Incident detection from Tweets by neural network with GPGPU

Twitter is an online social network service to supply the place to be released the short sentences as free. Recently, this service has a few hundred million users, and we can collect their Tweets easily. In this paper, we propose the climatic hazard detection method by using Twitter as a social sensor and by using neural network as a machine learning method. However the data size of the text classification is too large. So, it is required to propose the high-speed learning algorithm. On another front, GPU is the dedicated circuit to draw the graphics, so it has a characteristic that the many simple arithmetic circuits are implemented. This characteristic is hoped to apply the massive parallelism not only graphic processing. In this paper, the neural network is applied to be faster by using GPU. Some methods are considered, and the simple one is employed as comparison to compare with the proposed methods. As the result, the proposed method is 6 times faster than comparison method. This neural network learning method is used for the text classification. 35,379 Tweets were gathered and these were deconstructed to the words by using morphological analysis. The feature vectors were constructed by using the nouns and adjectives selected from the words of the Twitter. We used 860 dimensions feature vectors and classified the positive data or negative. As the result of the classification, we achieved 68 percent accuracy to classify the Tweet data.