This is my BSc dissertation, completed in the summer of 1997. For no-one in particular's delectation and amusement, I implemented a framework for Intelligent Agents, evolved using Genetic Algorithms, to learn about the user's Usenet reading habits, and suggest new and interesting articles in places the user would never think to look. The idea was for this to eventually become either a full-blown AI newsreader (decentralised), or an NNTP proxy with agent extensions (centralised).
Information Filtering and retrieval are rapidly becoming more important as the volume of publically available electronically stored information increases. An existing system, Beerud Dilip Sheth's NEWT is implemented and evaluated. Improvements are discussed, implemented and incorporated into a new system, Salamander. Both systems provide personalised information Filtering that adapts to a user's interests. A hybrid Genetic Algorithm is used to explore new domains for information the user might nd useful. Salamander's Genetic Algorithm also evolves better matches to the user's changing focus of attention. The two systems are discussed, evaluated and compared. Conclusions on Genetic Algorithms as an aid to information ltering are drawn, shortcomings identiﬁed and proposed future improvements are included.
The C++ source code, documentation source and experimental framework are still available, but I have not made them public yet. For now, you can have the document itself. Please note that the PDF version is converted from the original Postscript version many years after the fact. As such, some rendering issues may make an appearance, especially when reading the dissertation on-screen.