Next: Agent model Up: Future Work Previous: The Filtering Engine

Genetic Algorithm

The Genetic Algorithm promises to be an interesting approach for modeling adaptive filtering systems. However, the GA approach was not fully exploited in this work due to time constraints. There are many directions for future work relating to GAs for information filtering.

One area for future research is in automating the application of genetic operators based on user actions. If the the user starts providing too much negative feedback, increasing the frequency of newer generations may perhaps speed up the adaptation. The GA used in this system involves a number of parameters, such as the retention rate, mutation rate, etc. These could also be automatically modified based on user behavior. For example, if a user consistently provides positive feedback for documents retrieved serendipitiously, perhaps the the number of mutations should increase.

Another interesting problem is to maintain diversity in the population by preventing profiles from getting too ``close'' to each other. Different profiles should ideally cover non-overlapping regions of the information space. Calculating the proximity of profiles in the vector space representation is easy. The interesting problem is to devise a scheme which applies selective pressure on neighbouring profiles to move away from each other. The economic model for optimizing computational resources, as used by Baclace [2], could possibly provide a solution.

Users typically have special needs that must be integrated into the GA model. For example, a user might create a special high priority profile which must never be eliminated by the population, no matter what. Some users might not want any profile to be eliminated unless it gets explicit negative feedback. The GA model described in this thesis must be extended so that the ``special'' members of the population are handled in a special way. These issues need to be addressed in future work.



Next: Agent model Up: Future Work Previous: The Filtering Engine


MIT Media Lab - Autonomous Agents Group - agentmaster@media.mit.edu