A Novel Chinese Points of Interest Classification Method Based on Weighted Quadratic Surface Support Vector Machine

Points of interest (POIs) are some focused geographic entities or specific locations that a considerable group of persons find useful or interesting. They are always the basis for supporting location-based applications such as navigation systems, recommendation systems and so on. And these applications always rely on the accurate POIs classification. In this paper, a novel classification method based on weighted quadratic surface support vector machine (WQSSVM) is proposed to classify Chinese POIs from different websites. We first utilize the large number of Chinese POIs to build sparse feature vectors. Then, a weight function is designed to calculate the relative importance of each sample, which is the input to the WQSSVM model. Finally, the proposed WQSSVM model is trained to obtain a suitable classifier supporting by a small proportion of the high-quality samples, and classify the rest large portion of POIs automatically. The WQSSVM model avoids the disadvantages induced by the kernel functions used in classic support vector machine models with kernels. The numerical results on thirteen real-life Chinese POIs datasets indicate that the WQSSVM model not only outperforms the QSSVM model due to the designed weight function but also outperforms other state-of-the-art text classification models in terms of classification accuracy.