Patent Classification Using Parallel Min-Max Modular Support Vector Machine

The patent classification problem has a very large scale dataset. Traditional classifiers cannot efficiently solve the problem. In this work, we introduce an improved parallel Min-Max Modular Support Vector Machine (M³-SVM) to solve the problem. Both theoretical analysis and experimental results show that M³-SVM has much less training time than standard SVM^light. The experimental results also show that M³-SVM can achieve higher F₁ measure than SVM^light while predicting. Since the original M³-SVM costs too much time while predicting, in this work, we also introduce two pipelined parallel classifier selection algorithms to speed up the prediction process. Results on the patent classification experiments show that these two algorithms are pretty effective and scalable.