IT鈥橲 DARK. Surrounded by inky blackness, you are alone in a void. Away in the
distance loom several indistinct shapes which, as you zoom closer, resolve into
clusters of text. Words appear: 鈥淪ports鈥�, 鈥淓ntertainment鈥�, 鈥淣ews鈥� . . . As you
fly towards News, it sprouts bright lines that link it to still more words, such
as 鈥淭oday鈥檚 headlines鈥� and 鈥淵esterday鈥檚 features鈥�. Descending into the last of
these brings up more refined choices: Politics, Health, Science . . . You dive
into one of these, and there before you is just what you are looking for.
This surreal scene is not from a fantasy space game. It鈥檚 a demonstration of
Galaxy of News, a software tool designed to simplify tasks such as navigating
through a newspaper鈥檚 website. The idea is to give you an overview of the entire
content of a site on a single screen, so you can home in quickly on the article
you want. Galaxy of News is just one example of an idea being developed by a
small group of companies that aims to revolutionise the way we gather, analyse
and visualise information.
These pioneers all depend on giving computers a new interface geared to
spatial searching. Data will be displayed on screen in a way that lets you find
what you are after simply by reaching for it鈥攍ike pulling the item you
want from the shelves of a well-organised stock room or library. Clever graphics
give the impression of three dimensions, and the field of view changes at the
click of a mouse, allowing you to search from side to side and back and forth.
Long lists of files or hyperlinks will be banished for good.
Advertisement
But what鈥檚 on offer is not just fancy window dressing. These displays are
being linked to smart software that can 鈥渦nderstand鈥�, at least to some extent,
the concepts contained within document files. This clever code can take a mass
of unstructured data from, say, a web search, sort it by content and decide how
that content should appear on your screen. The upshot is a completely new view
of information, which is unlocking secrets that a conventional search would
never reveal.
Every computer system has some kind of user interface鈥攁 layer of
programming that sits between the raw data in a computer鈥檚 memory and the user,
and dictates how that data is presented on screen and how the user interacts
with it. Those in common use today, such as Microsoft Windows and the Macintosh
operating system, are based on metaphors that were chosen to be familiar to
people in the 1970s and 1980s, such as filing cabinets and paper documents. But
treating a computer like an old-fashioned office severely restricts the way
people can interact with it.
Windows, for example, closely parallels the way computers organise the data
they store. Users are still expected to negotiate a hierarchy made up of lists
of directories and subdirectories stuffed with lists of files. In effect, the
computer forces us to adapt to the way it works, instead of the other way
round.
The World Wide Web is little better. True, the advent of hyperlinks means it
no longer matters where in a computer a file is stored鈥攐r even on which
computer it鈥檚 stored. But too many websites expect surfers to click through
endlessly layered lists of hyperlinks to reach the information they are after.
鈥淭here is no reason why we have to access data in that way,鈥� says Earl Rennison,
who devised Galaxy of News (www.perspecta.com/indexCorp.html). 鈥淚t鈥檚
impressing the limitations of the physical world unnecessarily.鈥�
Human beings are used to locating things spatially. 鈥淭he human/computer
interface is much more effective if it works in a way that鈥檚 closer to the way
we think,鈥� says John Peterson, who is a co-founder of Object/FX
(www.objectfx. com), a Minnesota-based company that makes visual interfaces
for geographic information systems. But this goal of 鈥渉umanising鈥� the interface
does not need super-sophisticated 3D graphics. Sometimes it鈥檚 just a matter of
redrawing a hierarchy of directories and files in two dimensions in a way that
is more intuitive to its users.
