Machine Learning and Data Science
GAN-based EEG Signal Generation
Abstract: Processing and analysis of brain signals generally requires a large amount of data. But the acquisition of EEG signals is difficult while the sample size of the data set is small, and sometimes categories are unbalanced in the data set. Based on the challenge, we proposed the WGAN-GP method, a variant of GAN, to generate useful EEG signals. The experiments on single-channel and multi-channel model both show that the performance of WGAN-GP is stable, and the generated signals have close shape with the real signals, and have better spectrum performance than traditional methods. Our results and analysis show that WGAN-GP can generate accurate and diverse EEG signals, and thus, help extend the data set which is difficult to collect physically. We’ve made the code associated with this work available at https://github.com/warrenzha/GAN-EEG-generation.
Generated EEG signals and convergence rate of WGAN-GP.
Classification of Online Customer Reviews
Abstract: Customer reviews on e-commerce platforms contain valuable information, while sifting through them manually tends to dismay people because of the huge amount of data. This study implemented a machine learning-based algorithm to classify customer reviews. Our classifier extracts Chinese word segmentation and text frequency for feature extraction and scoring, and implements the classification with methods of Naive Bayesian and Support Vector Machines. Experimental results on the Taobao product review sentiment datasets show that our model based on two machine learning algorithms, though results in different performances, can provide suggestions on the selection of the identification classifier using a trade-off strategy and helps obtain fast and accurate classification on reviews of different categories.