How do I combine support vector machine and some kind of expert system method in building a hybrid algorithm for best extracting matching fields in a big pool of data like legal cases for making better decisions?
According to Bolle et al. [1] it is possible to distinguish three different fusion levels:
1) Data-Level Fusion (use of multiple sensors or mutliple traits to acquire)
2) Feature-Level Fusion (use of multiple features)
3) Decision-Level Fusion (use of multiple classifieres)
[1] Ruud M. Bolle et al.: "Guide to Biometrics", ISBN: 0-387-40089-3, Springer New York, 2003
What you are looking for sounds pretty much like a decision-level fusion. You have 'n' classifiers resulting in 'n' decisions which need to be combined.
A very simple example for an equally weighted decision-level fusion using two different classifieres would be a simple logical combination. For simplicity we stick to a 2-class classification scenario where the classification result is either 'true' or 'false'.
For example we could simply combine these two decisions using logical AND or OR operator: 'decision classifier 1' AND 'decision classifier 2' = 'fused decision'.
Not sure about SVM but i remember using RIPPER (J-RIP, WEKA) and getting pretty decent result on some vision stuff, a while back. You may want to check out the paper to see if it is relevant to your interests:
@inproceedings{
Cohen1995, author = {William W. Cohen},
booktitle = {Twelfth International Conference on Machine Learning},