2000 Volume 1 Issue 4 Pages 47-56
With the explosion in the amount of information on Internet, finding desired information put additional cognitive and repetitive burdens on the user. In order to overcome these drawbacks, we propose a layer of adaptive and collaborative agents between the layer of users and the layer of distributed information sources like Internet. The personalization is realized by learning the user model and using it at query formulation. Machine learning and information retrieval techniques were utilized to learn the user preferences and to provide support for well-formed personalized query reformulation. However, for learning agents working individually, they face two problems: (i) serendipity, i.e. they cannot deal properly with situations previously unseen in the past; and (ii) cold-start: they spend some time to relearn about new situations.
In order to deal with these problems, we add a layer of collaboration between the agents, where the selection of peers is based on the trust relationship among them. The collaborative aspect is obtained by exchange of information learned by the individual agents.