Intelligent Filtering of Computer-Mediated Human Communication
1. GOAL
Develop methods to aid users of asynchrounous group
communication systems (e.g. email, mailing lists, conference
systems, Usenet News) to select important messages.
2. MOTIVATION
Group communication systems are spreading increasingly fast as
they become part of the infrastructure in educational,
commercial and governmental organizations. The systems provide
effective means for fast coordination, dissemination and
retrieval of information. As with all new media they have
specific advantages, but with increased popularity a number of
disadvantages, such as information overload and a low ratio of
quality to quantity also appears.
3. METHOD
The KOM group conference system collects messages from several
sources (KOM, Usenet News, mailing lists and email) and
presents them in a coherent form. The KOM client enables the
user to process messages from or to these sources through a
single interface.
In the process of reading a stream of messages, the user
classifies some of them as highly interesting and some as
rubbish. The classification is individual to each user and
develops over time. Each message has a number of absolute or
inferred properties such as subject, length, sender, keywords,
conference etc. By solicitating the user's explicit
classification of typical messages, it is hypothesized that
machine learning techniques can be employed to perform
automatic classification of messages, using the absolute and
inferred properties of the messages.
4. EXPECTED RESULTS
Particular classes of messages are more easily identifiable
than others. The presence of important keywords, for instance,
is a significant factor in the determination of a message's
class. The automatic classification accuracy is highly
dependent on the machine learning technique employed, the set
of training examples given by the user and the concept or
class these training examples represent to the user.
5. APPLICATIONS
An automatized message classification module, coupled to the
KOM conference system. The module provide user's with a
priority estimate of their unseen messages, according to their
preferencs as stated through examples given to the system.