Data Type Taxonomy for Information Visualization.”
IEEE Conference on Visual Languages, 336-343.
Si, A., Leong, H. V., Lau, R. W. H. (1997). CHECK: a
document plagiarism detection system, Proc. of the 1997
ACM symposium on Applied computing, 70-77.