Xerox Scientists Automate Document
Classification
Xerox
scientists at the Grenoble research laboratories have created
software that models human thinking to classify documents and
information, automatically indexing, categorizing, and routing
electronic documents.
The
software is designed to help businesses keep their e-documents
collections orderly and easily accessible. Eric Gaussier, a
research scientist at the Xerox Research Centre Europe in Grenoble,
France says that the new software can help save time and money and
increase productivity. While current categorizing tools treat each
subject category independently, the Xerox system, based on
patented technologies, uses a hierarchical model that is able to
understand the dependency between categories and therefore make a
more informed.
-
The system needs little training. Using advanced machine-learning
techniques, with only a few examples (as few as ten per category),
it quickly learns by itself how to hierarchically classify
documents in existing categories.
-
It can turn unorganized e-files into cleanly labeled document
collections.
-
The system can learn entirely new categories on its own by
detecting new or emerging topics and dynamically suggesting new
categories to the people who are using the system.
In this it is as precise as humans.
The
categorizer handles documents written in up to 20 languages and
can be adapted for specific customer requirements. The software
routes documents to the right person based on a pre-set user
profile. Incoming
mail can be routed to the right person by topic (for example,
complaints to customer service and product queries to the right
marketing managers).
The
technology is based on linguistic analysis and machine-learning
techniques. The software is written in Java and can be deployed on
multiple platforms including UNIX, Linux and Windows.
The
company anticipates the technology to be licensed by software
vendors or corporations who wish to incorporate it into document
systems focused on areas such as customer relationship management,
information retrieval and data management.
(back to top)
