A study on method of noise reduction for situation visualization by using geo-tagged tweets

The study of geo-tagged tweet data has been increasing. In such a situation, the authors considered the method of noise reduction for improving the precision in the situation visualization by using geo-tagged tweets. Specifically, it is a study from the combination method of the autogeneration of noise reduction filter by using Natural Language Processing (NLP), and the noise reduction by the multiple people in the same time zone and near distance. In the NLP method, precision level at the about 53% was observed from the test by using the morpheme-3gram. And nominal significant difference at the 0.5% was observed in comparison with non-filter method. In the near distance method, precision level at the about 80% was observed, and nominal significant difference at the 0.5% were observed in comparison with NLP method. In the combination method of NLP and near distance, precision level at the about 84% was observed, and nominal significant difference at the 0.5% was observed in comparison with near distance method. Furthermore, in the verification by type of rainfall, it was revealed that the combination method can extract with higher accuracy than the NLP method or the near distance method from the extraction result with high accuracy exceeding 95% in the rainy situation. As a result, under the conditions in urban areas with many tweets, the results from this study on combination method indicated the certain effect.