Research Article
A Novel Feature Selection Strategy for Enhanced Biomedical Event Extraction Using the Turku System
Algorithm 1
AEE1 algorithm.
Step 1. Categorize features into feature classes and there are possibilities for all kinds of feature | combination. Denote , and run all of classifying test. -score is recorded for | each experiment. | Step 2. Sort feature combinations according to -score with the descending order. | Step 3. FOR to , to , | , ; | END FOR | // is used to calculate the occurrence of the th feature among experiments. | // evaluates the partial importance of the th feature after times experiments. | Step 4. FOR to , // For each feature combination in the th experiment. | FOR to , // For each feature. | IF the th feature occur in the th feature combination, | ++; | // For each experiment, the occurrence of the th feature is accumulated. | END IF | ; | // By dividing by , we get a numerical value to evaluate the importance | // of the th feature. Since the items are ranked by -score with a descending | // order, the first ranked features combination corresponds to a smaller and | // hence a bigger . | END FOR | END FOR | Step 5. FOR to , | AEE1() =; | END FOR | // By summing up for a fixed , we get the accumulated effect of the th feature. | Step 6. AEE1() is the objective function of greedy search algorithm, and we sort AEE1() | to get the most important feature. |
|