Enhancing Personalized Recommendation by Transductive Support Vector Machine and Active Learning

Algorithm 1

TSVM algorithm based on distance strategy and multiclassifier collaborative labeling (DCTSVM).

	Input: Labeled sample set L; unlabeled sample set U; the number of classifiers k.
	Output: The final classifier TSVM.
	Step 1: Apply the K-means algorithm to cluster the labeled sample set L, and extract samples from each cluster according to a specific proportion to form k sub-training sets, which are denoted as .
	Step 2: Utilize the SVM algorithm to train k training subsets to obtain k initial classifiers: .
	Step 3: Input unlabeled samples into , and obtain k output results: .
	Step 4: For any unlabeled sample xj, if the classification results of a k classifier meet equation (8), then label it as a positive class; if the classification results of a k classifier meet equation (9), then label it as a negative class.
	Step 5: If the currently labeled class of xj is inconsistent with the previously labeled class, cancel the labeling and delete it from the corresponding training set. If the currently labeled class is consistent with the previous and is inconsistent, then add the sample to Lj. If the sample is not labeled in the early stage, then find j that meets , and add the sample to Lj; otherwise, stop the iteration and go to Step 8.
	Step 6: Repeat steps 4 and 5 until all unlabeled samples are labeled.
	Step 7: Train the new training subset and obtain the new classifiers . If the sub-training sets of the previous and current iterations remain unchanged, the corresponding training should continue to use the classifier from the previous iteration; then, go to Step 3.
	Step 8: Combine each training subset to form the final training set, and retrain the sample set to obtain the final classifier.