Weak learning algorithm for multi-label multiclass text categorization
XU Yan-yong, Xianzhong Zhou, Zhong-Wei Guo · 2003
To handle the multi-label multiclass text categorization, a weak learning algorithm (WLA) is presented. The main idea of WLA is to find a highly accurate classification rule by combining many weak hypotheses, each of which may be only moderately accurate. We used a separate procedure, called the weak learner, to compute the weak hypotheses, and found a set of weak hypotheses by calling the weak learner repeatedly in a series of rounds. These weak hypotheses were then combined into a single rule called the final hypothesis, and the final hypothesis ranked the possible labels for a given document with the hope that the appropriate labels would appear at the top of the ranking. Using the three designed evaluation measures - ordinary-error, average-coverage and average-precision - our experiments show that the performance of WLA is generally better than the other algorithms on the same dataset.