That鈥檚 the plan at Inxight (www.inxight.com) of Palo Alto,
California, a spin-off from the nearby Xerox PARC. Inxight鈥檚 Hyperbolic Tree
software represents a directory as a wheel hub. Subdirectories or files
within it appear at the end of spokes radiating from it. Click on a subdirectory
and the image warps, bringing it centre-screen and magnifying it, while the hub
you began with moves to one side and gets smaller. 鈥淲e mimic the way a human
being works,鈥� says Ramana Rao, chief technology officer at Inxight. 鈥淲hen you
turn your head to look at something, that object becomes the centre of your
补迟迟别苍迟颈辞苍.鈥�
Thanks to these continual changes of perspective you can find the file you
want without losing sight of the whole hierarchical file structure. Rao
dismisses more advanced 3D visualisation packages as unnecessary. In any case,
most desktop computers do not yet have the processing power needed to run
them.
The potential uses of the Hyperbolic Tree are huge. Take the trend towards
Internet shopping. In today鈥檚 online supermarkets, even buying a bag of tomatoes
can be a trial. You head first to the Groceries section, which presents you with
a list. From there you select Fresh Foods, and wait for another list. Finally
you reach the vegetables, where you scroll down the list to pluck your tomatoes.
Each list can take ages to download. The Hyperbolic Tree avoids all this. It
displays the layout of the whole supermarket on a single screen, with every
section represented by a hub and every item it contains at the end of a spoke
(see Diagram).
Inxight designed the Hyperbolic Tree to be bolted onto other applications.
And this is already happening. Microsoft has included it in its Site Analyst
program, as a tool to give website designers an overview of the structure of
their sites. This helps, for example, to find broken links when a page is moved.
Less specialised applications are also starting to emerge. Later this month,
Inxight plans to release a file browser to complement Windows鈥檚 File Manager. It
will be available from Inxight鈥檚 website.
Of course, not all information comes as neatly organised as a computer鈥檚
files and directories. But the new interfaces can handle unstructured
information from memos, letters and newspaper articles, so long as it has first
been processed by one of the increasingly smart search engines and data mining
tools that decide by themselves how information should be structured. Last
October, for example, Cartia of Redmond, Washington, launched a program called
ThemeScape, which turns unstructured information from a database or Web search
into a topographic landscape. Just how the land lies is decided entirely by the
program.
Military intelligence
At the heart of ThemeScape is a text-processing engine called SPIRIX,
which combines statistical analysis and the simple grammatical rules used by
humans in natural language to locate key concepts. The methods it uses were
devised by American military intelligence and are still closely guarded by
Cartia. But at a simple level, it separates a sentence into its constituent
parts and looks for common themes. So, 鈥淭he cat sat on the mat鈥� would be broken
down into subject (cat), verb (sat) and object (mat). If the following sentence
read 鈥淚t licked its fur鈥�, the program would associate the 鈥渋t鈥� with the subject
of the previous sentence, and so make a connection between the cat and fur. In
this way, the concept that cats have fur and lick it becomes part of the
program鈥檚 鈥渦nderstanding鈥�.
SPIRIX extracts these concepts and creates a set of them for every document
in a database, which it then places within a multidimensional matrix. In
essence, it calculates multiple virtual 鈥渓ocations鈥� for each document, based on
its content. A document on terrorism might contain concepts such as violence,
fundamentalism and explosives, and will be represented in the matrix by these
concepts. Documents that share similar concepts are placed near to one another
in the matrix.
Conceptual navigation
ThemeScape then plots these concepts onto what looks like a relief map of an
island. It has light-coloured mountains, dark valleys and sharply defined
ridges. A mountain grows where a cluster of documents share similar concepts.
The more documents there are, the higher the peak grows. Each area of the map is
labelled to indicate the general concept it represents. And any documents that
have nothing to do with the mountain鈥檚 topics are dumped into the sea around the
island.
You use ThemeScape to explore the data using what Don Mason of Cartia calls
鈥渃onceptual navigation鈥�. Clicking the mouse on a particular area zooms you in on
it, and drills down into the data landscape to reveal details of the documents
beneath. In this way it鈥檚 possible to locate a single document. But the real
beauty of ThemeScape, says Mason, is that is gives you a bird鈥檚-eye view of a
body of information without having to read any individual documents.
