Adaptive agents for information retrieval and data-driven knowledge discovery
The recent proliferation of computers and communication networks has made it possible for individuals around the world to access a wide variety of information sources through the Internet. However, effective use of these information sources requires fairly sophisticated tools or software agents for locating, classifying, selectively retrieving and extracting knowledge from data. This dissertation addresses several related research issues in the design of such intelligent agents for information retrieval and knowledge discovery from distributed data and knowledge sources;Artificial neural networks, because of their potential for massive parallelism and fault tolerance, offer an attractive approach to the design of intelligent agents. Our work extended several single layer perceptrons and constructive neural networks of perceptrons in order to handle multi-category, real-valued patterns. In particular, we designed DistAl , a novel constructive neural network learning algorithm based on inter-pattern distance. DistAl is significantly faster than conventional neural network algorithms and has been demonstrated to perform well on a broad variety of benchmark data-driven knowledge discovery problems. The performance of DistAl was further improved by using it in conjunction with a genetic algorithm for automated selection of features used to encode the data. DistAl was also used for data-driven refinement of incomplete or inaccurate domain knowledge. Some of these algorithms were used in a design of a multi-agent system consisting of multiple cooperating customizable intelligent mobile agents for selective information retrieval and knowledge discovery from distributed data sources.