Real-time Haze Monitoring Based On Social Sensors

With the rapid development of China’s industry,new factors of environmental pollution are increasing.The traditional ways of finding pollutions,such as public tipoffs and physical instrument monitoring,has been unable to deal with the growing environmental problems.Therefore,it needs more accurate and real-time environmental monitoring data to help with the pollution detection.As an important channel for public expression of opinions,microblog has become an important platform for people to participate in political,social and economic life and a supervision by public opinions.The development of microblog provides us with a new breakthrough in monitoring the environmental situation.For example,when severe fog and haze happened somewhere,local people may complain about the weather quality by the microblogs.This paper implemented a real-time air monitoring method using microblog as a "social network sensor".The work of this paper can be divided into the following aspects:1.Corpus acquisition which related to target event——“Haze”.A crawler for the information of microblog is designed,which focus on the structure of web site pages and uses “haze” as the keyword to search for the whole network.Finally,the crawled relevant microblogs will be stored.The crawling content includes user name,microblog content,publishing time,publishing terminal,publishing location,etc.2.Propose an automatic classification method of "haze" indicator and the early warning model of "haze" based on "microblog active index".The model calculates the probability of haze occurrence from the results of the input "haze" indicator,and gives haze warning when the probability value exceeds a certain threshold.Experimental results show that the proposed method can automatically classify "haze" related microblogs and effectively identify "haze" indicators.The haze warning model can take into account the impact of Weibo activation index in an all-round way,and avoid the impact of "sleep period" for the prediction results to some extent.3.Apply an event embedding network which reconstructs event tuples based on relationship recognition in text database.It can transform the high-dimensional sparse word embedding into a low-dimensional dense event vector in order to solve the problems of less-information and lack of standard expression of microblog data.Besides,this paper realize the AQI trend prediction model with convolutional neural network(CNN)based on the event embedding.Experimental results show that the event embedding represent the event features in a micro-blog text more abstractly than using word embedding.In addition,compared with the general feedforward neural network,the prediction model based on CNN can extract the most representative global and local features,and establish the relationship between microblog events and the change of AQI,which performs better predicted results.