ThemeScape can analyse around 250 000 documents simultaneously. It allows
users to home in on the information they鈥檙e interested much more quickly than
they could by reading down a long list of keywords. Importantly, the overview
can also reveal general trends that would otherwise remain hidden.
Mason recalls demonstrating Theme-Scape to an American car maker, which used
the system to map its entire database. From the mix of e-mails, reports, memos,
news feeds and assorted data, an island began to emerge. One surprisingly
prominent peak on the island related to a technology that the company had
abandoned. In this mountain were old memos dismissing the idea, together with
recent news reports that a Japanese rival was filing patent after patent on the
technology. The discovery, Mason reveals, sent executives scurrying for their
mobile phones. 鈥淚t鈥檚 competitive intelligence,鈥� he says. 鈥淭his was a piece of
information that would have been lost among hundreds of thousands of documents,
but which the right visualisation technology could reveal.鈥�
As well as dealing with a large number of documents, ThemeScape can be put to
work on a single large document. It took less than 3 minutes, for example, to
map the Starr report into President Clinton鈥檚 inappropriate relationship with
Monica Lewinsky (www.cartia.com/screenshots/ screenshot_menu.html).
Another view of the Starr report has been concocted by Semio in San Mateo,
California. Like ThemeScape, SemioMap (http://demo.semio.com/public/discover.cgi)
carries out a conceptual search, but it lays out the results in a different way. Its findings appear as a vast
three-dimensional network of interlinked hubs set against a background of stars
(see Diagram). Click on a hub and you fly towards it. As you get closer,
other more distant hubs come into view. Double-click on one of these and it
reveals the passages of the report to which it refers.
One consequence of this trend towards conceptual navigation is that the
location of the data will become ever less important. Think about searching for
information stored in someone else鈥檚 computer system. As things are now, you
have only the directory and file names chosen by somebody else to guide
you鈥攏ames that are often too personal and cryptic to mean much. But a tool
able to read all the document files and extract the major concepts within them
would make a big difference. 鈥淭he new systems switch the focus from where
something is to what it鈥檚 about,鈥� says Mike Lynch, chief executive of Autonomy,
based in San Francisco, which produces a data categorisation and navigation tool
called AgentWare.
Taking this notion further, Lynch speculates that in future users will not
need to see any form of file structure at all. A natural-language analyser will
interpret the user鈥檚 needs and locate the information. And, unlike Web searches
conducted today, there will be no poring over list after list of hyperlinks: the
entire search will be laid out conceptually on a single screen.
At present, most of these tools are aimed squarely at corporate users. The
reason is simple. 鈥淚ndividuals won鈥檛 pay for it,鈥� says Peter Martin,
vice-president of sales at Semio. But companies that need to make sense of large
bodies of information will pay whatever it takes to give them a competitive
edge.
Eventually, Martin reckons, the new tools will trickle down the familiar
pathway to computers in people鈥檚 homes. When Lotus produced the first commercial
spreadsheets, he says, it marketed them to companies, not individuals. After
people became accustomed to using them at the office, demand grew for them to be
made available at home. 鈥淭hat鈥檚 what鈥檒l happen with graphic display devices,鈥� he
says. 鈥淭hey鈥檒l be tried out and paid for by corporations, then home users will
come in.鈥�
The first outward sign that the new search tools have arrived will be the
changing interface: the switch away from the old document metaphor. Rennison
compares today鈥檚 interface designers to the earliest film makers. 鈥淭o begin with
they shot movies just like a play, with no cuts or camera movements,鈥� he says.
鈥淚t took 50 years for the real language of the medium itself to be expressed.鈥�
If the change from document metaphor to spatial search is like the switch from,
say, silent movies to talkies, it鈥檚 clear we can expect some pretty spectacular
changes to the interface in the coming years.