Enhancing Air Quality Prediction with Social Media and Natural Language Processing

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2627–2632 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2627 Enhancing Air Quality Prediction with Social Media and Natural Language Processing Jyun-Yu Jiang†, Xue Sun†, Wei Wang† and Sean Young‡. †Department of Computer Science, University of California, Los Angeles, CA, USA ‡Department of Family Medicine, University of California, Los Angeles, CA, USA {jyunyu, cynosure, weiwang}@cs.ucla.edu, sdyoung@mednet.ucla.edu Abstract Accompanied by modern industrial develop- ments, air pollution has already become a ma- jor concern for human health. Hence, air quality measures, such as the concentration of PM2.5, have attracted increasing attention. Even some studies apply historical measure- ments into air quality forecast, the changes of air quality conditions are still hard to moni- tor. In this paper, we propose to exploit social media and natural language processing tech- niques to enhance air quality prediction. So- cial media users are treated as social sensors with their findings and locations. After fil- tering noisy tweets using word selection and topic modeling, a deep learning model based on convolutional neural networks and over- tweet-pooling is proposed to enhance air qual- ity prediction. We conduct experiments on 7- month real-world Twitter datasets in the five most heavily polluted states in the USA. The results show that our approach significantly improves air quality prediction over the base- line that does not use social media by 6.9% to 17.7% in macro-F1 scores. 1 Introduction In recent centuries, industrialization has consider- ably changed human society by providing a stim- ulus to economic growth and improved life qual- ity. However, the advancement is accompanied by the increase in air pollutant emissions and risks to public health. As a consequence, predicting real-time air quality information (AQI), such as the concentration of PM2.5, has attracted more and more attention. Air quality prediction may help the government and society to better protect their citizens from potentially harmful effects of poor air quality. To forecast AQI, one of the most conventional approaches is to exploit historical air quality and treat the task as a time series prediction prob- lem (Genc et al., 2010; Zheng et al., 2015). How- ever, the air quality information can be too so- phisticated to be predicted by only past AQI with- out any additional knowledge. For example, other environmental factors like humidity and temper- ature can affect the air quality when real-world events like wildfires may also play a role. To learn the additional information, most of the rel- evant studies collect data from additional sensors like images (Jiang et al., 2011) and ground sen- sors (Zheng et al., 2015). Nevertheless, these sen- sors are expensive in not only installation but also maintenance. As a result, exploiting sensors for air quality prediction may be too costly for most of the cities. To learn additional knowledge without physical sensors, one of the most effective approaches is to leverage the wisdom of the crowd on the in- ternet. For example, 81% of the …

Content truncated to 500 words