Related Experiment Videos
Marginal median SOM for document organization and retrieval
A Georgakis1, C Kotropoulos, A Xafopoulos
1Artificial Intelligence and Information Analysis Laboratory, Department of Informatics, Aristotle University of Thessaloniki, Box 451, Thessaloniki GR-54124, Greece. apostolos.georgakis@tfe.umu.se
Summary
This study introduces the self-organizing map (SOM) algorithm for document retrieval, enhancing its performance by using a marginal median adaptation rule. This novel approach requires fewer training iterations for improved document organization and retrieval accuracy.
Area of Science:
- Artificial Intelligence
- Information Retrieval
- Machine Learning
Background:
- Self-organizing map (SOM) algorithms are established for document organization.
- Existing SOM methods face limitations in document retrieval efficiency.
Purpose of the Study:
- To adapt the self-organizing map algorithm for effective document retrieval.
- To evaluate a novel SOM variant using a marginal median adaptation rule.
- To compare the performance of the enhanced SOM against traditional methods.
Main Methods:
- The self-organizing map algorithm was modified for document retrieval.
- A marginal median adaptation rule replaced the linear Least Mean Squares rule.
- Two implementations were tested: one with quantified real-valued feature vectors and one without.
- Performance was evaluated using mean square error and average recall-precision curves on two distinct corpora.
Main Results:
- The modified SOM demonstrated superior performance, requiring fewer training iterations to achieve a target mean square error.
- Both implementations of the marginal median variant outperformed the standard SOM-based method.
- The system achieved effective document organization and retrieval across diverse datasets, including web pages and the Reuters-21578 corpus.
Conclusions:
- The proposed self-organizing map variant with a marginal median adaptation rule significantly improves document retrieval efficiency.
- Quantifying feature vectors offers a viable implementation strategy for the enhanced SOM.
- The findings suggest a promising direction for developing more efficient document organization and retrieval systems.