Interactive patent classification based on multi-classifier fusion and active learning

Patent classification is of great importance to effective patent analysis. Traditional manual classification suffers from the problem of low efficiency and high expense. To address this issue, an interactive patent classification algorithm based on multi-classifier fusion and active learning is proposed in this paper, which comprises the construction and update of classification model. For model construction, a sub-classifier is trained for each class of the patents by means of support vector machine. Via multi-classifier fusion, the sub-classifiers are effectively combined to acquire enhanced classifiers, based on which the classification decision can be made. For model update, active learning is used to select the most informative patents for labeling, in which dynamic batch sampling is presented to cope with the problem of redundancy in traditional batch mode. Using dynamic certainty propagation, the selected patents become more informative for active learning. By iterating model construction and update, the classification performance can be gradually refined. The interactive classification algorithm is applied to both synthetic data and patents, and its effectiveness is demonstrated by the encouraging results.